INTRODUCTION TO THE SERIES

The aim of the Handbooks in Economics series is to produce Handbooks for various
branches of economics, each of which is a definitive source, reference, and teaching
supplement for use by professional researchers and advanced graduate students. Each
Handbook provides self-contained surveys of the current state of a branch of economics
in the form of chapters prepared by leading specialists on various aspects of this
branch of economics. These surveys summarize not only received results but also
newer developments, from recent journal articles and discussion papers. Some original
material is also included, but the main goal is to provide comprehensive and accessible
surveys. The Handbooks are intended to provide not only useful reference volumes for
professional collections but also possible supplementary readings for advanced courses
for graduate students in economics.

KENNETH J. ARROW and MICHAEL D. INTRILIGATOR

PUBLISHER'S NOTE

For a complete overview of the Handbooks in Economics Series, please refer to the
listing at the end of this volume.
INTRODUCTION

KOTARO SUZUMURA*
Trinity College, Cambridge, UK and Hitotsubashi University, Kunitachi, Tokyo, Japan

Contents

1. Historical background 1
2. Social choice and individual values 10
3. "Socialist planning" controversy 13
4. Significance of the subject and main lines of research 18
5. A disclaimer 25
References 26

1. Historical background

Social choice theory is concerned with the evaluation of alternative methods of

collective decision-making, as well as with the logical foundations of welfare
economics. In turn, welfare economics is concerned with the critical scrutiny of
the performance of actual and/or imaginary economic systems, as well as with the
critique, design and implementation of alternative economic policies. This being the
case, it goes without saying that the origin of social choice theory can be traced
back all the way to antiquity. Indeed, as soon as multiple individuals are involved in
making decisions for their common cause, one or other method of collective decision-
making cannot but be invoked. As a reflection of this obvious fact, there are numerous
examples in classic writings on the use and usefulness of alternative methods of
collective decision-making. Suffice it to quote Aristotle in ancient Greece, and Kautilya
in ancient India; they both lived in the fourth century B.C. and explored several

* Thanks are due to the co-editors of the Handbook of Social Choice and Welfare, Professors Kenneth
J. Arrow and Amartya K. Sen, whose encouragement, comments, and persuasion enabled me to
complete this Introduction. Thanks are also due to all the contributors to this Handbook, whose
willing collaboration made the completion of the project at all possible. In preparing several drafts
of this Introduction, I was greatly supported by helpful comments and encouragement provided by
Professors Nick Baigent, Walter Bossert, Marc Fleurbaey, Wulf Gaertner, Louis Gevers, Peter Hammond,
Herv6 Moulin, Maurice Salles, Koichi Tadenuma, John Weymark, and Yongsheng Xu. Needless to say,
I am solely responsible for any errors which may still remain.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K. Sen and K Suzumura
© 2002 Elsevier Science B. V All rights reserved
2 K. Suzumura

possibilities of collective decision-making in their books entitled, respectively, Politics

and Economics .
Likewise, as soon as any collective body designs and implements an economic
mechanism and/or an economic policy, paying proper attention to the costs and benefits
accruing to its constituent members, one or more social welfare judgements cannot be
avoided. In this sense, Joseph Schumpeter (1954, p. 1069) was certainly right when he
emphasized "the hallowed antiquity of welfare economics". He observed that "a large
part of the work of Carafa and his successors as well as of the work of the scholastic
doctors and their successors was welfare economics. We also know that the welfare
point of view was much in evidence in the eighteenth century ... For Bentham and the
English utilitarians generally this point of view was, of course, an essential element of
their creed. Hence, the positive spirit of Ricardian economics notwithstanding, we find
it also in the English 'classics', particularly in J.S. Mill. So far as this goes, modem
welfare economists merely revive the Benthamite tradition". It was in similar vein that
Paul Samuelson (1947, p. 203) began his famous Chapter VIII on Welfare Economics
in Foundations of Economic Analysis with the following remark: "Beginning as it did
in the writings of philosophers, teleologians, pamphleteers, special pleaders, and re-
formers, economics has always been concerned with problems of public policy and
welfare".
Without contradicting these authoritative verdicts on the long historical background
of social choice theory, we may nevertheless claim that the instrumental concern
with concrete methods of collective decision-making is one thing, and theoretical
investigation into their logical performance is another thing altogether. The former
concern may be as old as the origin of human society, but the latter development
seems to be of more recent origin. Indeed, it seems fair to say that the real origin
of the collective decision-making side of the coin can be attributed to the pioneering
contributions by two eminent French precursors around the time of French revolution,
viz. Marie-Jean de Condorcet, and Jean-Charles de Borda 2. It was in the intellectual
atmosphere of the European Enlightenment during the eighteenth century, with its
conspicuous concern with human rights and its reasoned design and implementation of
rational social order, that Condorcet (1785) addressed the mathematical discipline of

i See Sen (1999a, p. 350).

2 lain McLean and John London (1990, p. 107) maintained convincingly that they found "two medieval
thinkers, hitherto unknown to historians of social choice [viz. Ramon Lull (c. 1235 1315), who proposed
the Condorcet method of pairwise comparisons, and Nicolas Cusanus (1401-1464), who proposed the
Borda method of rank-order comparisons], who anticipated the work of Condorcet, Borda and Dodgson
by over 500 years". They aptly added, however, that "[n]either writer gives a mathematical or logical
justification for his scheme: such justifications had to await Condorcet and Borda [McLean and London
(1990, p. 106)]". It was for this reason that McLean (1995) later christened the period over which Borda,
Condorcet, and their contemporaries worked on the theoretical performance of voting schemes "the first
golden age of social choice".
Introduction 3

collective decision-making in terms of simple majority voting and related procedures 3.

He discovered the paradox of voting, or the Condorcetparadox, to the effect that the
method of pairwise simple majority voting may yield a social preference cycle - a
social alternative A defeating another alternative B by a simple majority, B defeating
the third alternative C again by a simple majority, and C in its turn defeating A by a
simple majority. This paradox sent an unambiguous signal that the logical performance
of voting and related procedures for collective decision-making must be the subject of
theoretical scrutiny. One of the logical implications of the Condorcet paradox is that,
once a simple majority cycle occurs in the set of social alternatives S = {A, B, C},
there exists no Condorcet winner - a feasible alternative which is undefeated by any
other feasible alternative - thereby excluding the possibility of basing social choice
on the seemingly democratic method of collective decision-making. It is worthwhile
to recollect in passing that Condorcet's first extended illustration of the paradox of
voting was taken from voting on economic policy. Indeed, the three policy alternatives
were 4

A = any restriction placed on commerce is an injustice;

B = only those restrictions placed through general laws can be just;
C = restrictions placed by particular orders can be just.

Condorcet's contribution seems to have been, at least partly and indirectly, inspired by
an earlier work by Borda (1781), who proposed what came to be known as the Borda
method of rank-order decision-making 5. For each voter, this method assigns a score
of zero to the last ranked alternative, a score of one to the penultimate alternative,
and so on all the way up to the top ranked alternative, which receives a score of n - 1
when there are n alternatives altogether. These individual scores are added for each
candidate over all voters, and the candidate which earned the largest sum-total becomes
the overall winner in the contest. According to Duncan Black (1958, p. 180), "[s]oon

3 It is worthwhile to recollect that "Condorcet's work on social choice (1785-94) spans the most active
constitution-making era in Western history until then, and the most active ever until 1989. Constitutions
for the United States, Poland, and France were written, and Condorcet was connected with all three ...
In 1792, Condorcet was made the chairman of a committee to draw up a Constitution for France ...
After the Jacobin coup d'6tat of June 1793, Condorcet was out of power. His constitution was dumped
in favor of one drawn up in great haste by Robespierre, who dropped all Condorcet's voting schemes
[McLean (1995, pp. 23-26)]". Condorcet's work on the theory of voting and human rights is translated
into English by lain McLean and Fiona Hewitt (1994).
4 See Emma Rothschild (2001, p. 181).
5 Borda's rank-order method was first proposed orally at the French Academy of Science in 1770, which
remained unpublished until 1784. Condorcet was well aware of this method, and immediately recognized
it to be an important challenge to his own pairwise comparison method. He stated in Condorcet (1785,
Discours pr6liminaire, p. clxxix) that he had heard of Borda's method orally, but that it was not published
until after his own work was in press. According to McLean (1995, p. 16), however, it was actually
Condorcet himself who published Borda's work.
4 K. Suzumura

after hearing Borda's paper in 1794 the [French] Academy [of Sciences] adopted his
method in elections to its membership. It remained in use until 1800, when it was
attacked by a new member and was modified soon afterwards. The new member was
Napoleon Bonaparte."
The same rank-order voting procedure was obtained from slightly different premises
by Pierre-Simon Laplace (1812)6. Laplace also acutely observed an obstacle to the
use of this procedure to the effect that "its working might be frustrated by electors
placing the strongest opponents to their favorite candidates at the foot of their list.
This would give a great advantage to candidates of mediocre merit, for while getting
few top places they would also get few lowest places [Black (1958, p. 182)]". As a
matter of fact, the same difficulty was confronted by Borda himself, who, when his
procedure was opposed precisely for this reason of strategic vulnerability, had retorted
by saying that his scheme is "only intended for honest men [Black (1958, p. 182)]".
This episode seems to show us unambiguously that the apprehension about the strategic
manipulability of voting schemes existed from the formative era of this side of social
choice theory.
There was intermittent exploratory work on voting schemes in the nineteenth
century, most notably by Charles Lutwidge Dodgson (1873, 1874, 1876), who is better
known by his literary pseudonym (Lewis Carroll). His works were circulated only
within a limited Oxford circle, and was virtually unknown in the outside world until
Black (1958, Appendix) made them widely accessible. Although ample circumstantial
evidence [Black (1958, pp. 192-194)] exists that Dodgson was acquainted neither
with Borda (1781) nor with Condorcet (1785), he was clearly aware of the ubiquity
of cyclical majorities as well as of the rank-order method of voting, most probably
through Isaac Todhunter (1865, Chapters XVII and XIX), which every late Victorian
scholar seems to have known about 7. His major logical concern was to devise a voting
procedure which would enable him to choose the Condorcet winner if one exists, and
to lexically supplement the simple majority voting if and when the Condorcet winner
failed to exist. Black seems certainly right in concluding that "Dodgson had been
caught in the grip of the theory of elections and committees and his understanding
of the subject was second only to that of Condorcet [Black (1958, p. 212)]".
In the last part of the nineteenth century and the first half of the twentieth century,
some sporadic contributions such as those by Edward J. Nanson (1882) and Francis
Galton (1907) notwithstanding, not much seems to have been done in the theory of
collective decisions, the major breakthrough having been accomplished only in the
late 1940s by Duncan Black (1948). He found a simple sufficient condition on the

6 For Laplace's theory of elections, those who are interested should refer to Isaac Todhunter (1865,
pp. 546-548) and Duncan Black (1958, pp. 180-183).
7 Although Black (1958, p. 193) went as far as to deny even the indirect influence of Borda and
Condorcet on Dodgson's theory of committees and elections through Todhunter's (1865) authoritative
account of Borda's and Condorcet's contributions, which "every mathematical lecturer in the country
ought to have studied" in Black's own admission, I found his argument less than persuasive.
Introduction 5

profile of voters' preferences, to be called the assumption of single-peakedpreferences,

under which simple majority voting will be able to determine a social outcome, since
there exists exactly one alternative which will receive a simple majority over any other
alternative, provided that the number of voters is odd, and the Black assumption of
single-peakedness is satisfied. This assumption has a simple geometric representation
to the effect that the utility indicators for the voters' preferences are such that the
social alternatives can be represented by a one-dimensional variable and that each of
the graphs of voters' utility indicators has a single peak. Black's theorem is the first
possibility result of this nature in social choice theory, and it opened up the gate wide
towards the modern development of the theory of voting.
Let us now turn to the welfare economics side of the coin. In this arena too, it
seems fair to say that the real origin of the critical and systematic approach to the
economic mechanism design and policy evaluation belongs to the relatively recent
past, and it may be safely attributed to the work of Jeremy Bentham (1789). He was
a contemporary in England of Borda and Condorcet 8. It is worthwhile to recollect
that Condorcet wrote enthusiastically of the new society of the United States that
"the spectacle of a great people where the rights of man are respected is useful to
all others ... It teaches us that these rights are everywhere the same". He wrote as
well as of the French Revolution that it had "opened up an immense scope to the hopes
of the human species ... [T]his revolution is not in a government, it is in opinions and
wills" 9. In sharp contrast, Bentham, a scholar in law and jurisprudence, was a stark
critic of the concept of inviolable natural rights 10. Indeed, it was in his harsh comment
on the French "Declaration of the Rights of Man and the Citizen", which was embodied
in the French Constitution of 1791, that he wrote the following famous passage:
"[N]atural rights is simple nonsense: natural and imprescriptible rights, rhetorical
nonsense, - nonsense upon stilts [Bentham (1843, p. 501)]". Instead of basing the
economic policies on the concept of inviolable human rights, Bentham took recourse
to the greatest happiness principle, so-called, to the effect that the ultimate criterion
for judging the goodness of an economic mechanism and economic policy is that it can

8 John Hicks (1975, p. 307) was certainly right when he asserted that "[the] 'official' history [of
welfare economics] begins with [Arthur Pigou's] The Economics of Welfare (1920). For it was certainly
Pigou who gave its name to the subject. If it existed before Pigou, it must then have been called
something else". However, the consideration of nomenclature alone should not vitiate the substantial
fact that Pigou's welfare economics is nothing other than the lineal descendent of the long tradition of
the Bentham-Mill-Marshall-Edgeworth-Sidgwick utilitarian calculus.
9 Both citations from Condorcet are due to Rothschild (2001, p. 6).
10 For Bentham, the only category of rights, whose existence he could recognize at all, were those which
depended on law and legislation; a natural right was for him nothing other than a contradiction in terms:
"[T]here are no such things as natural rights-no such things as rights anterior to the establishment of
government no such things as natural rights opposed to, in contradistinction to, legal; [T]he expression
is merely figurative: [W]hen used, in the moment you attempt to give it a literal meaning it leads to
error, and to that sort of error that leads to mischief-to the extremity of mischief [Bentham (1843,
p. 500)]".
6 K. Suzumura

bring about the "greatest happiness of the greatest number". In accordance with this
utilitarian view on the goodness of a state of affairs, the legislator's task is construed to
arrange law and other social and economic institutions so that each person in pursuit
of his own interest will be led to act so as to bring about the greatest happiness for
all persons involved. This utilitarian basis of economic policies permeated the work
of John Stuart Mill, Alfred Marshall, Francis Ysidro Edgeworth, and Henry Sidgwick,
and it served as a natural basis for the synthesis of this tradition by the hands of Arthur
Pigou (1920) in the early twentieth century.
Pigou's so-called "old" welfare economics, being based on the Benthamite-utilitarian
concept of economic welfare, presupposed that the utility of different individuals could
be added to, or subtracted from, one another to define the social objective of total
utility, viz. the greatest happiness . It was against this epistemological basis of Pigou's
"old" welfare economics that a harsh ordinalist criticism raged in the 1930s, kicked off
by a famous essay by Lionel Robbins (1935). Note, however, that Robbins' criticism
boils down to the categorical denial of the possibility of interpersonal comparisons
of utility with interobserver validity; careful reading of Robbins (1935, pp. 138-140,
pp. 149-150); Robbins (1938, pp. 636-637); Robbins (1981, p. 5) convinces us that
he did not reject the possibility of making "subjective" interpersonal comparisons of
utility, nor did he claim that economists should not make "subjective" interpersonal
comparisons of their own. What he actually asserted is that "subjective" interpersonal
comparisons cannot claim any "objective" interpersonal validity.
By the end of the 1930s, it became widely recognized that the foundations of
Pigou's "old" welfare economics were hopelessly eroded, and new foundations for
welfare economics had to be discovered on the basis of ordinal and interpersonally
non-comparable utility information, and nothing else, in order to salvage something
of substance from the vestige of Pigou's theoretical superstructure. This is the same

I1 At this juncture, two remarks seem to be in order. In the first place, while Pigou in principle subscribed
to the utilitarian viewpoint, careful reading of The Economics of Welfare reveals how discriminatingly
was the use he actually made of it. Having said this, however, it should be pointed out that Pigou's
discussions of tax-subsidy policies related to externalities, with which he is much associated, were
directly derived through a utilitarian way of reasoning. It is true that Pigou's use of the utilitarian
principle is not as conspicuous in reference to income distribution as was the case with Edgeworth, but
it was in fact Pigou who inspired Hugh Dalton's (1920) famous utilitarian measure of inequality. In
the second place, unlike Bentham, who was strongly and outspokenly against the idea of natural rights,
which goes squarely against the foundations of utilitarianism, Pigou (1920, 1952 edition, p. 759) made an
early use of the non-welfarist notion of individual rights when he discussed people's claim to "minimum
standard of real income", which "must be conceived, not as a subjective minimum of satisfaction, but as
an objective minimum of conditions". Pigou's characterization of "an objective minimum of conditions"
is close to what we now call the "basic needs", which consist of "some defined quantity and quality of
house accommodation, of medical care, of education, of food, of leisure, of the apparatus of sanitary
convenience and safety where work is carried on... " Pigou might have thought that such rights could be
justified on utilitarian grounds in the Benthamite tradition of regarding rights as intrinsically unimportant,
but instrumentally crucial, but The Economics of Welfare is completely reticent concerning the utilitarian
justification of these rights.
Introduction 7

informational basis as that of the Borda-Condorcet theory of collective decision-

making, which is a slightly ironical fact in view of the sharply contrasting background
of the Borda-Condorcet theory on the methods of collective decision-making, on the
one hand, and the Bentham-Pigou theory on the enhancement of social welfare, on
the other.
The first ordinalist response to this plea was to go back to the ordinalist tradition
pioneered by Vilfredo Pareto (1906, 1913), and invoke the seminal concept of the
Pareto principle to the effect that a change from one social state to another social
state can be judged as socially good if at least one individual is thereby made
better off without making anybody else worse off in return. The characterization and
implementation of the Pareto efficient resource allocation became the central exercise
in this phase of the "new" welfare economics, which may be duly represented by John
Hicks (1939). Note, however, that almost every economic policy cannot but favour
some individuals at the cost of disfavouring some others, so that there would be almost
no situation of real importance where the Pareto principle could claim relevance in
isolation.
It was against this background that two distinct approaches were explored to
rectify the unsatisfactory state of the post-Pigovian "new" welfare economics. The
first approach was the introduction of compensation criteria by Nicholas Kaldor
(1939), John Hicks (1940), Tibor Scitovsky (1941) and Paul Samuelson (1950),
which endeavoured to expand the applicability of the Pareto principle by introducing
hypothetical compensatory payments between gainers and losers from a change in
economic policy 1 2 . According to Johannes de V Graaff (1957, pp. 84-85), "[t]he
compensation tests all spring from a desire to see what can be said about social
welfare or 'real national income' ... without making interpersonal comparisons of
well-being ... They have a common origin in Pareto's definition of an increase in
social welfare ... but they are extended to situations in which some people are made
worse off".
The second approach was the introduction of the concept of a social welfare
function by Abram Bergson (1938) and Paul Samuelson (1947, Chapter VIII), which
is deeply rooted in the belief that the pursuit of the logical consequences of any value
judgements, irrespective of whose ethical beliefs they represent, whether or not they are
widely shared in the society, or how they are generated in the first place, is a legitimate
task of welfare economics. The social welfare function is meant to be the formal way
of encompassing such an ethical belief. It was in terms of this concept of a social
welfare function that Bergson and Samuelson tried to separate what belongs to the
area of ethics, about which economists qua scientists do not have any qualification to

12 According to John Chipman and James Moore (1978, p. 548, footnote 2), Enrico Barone (1908,
1935) had developed the compensation principle much earlier than Kaldor and Hicks, "who mentioned
it no less than four times". Barone's pioneering contribution was left unnoticed among English speaking
economists, however, even after the Italian original was translated into English in von Hayek (1935).
8 K. Suzulmura

say anything objective whatsoever, from what belongs to the area of welfare economics,
about which economists as scientists have every reason as well as obligation to say
something of objective validity 13.
Between these two schools of the "new" welfare economics, the former compensa-
tionist school met serious logical difficulties. Even before the scaffolds for construction
were removed from the construction site, serious logical contradictions in the form
of either the lack of asymmetry, or the lack of transitivity, could be found in the
social welfare judgements based on the Kaldor-Hicks-Scitovsky compensation criteria
by Tibor Scitovsky (1941), William Gorman (1955) and many others, which fatally
vitiated the credibility of the "new" welfare economics of the compensationist school.
The verdict on the Samuelson compensation principle, which was defined in terms of
a uniform outward shift of the utility possibility frontier, is quite different. Indeed,
the Samuelson compensation principle can always generate transitive social welfare
judgements, so that its logical performance in isolation is impeccable. Nevertheless,
it may still generate contradictory social welfare judgements in combination with
the Pareto principle 14. On the other hand, the second school of the "new" welfare
economics, which is founded on the Bergson-Samuelson social welfare function, has
been widely praised as the culmination of the ordinalist "scientific" approach to welfare
economics 15
Broadly speaking, this was the intellectual atmosphere surrounding social choice
theory when Kenneth Arrow published his Ph.D. Dissertation, Social Choice and

13 The genesis of the Bergson-Samuelson social welfare function was traced as far back as Pareto (1913)
by Chipman (1976) and Chipman and Moore (1978). True enough, Pareto was remarkably ahead of his
time, and sympathetic eyes may catch the glimpse of social welfare function in Pareto's early writings.
Nevertheless, it seems fair to say that, without Bergson (1938) and Samuelson (1947, Chapter VIII),
the concept of social welfare function could not have established itself as the central piece of modern
welfare economics. It is in this sense that Samuelson (1981, p. 248) is absolutely right when he wrote in
a related context that afterte, and only after, you have worked out a clear understanding of this subject
are you able to recognize the bits of the puzzle that Pareto had already discerned".
14 Let Pp, Ps and P stand, respectively, for the Pareto superiority relation, the Samuelson superiority
relation, and the social preference relation. The social preference relation is said to respect the Pareto
superiority relation as well as the Samuelson superiority relation if and only if it satisfies Pp C P and
Ps C P. It was shown by Suzumura (1980, 1999b) that there exists a situation, which is not concocted at
all, where we have four social states, say x, y, z and w, such that xPpy, zPpw, yPsz and wPx hold. If the
social preference relation respects the Pareto superiority relation as well as the Samuelson superiority
relation, then we cannot but obtain xPv, yPz, zPw and wPx, which clearly vindicate the social preference
cycle.
t5 Thus, Samuelson (1981, p. 223) could assert without any reservation the following: "As I write, the
new welfare economics is just over four decades old. This subject, in its essentials as we know it today,
was born when the 24-year-old Abram Bergson - then still a Harvard graduate student - wrote his
classic 1938 QuarterlyJournal of Economics article. To one like myself, who before 1938 knew all the
relevant literature on welfare economics and just could not make coherent sense of it, Bergson's work
came like a flash of lightning, describable only in the words of the pontifical poet:
Nature and Nature's laws lay hid in night:
God said, Let Newton be! and all was light."
Introduction 9

Individual Values, in 1951. In view of its innovative nature as well as the revolutionary
influence it exerted on the whole field of social choice theory, it will be justifiable to
devote the next section in its entirety to this work.
Quite apart from the Robbinsian criticism, which is epistemological in nature, there
is a fundamental criticism of, and a proposal for a serious alternative to, the Benthamite
utilitarianism by John Rawls (1962, 1963, 1971), which is focused directly on the
ethical nature of the Benthamite outcome morality. According to Rawls (1971, p. 22),
the main idea of classical utilitarianism is that "society is rightly ordered, and therefore
just, when its major institutions are arranged so as to achieve the greatest net balance of
satisfaction summed over all the individuals belonging to it". Not only is this classical
principle based on welfarism to the effect that "[t]he judgment of the relative goodness
of alternative states of affairs must be based exclusively on, and taken as an increasing
function of, the respective collections of individual utilities in these states", but also
it invokes the aggregation rule of sum-ranking to the effect that "[o]ne collection
of individual utilities is at least as good as another if and only if it has at least as
large a sum total [Sen (1979, p. 468)]". Rawls criticises the informational basis of
welfarism and proposes the alternative informational basis of social primary goods,
viz. "things that every rational man is presumed to want", which "normally have a use
whatever a person's rational plan of life [Rawls (1971, p. 62)]". Rawls also criticises
the utilitarian aggregation rule of sum-ranking for its being "indifferent as to how a
constant sum of benefits is distributed [Rawls (1971, p. 77)]". His proposed alternative
to the Benthamite utilitarianism is such that "[a]ll social primary goods - liberty and
opportunity, income and wealth, and the bases of self-respect - are to be distributed
equally unless an unequal distribution of any or all of these goods is to the advantage
of the least favored [Rawls (1971, p. 303)]". His own justification of this principle
of justice makes use of a hypothetical situation called the original position, where
individuals choose the basic principles of the society behind the veil of ignorance,
viz. without knowing their own position in the resulting social order as well as being
ignorant of their personal identities. In such a situation of primordial equality, Rawls
claims that his principles of justice would be generally accepted as a fair agreement
in the absence of ethically irrelevant vested interests 16
The invocation of the logical device of primordial stage of ignorance with the
purpose of securing a fair field for designing a set of social rules is not original to
Rawls. Other notable examples are William Vickrey (1945, 1960) and John Harsanyi
(1953, 1955, 1977), who respectively made use of the same device to find ajustification
for the Benthamite utilitarianism. Vickrey (1945) gave a brief, yet clear first statement

16 Rawls' theory of "justice as fairness" exerted a strong influence on the contemporary welfare
economics ingeneral, and social choice theory inparticular. But it is predominantly, if not exclusively,
in the modified welfaristic version in which the Rawlsian concern with the well-being of the least
favored individual is expressed with reference to the individuals' welfare levels, which are assumed to
be interpersonally comparable. Needless to say, Rawls' own "difference principle" focuses directly on
the minimal availability of "social primary goods", and not on the minimal individual welfare.
10 K. Suzumura

of the original position idea. Harsanyi (1955) proved the following important theorem:
Suppose that social preferences as well as individual preferences satisfy the von
Neumann-Morgenstern postulates of rationality, and if all individuals being indifferent
implies social indifference, then social welfare must be the weighted sum of individual
utilities. Under the additional requirement of anonymity, the Harsanyi representation
for social welfare boils down to the unweighted sum-total of individual utilities, viz.
the classical utilitarianism 17.

2. Social choice and individual values

Without denying the importance of those pioneering contributions made by many

precursors, it seems fair to say that Kenneth Arrow's Social Choice and Individual
Values elevated social choice theory to a stage which is qualitatively different
altogether.
To lend concrete substance to our sweeping assertion, let us start by referring to
the pioneering studies of voting schemes by Condorcet, Borda, Dodgson, Black, and
many others again. Important though these celebrated works are, it is undeniable that
their studies were concerned exclusively with some specified voting schemes such as
the method of simple majority voting, the Borda method, the Dodgson method, and
so forth. In sharp contrast, Arrow (1950, 1951, 1984) developed an analytical method
which allowed him to treat all conceivable voting schemes simultaneously within one
unified conceptual framework. To bring the importance of this development into clearer
relief, consider the simplest imaginable society with only two individuals, say 1 and
2, and three alternative social states, say x, y and z. Let us simplify our arena further
by assuming away individual as well as social indifference relations altogether. It is
clear, then, that there exist six distinct preference orderings of three social states 18:

a: x,y,z /3: x,z,y y: y,x,z 5: yzx C: z,x,y : z,y,x.

Each one of these orderings can represent an individual preference ordering for 1
and 2 over three social states. What Arrow christened the social welfare function, or
constitution in his more recent terminology, is a function which maps each profile
of individual preference orderings into a unique social preference ordering, which
is meant to denote the process or rule for aggregating each profile of individual

17 However, as Sen (1976c) acutely pointed out, utility is only used to represent preferences in the
theorem of Harsanyi (1955). Thus, there is ample room for reservation on the claim that Harsanyi's
argument can be interpreted as being an argument in support of utilitarianism. See also Prasanta
Pattanaik (1968).
18 Alternatives are arranged horizontally, the more preferred alternative being to the left of the less
preferred. Thus, the preference ordering a means that x is preferred to y, y is preferred to z, hence x is
preferred to z.
Introduction 11

preference orderings into a social preference ordering. In other words, a social welfare
function is a mapping defined on the Cartesian product A x A, where A = {a, 1i, y, 6,
e, }, and takes its values on A. Thus, even in our simplest conceivable society, there
exist 636 social welfare functions in the sense of Arrow, which is an astronomically
large number indeed (roughly 1028). It is clearly impossible to check all these Arrovian
social welfare functions one by one for their democratic legitimacy, on the one hand,
and for informational efficiency, on the other. Instead of attempting to cope with this
clearly hopeless task, Arrow pioneered the axiomatic approach in social choice theory,
which enabled him to analyse these 636 Arrovian social welfare functions all at once,
by imposing a set of axioms which are deemed necessary for the Arrovian social
welfare functions to be reasonable, hence acceptable. It is this novel methodology
which enabled him to analyse all the relevant social welfare functions at one stroke,
and led him to the celebrated generalpossibilitytheorem, or the Arrovian impossibility
theorem in the currently prevailing terminology, to the effect that there exists no social
welfare function satisfying a set of conditions necessary for democratic legitimacy and
informational efficiency.
The novelty of Arrow's approach is no less conspicuous in the context of the "new"
welfare economics as well. For Bergson and Samuelson, their social welfare function
was an analytical device for separating what should duly belong to economics from
what should duly be relegated to ethics. According to Samuelson (1947, p. 220-221),
"[i]t is a legitimate exercise of economic analysis to examine the consequences of
various value judgments, whether or not they are shared by the theorist, just as the
study of comparative ethics is itself a science like any other branch of anthropology".
It was as an analytical vehicle for implementing this "scientific" research program
of "new" welfare economics that Samuelson invoked what came to be known as the
Bergson-Samuelson social welfare function: "Without inquiring into its origins, we
take as a starting point for our discussion a function of all the economic magnitudes
of a system which is supposed to characterize some ethical belief- that of a benevolent
despot, or a complete egoist, or 'all men of good will', a misanthrope, the state, race,
or group mind, God, etc. Any possible opinion is admissible ... We only require that
the belief be such as to admit of an unequivocal answer as to whether one configuration
of the economic system is "better" or "worse" than any other or "indifferent", and that
the relationships are transitive ... "
In contrast with the Bergson-Samuelson social welfare function, which Bergson
and Samuelson assumed to be given from outside of economics, Arrow was of the
conviction that the process or rule through which the social value to be represented by
the Bergson-Samuelson social welfare function is formed should also be the subject
of logical scrutiny. In other words, in order for the economic analysis not to lose
social relevance, it is necessary that the process or rule for constructing the Bergson-
Samuelson social welfare function on the basis of individual judgments of the goodness
of the social states, viz. the Arrow social welfare function in this arena, must satisfy
the minimal requirements of democratic legitimacy and informational efficiency.
Interpreted in this new arena, the Arrow impossibility theorem turns out to be a
12 K. Suzumtnura

basic criticism against the foundations of "new" welfare economics of the Bergson-
Samuelson family. No wonder Arrow's theorem caused a stir among many reputable
economists who created and promoted the "new" welfare economics. For example, Ian
Little (1952, pp. 423-424) contrasted Bergson's and Arrow's social welfare functions
with the purpose of criticizing the latter as follows: "Bergson's welfare function was
meant as a 'process or rule' which would indicate the best economic state as a function
of a changing environment (i.e. changing sets of possibilities defined by different
economic transformation functions), the individuals' tastes being given. ... If tastes
change, we must expect a new ordering of all the conceivable states; but we do not
require that the difference between the new and the old orderings should bear any
particular relation to the changes of taste which have occurred. We have, so to speak, a
new world and a new order; and we do not demand correspondence between the change
in the world and the change in the order. ... Traditionally, tastes are given; indeed, one
might almost say that the given individuals are traditionally defined as the possessors
of the given tastes and that no sense is attached to the notion of given individuals with
changing tastes" t9. Samuelson (1967, p. 42), who has always been the most eloquent
advocate of the Bergson-Samuelson school of "new" welfare economics, went as far as
to declare that "the Arrow result is much more a contribution to the infant discipline of
mathematical politics than to the traditional mathematical theory of welfare economics.
I export Arrow from economics to politics because I do not believe that he has proved
the impossibility of the traditional Bergson welfare function of economics, even though
many of his less expert readers seem inevitably drawn into thinking so" 20 .
What, then, are the axioms of democratic legitimacy and informational efficiency
which Arrow demonstrated to be logically incompatible? In the 1963 revised version
of the theorem [Arrow (1963, pp. 96-97; 1987)], there are four transparent axioms
altogether. The first axiom is that each and every individual is free to form and
express whatever preference ordering he/she cares to specify, which represents his/her
evaluations of the goodness of social states, and the Arrow social welfare function
must be robust enough to be able to aggregate the profile of these individual preference
orderings into a social preference ordering. The second axiom requires that the Arrow

19 Little's criticism to this effect was strongly supported by Samuelson (1967, pp. 48-49): "For Bergson,
one and only one of the ... possible patterns of individuals' orderings is needed. It could be any one,
but it is only one. From it (not from each of them all) comes a social ordering ... The only Axiom
restricting a Bergson Social Welfare Function (of individualistic type) is a 'tree' property of Pareto-
optimality type". It is this sharp contrast between the Arrow social welfare function and the Bergson
social welfare function that created the widespread perception that the Arrow impossibility theorem,
which requires the full force of multiple profiles of individual preference orderings, does not apply to
the Bergson social welfare function which is rooted in the single profile framework.
20 To keep the record straight, let us emphasize that the Arrovian impossibility theorem is not a theorem
which negates the existence of the Bergson-Samuelson social welfare ordering; it is a theorem which
negates the existence of a "reasonable" process or rule which can associate a Bergson-Samuelson social
welfare ordering with each profile of individual preference orderings. See Suzumura (1976, 1987) and
Arrow (1983).
Introduction 13

social welfare function must faithfully reflect the unanimous preference expressed by
all individuals over a pair of social states, which makes the process or rule of preference
aggregation minimally democratic. The third axiom requires that the Arrow social
welfare function must be informationally efficient in that, in deciding whether one
social state is better than, or worse than, or indifferent to another social state, it is
necessary and sufficient to know how individuals rank just these two alternative social
states vis-A-vis each other. The fourth and the least controversial axiom requires that
there should be no dictator in the society, who can decide a strict social preference for
a social state vis-A-vis another social state simply by expressing his personal preference
for the former state against the latter.
It is worth emphasizing that these demonstrably contradictory axioms are nothing
other than the lineal descendents of what preceded Social Choice and Individual
Values. Indeed, in the context of the methods of collective decision-making, the method
of simple majority voting satisfies all of the Arrovian conditions except that the
generated social preference relation lacks the general assurance of transitivity by virtue
of the Condorcet paradox. In the alternative context of the foundations of welfare
economics, the "new" welfare economics of the compensationist school of thought,
as well as of the Bergson-Samuelson school of thought, is founded squarely on the
ordinal and interpersonally non-comparable informational basis; it is also deeply rooted
in the Paretian tradition to the effect of requiring social preference to reflect unanimous
individual preferences faithfully. Because it respected the preceding tradition, the
Arrow impossibility theorem was made not only more relevant, but also a clear
indicator of the need of systematic scrutiny in the search for reasonable resolutions
of the logical contradiction thereby identified. It is in this sense that the message of
Arrow's general impossibility theorem is clearly positive, rather than negative.
Arrow (1951, Chapter VII) also made another important contribution by developing
a systematic logical method in the analysis of simple majority voting, which enabled
him to pursue Black's geometric idea of single peaked preferences in the general case
of any number of alternatives. This neat method of analysis enabled his successors
to introduce some other restrictions on the admissible profiles of voters' preferences
under which the method of simple majority voting can escape from the Condorcet
paradox. Indeed, it was this method of analysis which eventually led Ken-Ichi Inada
(1969), on the one hand, and Sen and Pattanaik (1969) on the other, to discover the
necessary and sufficient conditions for this method of collective decision-making to
work satisfactorily.

3. "Socialist planning" controversy

There is another controversy of historical importance, which was fought mainly in

the 1930s. Maurice Dobb (1969, p. 183) had the strong opinion that "[t]he old debate
about Wirtschaftsrechnung ... is nowadays sufficiently familiar ... for any suggestion
of revisiting it to invite disinclination rather than attention". Nevertheless, it seems to
14 K. Suzurnul

us that there are several lessons of this harsh controversy with lasting importance in
the evolution and orientation of the theory of decentralized planning procedures la
Edmond Malinvaud (1967) and Geoffrey Heal (1973), as well as of the related branch
of social choice theory called the implementation theory, or of the theory of mechanism
design, la Leonid Hurwicz (1960, 1972, 1973) and Eric Maskin (1979, 1999).
It was Ludwig von Mises (1920) who kicked off this controversy. In his under-
standing, rational economic calculation is possible only when monetary prices exist,
not only for consumption goods, but also for production goods of any order, since it is
monetary calculation which "affords us a guide through the oppressive plentitude of
economic potentialities ... It renders their value capable of computation and thereby
gives us the primary basis for all economic operations with goods of a higher order
[von Hayek (1935, p. 101)]." According to von Mises, however, it is impossible to
find necessary monetary prices for production goods of a higher order in a socialist
state, because no production good will ever become the object of market exchange
in a socialist state where, by definition, collective ownership prevails for all means of
production.
It is clear that the impossibility thesis Ala von Mises holds if and only if there are no
prices for production goods in a socialist state with collective ownership of the means
of production. It seemed obvious to Oskar Lange (1938, p. 61) that the latter thesis was
clearly false: "Professor Mises seems to have confused prices in the narrower sense,
i.e. the exchange ratios of commodities on a market, with prices in the wider sense of
'terms on which alternatives are offered' ... It is only in the latter sense that 'prices'
are indispensable for the allocation of resources ... " As Lange correctly pointed
out, "prices in the generalized sense," or "efficiency prices" in the circumlocution of
modern economic theory, exist irrespective of the ownership structure of the means of
production. This fact alone was enough to eradicate the impossibility thesis A la von
Mises.
However, the controversy resurged in the hands of Friedrich von Hayek (1935,
1944, 1948), taking a more sophisticated form. Unlike von Mises, von Hayek never
denied the theoretical existence of efficiency prices for all goods including the means
of production, which, if made available, would enable a socialist state to attain a
rational allocation of resources. The problem which von Hayek pointed out, and made
the foundations of his impossibility thesis, was how such efficiency prices could be
made available in practice: "[T]his is not an impossibility in the sense that it is
logically contradictory. But to argue that a determination of prices ... being logically
conceivable in any way invalidates the contention that it is not a possible solution,
only proves that the real nature of the problem has not been perceived [von Hayek
(1935, pp. 207-208)]." To understand why, von Hayek urges us to visualize what the
determination of efficiency prices by computational method would imply in practice:
"It is clear that any such solution would have to be based on the solution of some
such system of equations [for general economic equilibrium] as that developed in
[Enrico] Barone's article [Barone (1908)] ... [W]hat is practically relevant ... is not
the formal structure of this system, but the nature and amount of concrete information
Introduction 15

required if a numerical solution is to be attempted and the magnitude of the task

which this numerical solution must involve ... [von Hayek (1935, p. 208)]". To
calculate efficiency prices by solving the general equilibrium equations, we must gather
information about technology, primary and intermediate resources, and consumers'
preferences, which are widely dispersed and privately owned by numerous economic
agents. Given the nature and complexity of this privately held information, it would be
prohibitively difficult, if not logically impossible, to motivate numerous private agents
to comply with the request from the central planning board and submit this information
faithfully for the purpose of computing efficiency prices. Thus, von Hayek concludes,
"[i]t is probably evident that the mere assembly of these data is a task beyond human
capacity [von Hayek (1935, p. 211)]". To make this situation even worse, "[m]ost
of [the technical information] consists in a technique of thought which enables the
individual engineer to find new solutions rapidly as soon as he is confronted with new
constellations of circumstances [von Hayek (1935, pp. 210-211)]". This is the essence
of the impossibility thesis a la von Hayek.
Once again, Lange was ready to confront von Hayek's impossibility thesis.
Capitalizing and elaborating on the earlier works by Enrico Barone (1908) and Fred
M. Taylor (1929), Lange developed a sophisticated trial and error method of price
adjustment in a socialist state. To see how he designed this scheme, the so-called
Lange-Lerner market socialism after Oscar Lange (1936-1937) and Abba Lerner
(1944), and how this scheme fares with respect to some performance criteria, is
useful in identifying the areas of research called the theory of decentralized planning
procedures and the theory of mechanism design.
Lange assumed a socialist state where freedom of choice in consumption and
freedom of choice of occupation are guaranteed, and the preferences of consumers are
the guiding criteria in production and in the allocation of resources. In this system,
there exist market prices for consumption goods and for labour services, but the
prices for capital goods and productive resources other than labour are prices in the
generalized sense, i.e. mere accounting prices. Some appropriate rules are applied to
the distribution of social dividend to the consumers. Subject to these rules of income
formation and given market prices, the consumers are free to choose their demand for
consumption goods and supply of labour services. Likewise, some appropriate rules
are applied to the production units (in industry with many firms incurring set-up costs)
so that average cost of production will be minimized, and marginal cost will be made
equal to the price of the product for each and every good produced. The accounting
prices for capital goods and productive resources other than labour are formed and
adjusted by the Central Planning Board through the instrumental use of the Walrasian
tatonnement process, where the Central Planning Board plays the role of the Walrasian
auctioneer. The modus operandi of this successive trial and error process is exactly
the same as the well-known Walrasian ttonnement process, and the adjustment of the
market price or the accounting price for each good and service are made in accordance
with the aggregate excess demand for the good and service in question.
Two properties of this pseudo-Walrasian tatonnement process deserve particular
16 K. Suzunlura

attention. In the first place, it enables the Central Planning Board to escape from the
Hayekian task of gathering dispersed private information for computing accounting
prices at the centre, which von Hayek maintained to be practically impossible to
perform, since the necessary computation is in effect performed by each and every
holder of private information. In the second place, the accounting prices found at the
equilibrium of this pseudo-Walrasian tatonnement process in a socialist state "have
quite the same objective character as the market prices in the regime of competition.
Any mistake made by the Central Planning Board in fixing prices would announce
itself in a very objective way - by a physical shortage or surplus of the quantity
of the commodity or resources in question - and would have to be corrected in
order to keep production running smoothly [Lange (1938, p. 82)]". On the basis
of these nice properties of his scheme, Lange concluded that "a substitution of
planning for the functions of the market is quite possible and workable", and the
immediate successors of the lessons of the controversy gladly concurred. Indeed, "[a]s
far as economics profession is concerned", wrote Paul Sweezy (1949, p. 232) in
the Economics Handbook Series edited by Seymour Harris, "Lange's paper may be
regarded as having finally removed any doubts about the capacity of socialism to utilize
resources rationally". Upon careful scrutiny, however, this sweeping verdict turns out
to be untenable, to say the least.
To begin with, for the quasi-Walrasian ttonnement process to serve as an algorithm
for finding right market prices and accounting prices, it must be guaranteed to converge
surely and rapidly to the system of general equilibrium prices. Unless some very
special assumptions, such as gross substitutability, or the weak axiom of revealed
preference, are imposed on the aggregate excess demand functions, however, there
is no guarantee for the global stability of the Lange process of price adjustment2 .
In a postscript to the controversy written thirty years later, Lange (1967, p. 158)
wrote that "[i]t was assumed without question that the ttonnement process in fact
converges to the system of equilibrium prices." Since there is no general guarantee of
such a convergence property, the Lange-Lerner scheme of market socialism offers no
assurance of non-wasteful workability 22 . More remarkably, Lange went on to maintain
that "[w]ere I to rewrite my essay today my task would be much simpler. My answer
to Hayek and Robbins would be: so what's the trouble? Let us put the simultaneous
equations on an electric computer and we shall obtain the solution in less than a
second. The market process with its cumbersome ttonnements appears old-fashioned.
Indeed, it may be considered as a computing device of the pre-electronic age". This

21 Herbert Scarf (1960) constructed an explicit example where the competitive equilibrium is globally
unstable. See also Takashi Negishi (1962).
22 As far as the relative performance of the competitive market economy and the Lange-Lerner scheme
of market socialism is concerned, this objection is a double-edged sword; it applies not only to the
Lange-Lerner scheme of market socialism, but also to the competitive market economy. But the basic
fact remains that the Lange-Lerner scheme is not successful as a decentralized algorithm for computing
a general equilibrium solution in a socialist state, as it was originally meant to be.
Introduction 17

statement is truly remarkable, as it "proves that the real nature of the problem has
not been perceived". Recollect that the impossibility thesis a la von Hayek was based
not on the limitation of computational capacity on the part of the Central Planning
Board, but on the prohibitive difficulty of gathering dispersed and privately owned
information for the purpose of central computation. Needless to say, no computer
with whatever capacity can work without being provided with the relevant data.
Interestingly enough, Abram Bergson (1967, pp. 663-664) also posed the possibility
of avoiding trial and error procedure by solving pertinent equations by means of
mathematical techniques: "[B]oth Lange and [Dickinson] wrote before the age of
electronic computers. Given this technology, could not the [Central Planning Board],
in performing its cardinal task of fixing prices, confute Hayek after all simply by using
mathematical techniques?" However, Bergson was far more careful in answering this
question than Lange: "[S]hould the Board seek to employ mathematical procedures
in fixing prices comprehensively and in detail, its undertaking surely could become
burdensome for managers of production units, who might be called on to predict and
articulate in inordinately concrete detail the complex and ever changing constraints
and opportunities that confront them, and on this basis to communicate to the Board
such data on these matters as the Board would require; and for the Board itself, which
promptly would have to digest such information and to communicate the results of
its deliberations to the managers. The capacities of managers as well as of the Board
to grapple with these tasks might often be enhanced by use of computers, but not
always".
Secondly, there is no systemic device in the Lange-Lerner scheme of market
socialism to confront the possibility of strategic behaviour by private agents. As Lange
(1938, p. 81) rightly observed, "[o]n a competitive market the parametric function of
prices results from the number of competing individuals being too large to enable
any one to influence prices by his own action. In a socialist economy, production and
ownership of the productive resource outside of labour being centralized, the managers
certainly can and do influence prices by their decisions. Therefore, the parametric
function of prices must be imposed on them by the Central Planning Board as an
accounting rule. All accounting has to be done as if prices were independent of the
decisions taken. For purposes of accounting, prices must be treated as constant, as
they are treated by entrepreneurs on a competitive market". Since there is nothing
in the Lange-Lerner scheme to make this accounting rule compatible with the private
incentives of individual agents, we cannot but conclude that the Lange-Lerner scheme
of market socialism lacks the important property of incentive compatibility.
Thirdly, the Lange-Lerner market socialism is designed for the single-minded
purpose of enabling a socialist state to use its endowed scarce resources efficiently.
As was aptly observed by Sweezy (1949, p. 233), "[p]erhaps the most striking feature
of Lange's model is that the function of the Central Planning Board is virtually confined
to providing a substitute for the market as the coordinator of the activities of the various
plants and industries. The truth is that Lange's Board is not a planning agency at all
but a price-fixing agency; in his model production decisions are left to a myriad of
18 K. Sutzumura

essentially independent units, just as they are under capitalism". It is true that achieving
the efficient use of scarce resources is a task of no mean difficulty, but "[t]he common
features of all collectivist systems may be described ... as the deliberate organisation
of the labours of society for a definite social goal. That our present society lacks such
'conscious' direction towards a single aim, that its activities are guided by the whims
and fancies of irresponsible individuals, has always been one of the main complaints
of its socialist critics [von Hayek (1944, p. 42)]". If we take this observation at all
seriously, we must go beyond mere efficiency and proceed to optimality with reference
to the single social goal in order to have a fully-fledged design of a rational collectivist
society.
If we retain, as in the Lange-Lerner scheme of market socialism, the crucial value
premise of consumers' sovereignty and want to orient a socialist state towards a definite
social goal beyond the mere attainment of efficient allocation of scarce resources,
we must find a process or rule to construct a conscious social goal on the basis
of individual judgements on what constitutes social goods, since "[t]he effect of the
people agreeing that there must be central planning, without agreeing on the ends, will
be rather as if a group of people were to commit themselves to take a journey together
without agreeing where they want to go [von Hayek (1944, p. 46)]". This is precisely
the same problem posed and settled in the negative by Arrow in a related but distinct
context of collective choice and social welfare. Interestingly enough, von Hayek (1944,
p. 44) observed that forming "a definite social goal" for its use in orienting central
planning "would be impossible for any mind to comprehend the infinite variety of
different needs of different people which compete for the available resources and to
attach a definite weight to each". See also Leif Johansen (1969), who examined the
relevance of Arrow's impossibility theorem in the context of economic planning.
These negative observations notwithstanding, it should be emphasized that the
"socialist planning" controversy, in which both Lange and von Hayek played major
roles, was the first serious attempt at designing an alternative economic mechanism
with the purpose of satisfying some concrete performance characteristics. In so doing,
they became the modem forerunners in the theory of decentralized planning procedures
and the theory of mechanism design.

4. Significance of the subject and main lines of research

Enough has been said so far about the historical background of social choice theory.
It remains for us to emphasize the significance of the subject, and identify the major
lines of research in this broad and interdisciplinary area.
Ever since the appearance of Social Choice and Individual Values, the growth of
social choice theory along many distinct lines of research has been quite conspicuous,
especially after the 1960s. By now, there is an extensive Social Choice Bibliography
prepared and regularly updated by Jerry Kelly (htpp://www.maxwell.syr.edu/maxpages/
faculty/jskelly/biblioho.htm), which is more than 300 pages in single-space printout.
Introduction 19

Even this extensive and invaluable Kelly Bibliography does not cover some of the
issues treated in this Handbook of Social Choice and Welfare in full, whereas there
are many other issues which are included in the Kelly Bibliography but not in this
Handbook. The plan of this Handbook clearly reflects our perception of the special
significance of the development along the three lines of research which we have
identified in our account of the historical evolution of social choice theory: the methods
of collective decision-making, the theoretical foundations of welfare economics, and
the theory of incentive compatibility and mechanism design. To explain why we believe
these issues to be of special significance, it is useful to go back to Social Choice and
Individual Values once again.
To begin with, note that Arrow's theory connected social choice and a social
preference ordering, which the Arrow social welfare function associates with each
profile of individual preference orderings, through the assumption of collective
rationality: Given any set of available social states, the society chooses that available
social state which is at least as good as any other available social state, where the
judgements of the goodness of social states are performed in terms of the social
preference ordering. This crucial assumption has been one of the major targets for
critics of the Arrovian framework of social choice theory. Most notable is the criticism
by James Buchanan (1954, p. 116), according to whom "[t]he mere introduction of
the idea of social rationality suggests the fundamental philosophical issues involved.
Rationality or irrationality as an attribute of the social group implies the imputation
to the group of an organic existence apart from that of its individual components ...
We may adopt the philosophical bases of individualism in which the individual is the
only entity possessing ends or values. In this case no question of social or collective
rationality may be raised. A social value scale simply does not exist. Alternatively,
we may adopt some variant of the organic philosophical assumption in which the
collectivity is an independent entity possessing its own value ordering. It is legitimate
to test the rationality or irrationality of this entity only against this value ordering".
Two avenues of research were explored in response to this early criticism, in
order to check the robustness of the Arrovian impossibility theorems with respect
to the assumption of collective rationality. The first avenue maintained the definition
of social choice in terms of the optimization of the social preference relation, but
weakened the required degree of collective rationality. Weakening Arrow's requirement
of completeness as well as transitivity of social preference relation, one may want
to discard the exacting requirement of transitivity of the indifference relation, and
retain only the more defensible requirement of transitivity of the strict preference
relation (to be called quasi-transitivity);one may also go one step further and weaken
the requirement of quasi-transitivity, and settle with only the non-existence of any
strict preference cycle (to be called acyclicity). The second avenue went further and
discarded the assumption of collective rationality altogether; it focused directly on
social choice which has no underlying social preference relation, and imposed some
choice-consistency property, an important example thereof being path-independence:
"the independence of the final choice from the path to it [Arrow (1963, p. 120)]".
20 K. Suzinura

These two avenues were pioneered and vigorously explored by Sen (1969, 1970a
Chapter 4*; 1977a); his leading attempts were followed by Andreu Mas-Colell and
Hugo Sonnenschein (1972), Charles Plott (1973), Douglas Blair, Georges Bordes, Jerry
Kelly, and Kotaro Suzumura (1976), Suzumura (1983, Chapter 3), and many others.
Basically, however, these extensive researches confirmed the robustness of the Arrovian
impossibility theorems. As Arrow (1963, p. 109) has observed in a related but distinct
context, "[t]he paradox of social choice cannot be so easily exorcised".
The next crucial step in the search for an escape route from Arrow's impossibility
theorem was to explore the use and usefulness of interpersonal comparisons of
utilities, with or without cardinal measurability 23. The context in which we can
meaningfully talk about this potential escape route is one where an ethical observer
forms his own subjective interpersonal comparisons of utilities, and makes use of this
extended informational basis to define an essentially Arrovian social welfare function.
A fruitful and systematic method of analysis was developed mainly in the 1970s by
Sen (1970a, 1973, 1977b), Peter Hammond (1976), Claude d'Aspremont and Louis
Gevers (1977), and Eric Maskin (1978), among many others, which brought about a
neat axiomatization of the Rawlsian difference principle (in its welfaristic version) as
well as of the Benthamite principle of utilitarianism. This is a legitimate way out from
the Arrovian impossibility theorem in the context of forming someone's social welfare
judgements, but such an escape route is surely not available in the alternative context of
collective decision-making. Even in the context of forming social welfare judgements,
the phantom of Lionel Robbins cannot be exorcised so easily; if there are multiple
ethical observers who form their respective subjective interpersonal comparisons of
utilities, their social welfare judgements may well conflict with each other so much
so that some variants of the Arrovian impossibility theorems may well come back
strenuously, as was demonstrated by Kevin Roberts (1980a,b, 1995) and Suzumura
(1996b).
In passing, one particular type of interpersonal utility comparison deserves special
attention: "People seem prepared to make comparisons of the form: State x is better (or
worse) for me than state y is for you. ... Interpersonal comparisons of the extended
sympathy type can be put in operational form; the judgment takes the form: It is
better (in my judgment) to be myself in state x than to be you in state y [Arrow
(1963, pp. 114-115)]" 2 4 . This is indeed the type of interpersonal utility comparison

23 Note, in passing, that cardinality of individual utilities without interpersonal comparability does not
provide us with any escape route from the Arrovian impossibility theorems. Indeed, it was shown by
Sen (1970a, Theorem 8*2) that there exists no social welfare functional - which is "a mechanism that
specifies one and only one social ordering given a set of individual welfare functions, one function
for each individual [Sen (1970a, pp. 123-124)]" satisfying the following conditions: unrestricted
domain, independence of irrelevant alternatives, non-dictatorship, weak Pareto principle, cardinality, and
non-comparability.
24 The interpersonal comparisons of the extended sympathy type was first formulated with rich
applications by Patrick Suppes (1966).
Introduction 21

which formed the informational basis of, e.g. an analysis of economic inequality by
Sen (1973), as well as of an axiomatization of the Rawlsian difference principle by
Hammond (1976) and Sen (1977b). This is also the informational basis which enables
us to extend the celebrated fairness-as-no-envy approach in the theory of resource
allocation - developed most notably by Duncan Foley (1967), Serge-Christophe Kolm
(1972) and Hal Varian (1974) - to the theory of social choice, which was initiated by
Suzumura (1981a,b).
Still centering around the original Arrow impossibility theorem itself, one may try
to see how tight this remarkable theorem in fact is by carefully checking whether or
not any one of the constituting axioms can be weakened without upsetting the validity
of the theorem. One may also try to see the trade-off relationship which may hold
between different axioms, keeping the essential validity of the theorem intact. These
ideas have been pursued, e.g., by Julian Blau (1979) and Robert Wilson (1972), on
the one hand, and by Donald Campbell and Jerry Kelly (1994), on the other.
All the lines of research mentioned so far are, to a great extent, correctly describable
as being the lineal descendants of Arrow's seminal work. There are some other lines
of research which were mentioned, but not explored, in Social Choice and Individual
Values. One salient example is the strategic aspects of collective decision-making,
which we have briefly mentioned in the context of the Borda-Laplace rank-order
method of collective decision-making. Arrow (1951, p. 7) was careful enough to
point out that "once a machinery for making social choices from individual tastes
is established, individuals will find it profitable, from a rational point of view, to
misrepresent their tastes by their actions, either because such misrepresentation is
somehow directly profitable or, more usually, because some other individual will be
made so much better off by the first individual's misrepresentation that he could
compensate the first individual in such a way that both are better off than if everyone
really acted in direct accordance with his tastes". As a matter of fact, Samuelson (1954,
pp. 388-389) pointed out the ubiquity of strategic misrepresentation of preferences in
the specific context of the efficient provision of public goods: "[I]t is in the selfish
interest of each person to give false signals, to pretend to have less interest in a given
collective consumption activity than he really has, etc." This free-riderproblem, so-
called, can be traced back much further to Knut Wicksell (1896): "If the individual is
to spend his money for private and public uses so that his satisfaction is maximized,
he will obviously pay nothing whatsoever for public purposes (at least if we disregard
fees and similar charges). Whether he pays much or little will affect the scope of public
service so slightly, that for all practical purposes, he himself will not notice it at all. Of
course, if everyone were to do the same, the State would soon cease to function". In
the context of social choice theory, however, the first general treatment of the strategic
misrepresentation issue, of which Arrow was aware from the inception of social choice
theory, but left unexplored, had to wait until 1970s when Allan Gibbard (1973) and
Mark Satterthwaite (1975) came up with a general theorem on the manipulability of
22 K. SuzuLmulra

voting schemes 25 . Recollect that a voting scheme is a social choice mechanism which
assigns a single outcome to each and every profile of voters' preference orderings
over outcomes. As long as there are at least three alternative outcomes and at least
two voters, there exists no non-dictatorial voting scheme which is free from strategic
misrepresentation of preferences by individuals. It is worthwhile to point out that the
Arrow theorem is closely related to the Gibbard-Satterthwaite theorem in the sense that
the former theorem can provide the crucial step in proving the latter theorem. Given
the validity of the basic Gibbard-Satterthwaite theorem on the ubiquity of strategic
manipulation of voting schemes, it is natural that a huge literature was created in
the search for either the escape route from the Gibbard-Satterthwaite impossibility
theorem, or directions in which their theorem may be generalized.
Since the strategic misrepresentation of preferences is demonstrably ubiquitous,
there is a further problem to be tackled: "Even in a case where it is possible to
construct a procedure showing how to aggregate individual tastes into a consistent
social preference pattern, there still remains the problem of devising rules of the game
so that individuals will actually express their true tastes even when they are acting
rationally [Arrow (1951, p. 7)]". It was precisely in response to this plea that a fruitful
area of research, to be called the implementation theory, or the theory of mechanism
design, was created by Leonid Hurwicz (1960, 1972, 1973), Partha Dasgupta, Peter
Hammond and Eric Maskin (1979) and Eric Maskin (1979, 1999). A mechanism is
a game form, which is designed and managed by the helmsman of the economy, so
that it can attain the social objective at the equilibrium of the game by assigning to
each individual agent an appropriate set of admissible strategies and a payoff function.
In view of the Gibbard-Satterthwaite theorem and Hurwicz's (1972) theorem to the
same effect in economic environments, the constructed game forms are such that the
set of admissible strategies cannot be that of individual preference orderings, but that
of much wider nature. Although the public objective, which the helmsman tries to
optimize, is typically dependent on the private information, it need not be concordant
with the private incentives of individual agents. It follows that the requirement that
individual agents within the designed mechanism should be so induced as to bring
about the social objective optimization at equilibrium, cannot but impose a constraint
on the mechanisms to be designed and on the public objectives to be implemented.
Another game-theoretic background of social choice theory deserves to be men-
tioned, which can be traced back all the way to the cooperative game theory of John
von Neumann and Oscar Morgenstern (1944). Notable cooperative solution concepts
to the axiomatic bargaining problem by John Nash (1950) such as the Nash bargaining
solution, or the Kalai-Smorodinsky (1975) solution, as well as to the games of
characteristic function forms such as the Shapley value, the core, or the nucleolus,

25 See, however, an interesting earlier study on strategic behavior invoting by Robin Farquharson (1969).
See also Pattanaik (1978).
Introduction 23

provide social choice theory with a rich class of reasonable (fair) compromises in the
situation which mixes cooperation and competition among individual agents.
Not only Arrow's social choice theory, but also the Gibbard-Satterthwaite theorem
on the non-manipulability of voting schemes, as well as the Hurwicz-Maskin theory
on implementation, and the cooperative game-theoretic approach to fair compromises,
all make extensive use of axiomatic methods. Many of the strengths and weaknesses of
these theories hinge squarely on this common analytical character. As was observed by
Arrow (1951, p. 87), "[o]ne of the great advantages of abstract postulational methods
is the fact that the same system may be given several different interpretations". In
exchange for this great merit of interpretational versatility, however, the axiomatic
methods tend to be plagued with the potential weakness of a formal neglect of
substantial issues. A case in point is a warning by Leif Johansen (1977) to the effect
that the theoretically undeniable ubiquity of "playing down one's preferences for a
public good in order to get a lower share in the costs of providing the good" does
not seem "likely to succeed in an open political decision-making process involving
elected representatives." According to Johansen, "the two-tier system of electors and
representatives tends to diminish the significance and relevance of the theoretical
problem of unwillingness to reveal preferences for public goods." This warning seems
to urge us to examine in concrete detail the institutional structures of the society,
political as well as economic, in search of the empirical relevance of purely theoretical
results obtained in a general axiomatic framework. This is an interesting step to
take if one wants to verify that the paradox of voting is not just a theoretical
curiosity, but a phenomenon of substantial empirical relevance; it also motivates
us to analyse the logical performance of representative democracy vis-A-vis direct
democracy. Furthermore, instead of merging "voting, typically used to make 'political'
decisions, and the market mechanism, typically used to make 'economic' decisions
[Arrow (1951, p. 1)]" into one and the same axiomatic system, it may prove useful
to develop an idiosyncratic model of social choice in economic environments, along
with developing a separate model of political decision-making. All these steps have
been taken vigorously in the social choice literature with rich ramifications of specific
results.
There is yet another crucial point of departure from Arrow's original formulation
of social choice theory. Not only the traditional welfare economics, "old" as well as
"new", but also the Arrovian social choice theory itself, are deeply rooted in the
philosophical approach of welfarist-consequentialism in that they are based on the
assessment of the goodness of states of affairs in terms of individual utilities obtained
from these states of affairs. It was Sen's (1970a Chapter 6*, 1970b, 1976a, 1992)
impossibility ofa Paretianliberal which casted a serious doubt on this long tradition by
establishing an impossibility theorem to the effect that the weak welfaristic requirement
of the Pareto principle cannot but conflict with the non-welfaristic requirement of the
respect for minimal individual liberty. Sen's seminal analysis can be traced back to
the problem which John Stuart Mill (1859, 1861) had to face in his simultaneous
belief in the utilitarian outcome morality, on the one hand, and in the sanctity of
24 K. Szuura

libertarian rights, on the other. In view of the remarkable pervasiveness of welfarist-

consequentialism in the whole spectrum of normative economics, it is natural to
find many attempts in the literature to try to find an escape route from Sen's
impossibility theorem, e.g. Gibbard (1974), Blau (1975), Sen (1976a, Sections III-
XI), and Suzumura (1978, 1979); to gauge the robustness of Sen's liberal paradox, so-
called, e.g. David Kelsey (1985, 1988), and Sen (1976a, Section II and Appendix A2);
and to examine critically Sen's original articulation of individual liberty, e.g. Peter
Grdenfors (1981), Robert Sugden (1985), Peter Hammond (1986), Wulf Gaertner,
Prasanta Pattanaik and Kotaro Suzumura (1992), and Pattanaik and Suzumura (1994,
1996). The implications and relevance of these works on the impossibility of a Paretian
liberal are critically evaluated by Suzumura (1996a) who distinguished the three related
but distinct issues in the social choice-theoretic analysis of welfare and rights: the issue
of the analytical articulation of rights, the issue of the realization of rights, and the
issue of the initial conferment of rights. There are also many criticisms of welfarist-
consequentialism in terms of the counter-intuitive implications of this informational
constraint in some paradigmatic cases, e.g. Ronald Dworkin (1981 a), Amartya Sen and
Bernard Williams (1982), Jon Elster (1983), Amartya Sen (1985), and many others.
Once Pandora's box is opened, and we are given a glimpse of the possibilities
which lie beyond the narrow confines of welfarist-consequentialism, nothing prevents
us from asking questions which can be properly posed only when we are ready to
go beyond the traditional informational basis of welfarist-consequentialism. In the
analysis of individual well-being, for example, we need not necessarily analyse it only
through the looking glass of individual welfares. Alternative articulations of individual
advantages have been proposed, which have opened new possibilities in welfare
economics in general, and social choice theory in particular. Representative proposals
to this effect include socialprimary goods in Rawls' (1971) theory of justice, resources
in Dworkin's (1981b, 2000) theory of equality, and capabilitiesin Sen's (1985, 1999b)
theory of well-being. The new vistas thereby opened have far-reaching implications
with innovative perspectives on the theory and policy of economic development, as
expounded in Sen (1999b). We may even proceed beyond consequentialism as such,
and pose some questions such as the intrinsic value of opportunities to choose and/or
the intrinsic value of procedures for choice, along with their instrumental values.
Indeed, it is only with these new developments in clear perspective that we can gauge
the true usefulness and limitations of the traditional informational basis of welfarist-
consequentialism. Some of these new vistas opened in this direction are expounded
in Sen (2001), Suzumura (1999a, 2000, 2001), Kotaro Suzumura and Yongsheng Xu
(2001a,b) and Reiko Gotoh and Kotaro Suzumura (2001).
Overlapping partly with this trend to go beyond welfarist-consequentialism as the
informational basis of social welfare analysis, there were conspicuous developments
in the theory of how to measure economic well-being. It was Serge-Christophe Kolm
(1969) and Anthony Atkinson (1970) who kicked off the modern resurgence of interest
in the measurement of income inequality. Soon afterwards, Sen (1976b) axiomatized a
new measure of income poverty, which went substantially beyond the crude traditional
Introduction 25

measure such as the head count ratio, and incorporated a new distributional dimension
into the measurement of poverty. More recently, Prasanta Pattanaik and Yongsheng Xu
(1990) started a new area of research concerning how to measure freedom of choice.
Each one of these seminal works generated substantial follow-up works of their own,
which are enriching our theoretical tool box for the measurement of well-being.
This Handbook of Social Choice and Welfare is a systematic attempt to provide, in
two volumes, an up-to-date overview of the current state of the art in social choice
theory and welfare economics, encompassing all these issues we have so far identified
and even more 26. Plenty of dishes are on the table. It is our sincere hope that the readers
will enjoy them and be motivated to participate in the vigorous research activities
which are currently taking place.

5. A disclaimer

It has been said that social choice theory is "a science of the impossible". This
statement contains an element of the truth only to the limited extent that the
development of modern social choice theory received strong momentum from
many impossibility theorems. Arrow's monumental theorem on the impossibility of
democratic and informationally efficient preference aggregation procedures, Sen's
theorem on the impossibility of a Paretian liberal, and the Gibbard-Satterthwaite
theorem on the impossibility of non-manipulable and non-dictatorial voting schemes,
to cite only a few most salient examples, have served us positively by sending an
unambiguous signal that there are logical problems which await our careful scrutiny
and serious attempt for resolution. In the process of understanding these impossibility
theorems, we are brought to the far deeper perception of what underlies social conflicts
of important values than ever. Likewise, in the process of finding some meaningful
escape routes from these logical impasses, we are brought to much richer understanding
on what makes several social values mutually compatible than otherwise. In this
sense, there is nothing intrinsically negative about social choice theory in general, and
impossibility theorems in particular.
It has also been said that welfare economics is plagued with elegance nihilism. In
this context, it is worthwhile to recollect that Pigou's "old" welfare economics started
with the following manifest: "The complicated analyses which economists endeavour
to carry through are not mere gymnastic. They are instruments for the bettering of
human life. The misery and squalor that surround us, the dying fire of hope in many
millions of European homes, the injurious luxury of some wealthy families, the terrible
uncertainty overshadowing many families of the poor - these are evils too plain to
be ignored. By the knowledge that our science seeks it is possible that they may be
restrained [Pigou (1920, p. vii)]". Forty years later, however, Edward Mishan (1960,

26 See also Arrow, Sen and Suzumura (1996-1997).

26 K. Suzumura

p. 197) commenced his survey of welfare economics over the period 1939-1959 with
the following remark: "While it continues to fascinate many, welfare economics does
not appear at any time to have wholly engaged the labours of any one economist. It is
a subject which, apparently, one dabbles in for a while, leaves and, perhaps, returns to
later in response to troubled conscience ... " Since Mishan's survey covered the period
over which the "new" welfare economics was created so as to replace the crumbling
"old" welfare economics only to receive harsh criticisms on their logical foundations
even before the scaffolds of construction were removed from their construction sites,
Mishan's cynicism may be understandable at least to some extent. But the cynicism
persisted ever since, and Atkinson (2001) felt it necessary to talk about "The Strange
Disappearance of Welfare Economics" from the mainstream economics. However, as
we have observed at the beginning of this Introduction, "as soon as any collective body
designs and implements an economic mechanism and/or an economic policy, paying
proper attention to the costs and benefits accruing to its constituent members, one or
more social welfare judgements cannot be avoided." Since social choice theory is partly
concerned with the logical foundations of welfare economics, we cannot but maintain
that the study of social choice theory and welfare economics is indispensable as long
as one is interested in the problem of any economic policy, be that macroeconomic
or microeconomic in nature. Pigou thought that welfare economics was a potent
instrument for the bettering of human life. The same can be said of social choice
theory.

References

Arrow, K.J. (1950), "A difficulty in the concept of social welfare", Journal of Political Economy
58:328-346. Reprinted with an introduction in 1984 in: Collected Papers of Kenneth J. Arrow, Vol. 1
(Basil Blackwell, Oxford) pp. 1-29.
Arrow, K.J. (1951), Social Choice and Individual Values, 1st Edition (Wiley, New York).
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York); with "Notes
on the Theory of Social Choice, 1963".
Arrow, K.J. (1983), "Contributions to welfare economics", in: E.C. Brown and R.M. Solow, eds., Paul
Samuelson and Modern Economic Theory (McGraw-Hill, New York) pp. 15 30.
Arrow, K.J. (1984), Social Choice and Justice, Vol. 1, Collected Papers of Kenneth J. Arrow (Basil
Blackwell, Oxford).
Arrow, K.J. (1987), "Arrow's Theorem", in: J. Eatwell, M. Milgate and P. Newman, eds., The New
Palgrave: A Dictionary of Economics, Vol. 1 (Macmillan, London) pp. 124-126.
Arrow, K.J., A.K. Sen and K. Suzumura, eds. (1996-1997), Social Choice Re-examined (Macmillan,
London) Vol. 1, 1997; Vol. 2, 1996.
Atkinson, A.B. (1970), "On the measurement of inequality", Journal of Economic Theory 2:244-263.
Atkinson, A.B. (2001), "The strange disappearance of welfare economics", Kyklos 54:193-206.
Barone, E. (1908), "II ministro della produzione nello stato collectivista", Gionale degli Economisti e
Rivista di Statistica 37:267-293.
Barone, E. (1935), "The Ministry of Production in the collectivist state", in: A. von Hayek, ed.,
Collectivist Economic Planning (Routledge, London) pp. 245-290.
Bentham, J. (1789), An Introduction to the Principles of Morals and Legislation (Payne, London).
Republished in 1907 by Clarendon Press, Oxford.
Introduction 27

Bentham, J. (1843), "Anarchical fallacies". First published in English in: J. Bowring, ed., The Works of
Jeremy Bentham, Vol. II (William Tait, Edinburgh). Republished in 1995 (Theommes Press, Bristol)
pp. 489-534.
Bergson, A. (1938), "A reformulation of certain aspects of welfare economics", Quarterly Journal of
Economics 52:310-334.
Bergson, A. (1967), "Market socialism revisited", Journal of Political Economy 75:655 673.
Black, D. (1948), "On the rationale of group decision-making", Journal of Political Economy 56:23-34.
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, Cambridge).
Blair, D.H., G.A. Bordes, J.S. Kelly and K. Suzumura (1976), "Impossibility theorems without collective
rationality", Journal of Economic Theory 21:361-379.
Blau, J.H. (1975), "Liberal values and independence", Review of Economic Studies 42:395-401.
Blau, J.H. (1979), "Semiorders and collective choice", Journal of Economic Theory 21:195-206.
Borda (J.-C. de Borda) (1781), "Memoire sur les elections par scrutin", Mrmoires de l'Academie Royale
des Sciences annre 1781, pp. 657-665. Translated in English in 1953 by A. de Grazia: "Mathematical
derivation of an election system", Isis 44:42-51.
Buchanan, J.M. (1954), "Social choice, democracy, and free markets", Journal of Political Economy
62:114-123.
Campbell, D.E., and J.S. Kelly (1994), "Trade-off theory", American Economic Review: Papers and
Proceedings 84:422-426.
Chipman, J.S. (1976), "The Paretian heritage", Revue Europeene des Sciences Sociales et Cahiers
Vilfredo Pareto 14:65-171.
Chipman, J.S., and J.C. Moore (1978), "The new welfare economics 1939-1974", International Economic
Review 19:547-584.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Dalton, H. (1920), "The measurement of the inequality of incomes", Economic Journal 30:348-361.
Dasgupta, P, PJ. Hammond and E. Maskin (1979), "Implementation of social choice rules: some general
results on incentive compatibility", Review of Economic Studies 46:181-216.
d'Aspremont, C., and L. Gevers (1977), "Equity and informational basis of collective choice", Review
of Economic Studies 44:199-209.
Dobb, M. (1969), Welfare Economics and the Economics of Socialism (Cambridge University Press,
Cambridge).
Dodgson, C.L. (Lewis Carroll) (1873), A Discussion of the Various Methods of Procedure in Conducting
Elections (Imprint by E.B. Gardner, E. Pickard Hall and J.H. Stacy, Printers to the University, Oxford).
Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections (Cambridge University
Press, Cambridge) pp. 214-222.
Dodgson, C.L. (Lewis Carroll) (1874), Suggestions as to the Best Method of Taking Votes, Where
More than Two Issues are to be Voted on (Imprint by E. Pickard Hall and J.H. Stacy, Printers to
the University, Oxford). Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections
(Cambridge University Press, Cambridge) pp. 222-224.
Dodgson, C.L. (Lewis Carroll) (1876), A Method of Taking Votes on More Than Two Issues (Clarendon
Press, Oxford). Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections (Cambridge
University Press, Cambridge) pp. 224-234.
Dworkin, R. (1981a), "What is equality? Part 1: Equality of welfare", Philosophy and Public Affairs
10:185-246.
Dworkin, R. (1981b), "What is equality? Part 2: Equality of resources", Philosophy and Public Affairs
10:283-345.
Dworkin, R. (2000), Sovereign Virtue: The Theory and Practice of Equality (Harvard University Press,
MA).
28 K. Suzuntlra

Elster, J. (1983), Sour Grapes: Studies in the Subversion of Rationality (Cambridge University Press,
Cambridge).
Farquharson, R. (1969), Theory of Voting (Yale University Press, New
Haven).
Foley, D.K. (1967), "Resource allocation and the public sector", Yale
Economic Essays 7:45-98.
Gaertner, W., P.K. Pattanaik and K. Suzumura (1992), "Individual rights
revisited", Economica 59:
161-177.
Galton, E (1907), "One vote, one value", Nature 75:414.
Girdenfors, P (1981), "Rights, games and social choice", Nois
15:341-356.
Gibbard, A.E (1973), "Manipulation of voting schemes: a general result",
Econometrica 41:587-601.
Gibbard, A.F. (1974), "A Pareto-consistent libertarian claim", Journal
of Economic Theory 7:388-410.
Gorman, WM. (1955), "The intransitivity of certain criteria used in welfare
economics", Oxford Economic
Papers 7:25-35.
Gotoh, R., and K. Suzumura (2001), "Constitutional democracy and public
judgements", paper presented
at the Conference in Honour of Amartya K. Sen, Center for Interdisciplinary
Research, University of
Bielefeld, June 21-23, 2001.
Graaff, J. de V (1957), Theoretical Welfare Economics (Cambridge University
Press, London).
Hammond, P.J. (1976), "Equity, Arrow's conditions and Rawls' difference
principle", Econometrica
44:793-804.
Hammond, PJ. (1986), "Consequentialist social norms for public decisions",
in: WP. Heller, R.M. Starr
and D.A. Starrett, eds., Social Choice and Public Decision Making:
Essays in Honor of Kenneth J.
Arrow, Vol. I (Cambridge University Press, Cambridge) pp. 3-27.
Harsanyi, J.C. (1953), "Cardinal utility in welfare economics and in
the theory of risk-taking", Journal
of Political Economy 61:434-435.
Harsanyi, J.C. (1955), "Cardinal welfare, individualistic ethics and interpersonal
comparisons of utility",
Journal of Political Economy 63:309-321.
Harsanyi, J.C. (1977), Rational Behavior and Bargaining Equilibrium
in Games and Social Situations
(Cambridge University Press, Cambridge).
Heal, G.M. (1973), The Theory of Economic Planning (North-Holland,
Amsterdam).
Hicks, J.R. (1939), "The foundations of welfare economics",
Economic Journal 49:696-712.
Hicks, JR. (1940), "The evaluation of the social income",
Economica 7:105-124,
Hicks, J.R. (1975), "The scope and status of welfare economics", Oxford
Economic Papers 27:307-326.
Hurwicz, L. (1960), "Optimality and informational efficiency in resource
allocation processes", in:
K.J. Arrow, S. Karlin and P. Suppes, eds., Mathematical Methods
in the Social Sciences 1959
(Stanford University Press, Stanford) pp. 27-46.
Hurwicz, L. (1972), "On informationally decentralized systems", in: C.B.
McGuire and R. Radner, eds.,
Decision and Organization (North-Holland, Amsterdam) pp. 297-336.
Hurwicz, L. (1973), "The design of resource allocation mechanisms",
American Economic Review
58:1-30.
Inada, K. (1969), "The simple majority decision rule", Econometrica 37:490-506.
Johansen, L. (1969), "An examination of the relevance of Kenneth Arrow's
general possibility theorem for
economic planning", Economics of Planning 9:5-41. Reprinted, 1987,
in: FR. Frsund, ed., Collected
Works of Leif Johansen, Vol. 2 (North-Holland, Amsterdam) pp. 439-475.
Johansen, L. (1977), "The theory of public goods: misplaced emphasis?"
Journal of Public Economics
7:147-152. Reprinted, 1987, in: ER. Frsund, ed., Collected Works
of Leif Johansen, Vol. 2 (North-
Holland, Amsterdam) pp. 663-668.
Kalai, E., and M. Smorodinsky (1975), "Other solutions to Nash's
bargaining problem", Econometrica
43:513-518.
Kaldor, N. (1939), "Welfare propositions in economics and interpersonal
comparisons of utility",
Economic Journal 49:549-552.
Kelsey, D. (1985), "The liberal paradox: a generalization", Social Choice
and Welfare 1:245 252.
Introduction 29

Kelsey, D. (1988), "What is responsible for the 'Paretian Epidemic'?" Social Choice and Welfare
5:303-306.
Kolm, S.-Ch. (1969), "Optimum production of social justice", in: J. Margolis and H. Guitton, eds., Public
Economics (Macmillan, New York) pp. 145-200.
Kolm, S.-Ch. (1972), Justice et Equit6 (Editions du Centre National de la Recherche Scientifique, Paris).
1997 English translation: Justice and Equity (MIT Press, Cambridge, MA).
Lange, 0. (1936-1937), "On the economic theory of socialism", Review of Economic Studies 4:53-71,
123-142. Revised and reprinted in 1938.
Lange, 0. (1938), "On the economic theory of socialism", in: B.E. Lippincott, ed., with O. Lange and
EM. Taylor, On the Economic Theory of Socialism (The University of Minnesota Press, Minneapolis)
pp. 57-143 (revised).
Lange, 0. (1967), "The computer and the market", in: C.H. Feinstein, ed., Socialism, Capitalism and
Economic Growth (Cambridge University Press, London) pp. 158-161.
Laplace, P.-S. (1812), "Lecons de math6matiques, donn6es a l'Ecole Normale en 1795", Journal de
l'Ecole Polytechnique, Tome II, Septieme et Huiti6me cahiers, Paris.
Lerner, A.P. (1944), The Economics of Control (Macmillan, New York).
Little, I.M.D. (1952), "Social choice and individual values", Journal of Political Economy 60:422-432.
Malinvaud, E. (1967), "Decentralized procedures for planning", in: E. Malinvaud and M.O.L. Bacharach,
eds., Activity Analysis in the Theory of Growth and Planning (Macmillan, London) pp. 170-208.
Mas-Colell, A., and H. Sonnenschein (1972), "General possibility theorems for group decisions", Review
of Economic Studies 39:185-192.
Maskin, E. (1978), "A theorem on utilitarianism", Review of Economic Studies 45:93-96.
Maskin, E. (1979), "Implementation and strong Nash equilibrium", in: J.-J. Laffont, ed., Aggregation
and Revelation of Preferences (North-Holland, Amsterdam) pp. 433-439.
Maskin, E. (1999), "Nash equilibrium and welfare optimality", Review of Economic Studies 66:23-38.
McLean, I. (1995), "The first golden age of social choice, 1784-1803", in: W.A. Barnett, H. Moulin,
M. Salles and N.J. Schofield, eds., Social Choice, Welfare, and Ethics (Cambridge University Press,
Cambridge) pp. 13-33.
McLean, I., and E Hewitt (1994), Condorcet: Foundations of Social Choice and Political Theory (Edward
Elgar, Hants).
McLean, I., and J. London (1990), "The Borda and Condorcet principles: three medieval applications",
Social Choice and Welfare 7:99-108.
Mill, J.S. (1859), On Liberty (Parker, London). Reprinted, 1977, in: J.M. Robson, ed., The Collected
Works of John Stuart Mill, Vol. XVIII (The University of Toronto Press, Toronto).
Mill, J.S. (1861), Utilitarianism (Collins, London).
Mishan, E.J. (1960), "A survey of welfare economics, 1939-59", Economic Journal 70:197-265.
Nanson, E.J. (1882), "Methods of election", Transactions and Proceedings of the Royal Society of
Victoria 19:197-240. Reprinted, 1907, in British Government blue book, Misc. No. 3, Cd. 3501.
Nash, J.E (1950), "The bargaining problem", Econometrica 18:155-162.
Negishi, T. (1962), "The stability of a competitive economy: a survey article", Econometrica 30:
635-669.
Pareto, V (1906), Manuale di Economia Politica (Societa Editrice Libraria, Milan). French translation
(revised), 1909, Manuel d'Economie Politique (M. Giard, Paris). English translation, 1927, Manual
of Political Economy (A.M. Kelley, New York).
Pareto, V (1913), "II massimo di utilitA per una collettiviti in Sociologia", Giornale Degli Economisti
e Revista di Statistica 46:337-341.
Pattanaik, P.K. (1968), "Risk, impersonality, and the social welfare function", Journal of Political Economy
76:1152-1169.
Pattanaik, PK. (1978), Strategy and Group Choice (North-Holland, Amsterdam).
Pattanaik, PK., and K. Suzumura (1994), "Rights, welfarism and social choice", American Economic
Review: Papers and Proceedings 84:435-439.
30 K. Suzumura

Pattanaik, P.K., and K. Suzumura (1996), "Individual rights and social evaluation: a conceptual
framework", Oxford Economic Papers 48:194-212.
Pattanaik, PK., and Y. Xu (1990), "On ranking opportunity sets in terms of freedom of choice",
Recherches Economiques de Louvain 56:383-390.
Pigou, A.C. (1920), The Economics of Welfare (Macmillan, London).
Plott, C.R. (1973), "Path independence, rationality, and social choice", Econometrica 41:1075-1091.
Rawls, J. (1962), "Justice as fairness", in: P. Laslett and W.G. Runciman, eds., Philosophy, Politics and
Society, 2nd series (Basil Blackwell, Oxford) pp. 132-157. Reprinted, 1999, in: S. Freeman, ed., John
Rawls: Collected Papers (Harvard University Press, MA) pp. 47-72.
Rawls, J. (1963), "Constitutional liberty and the concept of justice", in: C.J. Friedrich and J.W. Chapman,
eds., Nomos VI: Justice (Atherton Press, New York) pp. 98 125. Reprinted, 1999, in: S. Freeman,
ed., John Rawls: Collected Papers (Harvard University Press, MA) pp. 73-95.
Rawls, J. (1971), A Theory of Justice (Harvard University Press, MA). Revised edition, 1999.
Robbins, L. (1935), An Essay on the Nature and Significance of Economic Science, 2nd Edition
(Macmillan, London). Ist Edition published in 1932.
Robbins, L. (1938), "Interpersonal comparisons of utility", Economic Journal 48:635-641.
Robbins, L. (1981), "Economics and political economy", American Economic Review 71:1-10.
Roberts, K.WS. (1980a), "Possibility theorems with interpersonally comparable utility levels", Review
of Economic Studies 47:409-420.
Roberts, K.WS. (1980b), "Interpersonal comparability and social choice theory", Review of Economic
Studies 47:421-439.
Roberts, K.WS. (1995), "Valued opinions or opinionized values: the double aggregation problem", in:
K. Basu, P.K. Pattanaik and K. Suzumura, eds., Choice, Welfare and Development: A Festschrift in
Honour of Amartya Sen (Clarendon Press, Oxford) pp. 141-185.
Rothschild, E. (2001), Economic Sentiments: Adam Smith, Condorcet, and the Enlightenment (Harvard
University Press, MA).
Samuelson, P.A. (1947), Foundations of Economic Analysis (Harvard University Press, MA).
Samuelson, P.A. (1950), "Evaluation of real national income", Oxford Economic Papers 2:1-29.
Samuelson, PA. (1954), "The pure theory of public expenditure", Review of Economics and Statistics
36:387-389. Reprinted, 1966, in: J.E. Stiglitz, ed., Collected Scientific Papers of Paul A. Samuelson,
Vol. II (The MIT Press, MA) pp. 1223-1225.
Samuelson, P.A. (1967), "Arrow's mathematical politics", in: Hook, S., ed., Human Values and Economic
Policy (New York University Press, New York) pp. 41-52.
Samuelson, PA. (1981), "Bergsonian welfare economics", in: S. Rosefielde, ed., Economic Welfare and
the Economics of Soviet Socialism: Essays in Honor of Abram Bergson (Cambridge University Press,
MA) pp. 223-266.
Satterthwaite, M.A. (1975), "Strategyproofness and Arrow's conditions: existence and correspondence
theorems for voting procedures and social welfare functions", Journal of Economic Theory 10:
187-217.
Scarf, H. (1960), "Some examples of global instability of the competitive equilibrium", International
Economic Review 1:157-172.
Schumpeter, J.A. (1954), History of Economic Analysis (Oxford University Press, New York).
Scitovsky, T. (1941), "A note on welfare propositions in economics", Review of Economic Studies
9:77-88.
Sen, A.K. (1969), "Quasi-transitivity, rational choice and collective decisions", Review of Economic
Studies 36:381-393.
Sen, A.K. (1970a), Collective Choice and Social Welfare (Holden-Day, San Francisco). Republished,
1979, North-Holland, Amsterdam.
Sen, A.K. (1970b), "The impossibility of a Paretian liberal", Journal of Political Economy 78:152-157.
Sen, A.K. (1973), On Economic Inequality, st Edition (Oxford University Press, Oxford). 1997 Expanded
edition with a substantial annexe by J.E. Foster and A.K. Sen (Oxford University Press, Oxford).
Introduction 31

Sen, A.K. (1976a), "Liberty, unanimity and rights", Economica 43:217-245.

Sen, A.K. (1976b), "Poverty: an ordinal approach to measurement", Econometrica 44:219-223.
Sen, A.K. (1976c), "Welfare inequalities and Rawlsian axiomatics", Theory and Decisions 7:243-262.
Sen, A.K. (1977a), "Social choice theory: a re-examination", Econometrica 45:53-89.
Sen, A.K. (1977b), "On weights and measures: informational constraints in social welfare analysis",
Econometrica 45:1539-1572.
Sen, A.K. (1979), "Utilitarianism and welfarism", Journal of Philosophy 76:463-489.
Sen, A.K. (1985), Commodities and Capabilities (North-Holland, Amsterdam).
Sen, A.K. (1992), "Minimal liberty", Economica 59:139-159.
Sen, A.K. (1999a), "The possibility of social choice", American Economic Review 89:349-378.
Sen, A.K. (1999b), Development as Freedom (Oxford University Press, Oxford).
Sen, A.K. (2001), Rationality, Freedom and Justice, Vol. I: Rationality and Freedom: Arrow Lectures
and Other Essays; Vol. II: Freedom and Justice: Dewey Lectures and Other Essays. To be published.
Sen, A.K., and PK. Pattanaik (1969), "Necessary and sufficient conditions for rational choice under
majority decisions", Journal of Economic Theory 1:178-202.
Sen, A.K., and B. Williams, eds. (1982), Utilitarianism and Beyond (Cambridge University Press,
Cambridge).
Sugden, R. (1985), "Liberty, preference, and choice", Economics and Philosophy 1:213-229.
Suppes, P. (1966), "Some formal models of grading principles", Synthese 6:284-306.
Suzumura, K. (1976), "Remarks on the theory of collective choice", Economica 43:381-390.
Suzumura, K. (1978), "On the consistency of libertarian claims", Review of Economic Studies 45:
329-342.
Suzumura, K. (1979), "A correction", Review of Economic Studies 46:743.
Suzumura, K. (1980), "On distributional value judgements and piecemeal welfare criteria", Economica
47:125-139.
Suzumura, K. (1981a), "On the possibility of 'fair' collective choice rule", International Economic
Review 22:307-320.
Suzumura, K. (1981b), "On Pareto-efficiency and the no-envy concept of equity", Journal of Economic
Theory 25:367-379.
Suzumura, K. (1983), Rational Choice, Collective Decisions and Social Welfare (Cambridge University
Press, New York).
Suzumura, K. (1987), "Social welfare function", in: J. Eatwell, M. Milgate and P Newman, eds., The
New Palgrave: A Dictionary of Economics, Vol. 4 (Macmillan, London) pp. 418420.
Suzumura, K. (1996a), "Welfare, rights, and social choice procedure: a perspective", Analyse & Kritik
18:20-37.
Suzumura, K. (1996b), "Interpersonal comparisons of the extended sympathy type and the possibility of
social choice", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Social Choice Re-examined, Vol. 2
(Macmillan, London) pp. 202-229.
Suzumura, K. (1999a), "Consequences, opportunities, and procedures", Social Choice and Welfare
16:17-40.
Suzumura, K. (1999b), "Paretian welfare judgements and Bergsonian social choice", Economic Journal
109:204-220.
Suzumura, K. (2000), "Welfare economics beyond welfarist-consequentialism", Japanese Economic
Review 51:1-32.
Suzumura, K. (2001), "On the concept of procedural fairness", Presidential Address presented at the
Society for Social Choice and Welfare, Alicante, Spain, July 2000.
Suzumura, K., and Y. Xu (2001a), "Characterizations of consequentialism and non-consequentialism",
Journal of Economic Theory 101:423-436.
Suzumura, K., and Y. Xu (2001b), "Welfarist-consequentialism, similarity of attitudes, and Arrow's
general impossibility theorem", Social Choice and Welfare, forthcoming.
Sweezy, P.M. (1949), Socialism (McGraw-Hill, New York).
32 K. Suzumura

Taylor, EM. (1929), "The guidance of production in a socialist state", American Economic Review
19:1-8. Reprinted, 1938, in: B.E. Lippincott, ed., Oscar Lange and Fred M. Taylor, On the Economic
Theory of Socialism (The University of Minnesota Press, Minneapolis) pp. 41 54.
Todhunter, I. (1865), A History of the Mathematical Theory of Probability from the Time of Pascal to
That of Laplace (Macmillan, London).
Varian, H.R. (1974), "Equity, envy, and efficiency", Journal of Economic Theory 9:63-91.
Vickrey, W.S. (1945), "Measuring marginal utility by reactions to risk", Econometrica 13:319-333.
Vickrey, W.S. (1960), "Utility, strategy, and social decision rules", Quarterly Journal of Economics
74:507-535.
von Hayek, EA., ed. (1935), Collectivist Economic Planning (Routledge, London).
von Hayek, EA. (1944), The Road to Serfdom (Routledge and Kegan Paul, London).
von Hayek, EA. (1948), Individualism and Economic Order (The University of Chicago Press, Chicago).
von Mises, L. (1920), "Die Wirtschaftsrechnung im sozialistischen Gemeinwesen", Archiv fr
Sozialwissenschaften und Sozialpolitik 47. English translation, 1935, in: EA. von Hayek, ed.,
Collectivist Economic Planning (Routledge, London) pp. 87-130.
von Neumann, J., and 0. Morgenstern (1944), Theory of Games and Economnic Behavior, 1st Edition
(Princeton University Press, Princeton) 2nd Edition, 1947.
Wicksell, K. (1896), "Ein neues Prinzip der gerechten Besteuerung", Finanztheoretische Untersuchungen,
Jena. English translation, 1958, "A new principle of just taxation", in: R.A. Musgrave and A.T. Peacock,
eds., Classics in the Theory of Public Finance (Macmillan, London) pp. 72-118.
Wilson, R.B. (1972), "Social choice without the pareto principle", Journal of Economic Theory 5:14-20.
CONTENTS OF THE HANDBOOK

VOLUME 1
Introduction
KOTARO SUZUMURA
Part 1 - ARROVIAN IMPOSSIBILITY THEOREMS
Chapter I
Impossibility Theorems in the Arrovian Framework
DONALD E. CAMPBELL and JERRY S. KELLY
Chapter 2
Categories of Arrovian Voting Schemes
FUAD ALESKEROV
Chapter 3
Domain Restrictions
WULF GAERTNER
Part 2 - VOTING SCHEMES AND MECHANISMS
Chapter 4
Voting Procedures
STEVEN J. BRAMS and PETER C. FISHBURN
Chapter 5
Implementation Theory
ERIC MASKIN and TOMAS SJOSTROM
Chapter 6
Axiomatic Cost and Surplus Sharing
HERVE MOULIN
Part 3 - STRUCTURE OF SOCIAL CHOICE RULES
Chapter 7
Positional Rules of Collective Decision-Making
PRASANTA K. PATTANAIK
Chapter 8
Game-Theoretic Analysis of Voting in Committees
BEZALEL PELEG
viii Contents of'the Handbook

Chapter 9
Representative Democracy and Social Choice Theory
NORMAN SCHOFIELD

Part 4 - WELFARE, JUSTICE AND POVERTY

Chapter 10
Social Welfare Functionals and Interpersonal Comparability
CLAUDE d'ASPREMONT and LOUIS GEVERS

Chapter 11
Utilitarianism and the Theory of Justice
CHARLES BLACKORBY, WALTER BOSSERT and DAVID DONALDSON

Chapter 12
Inequality, Poverty and Welfare
BHASKAR DUTTA
PREFACE TO THE HANDBOOK

This Handbook of Social Choice and Welfare presents, in two volumes, essays on
past and on-going work in social choice theory and welfare economics. The first
volume consists of four parts. In Part 1 (Arrovian Impossibility Theorems), various
aspects of Arrovian general impossibility theorems, illustrated by the simple majority
cycle first identified by Condorcet, are expounded and evaluated. It also provides a
critical survey of the work on different escape routes from impossibility results of this
kind. In Part 2 (Voting Schemes and Mechanisms), the operation and performance of
voting schemes and cost-sharing mechanisms are examined axiomatically, and some
aspects of the modem theory of incentives and mechanism design are expounded and
surveyed. In Part 3 (Structure of Social Choice Rules), the positional rules of collective
decision-making (the origin of which can be traced back to a seminal proposal by
Borda), the game-theoretic aspects of voting in committees, and the implications of
representative democracy in the theoretical arena of social choice theory are examined.
In Part 4 (Welfare, Justice and Poverty), the possibility and implications of making
use of interpersonal comparisons of welfare (with or without cardinal measurability)
are expounded, and the status of utilitarianism as a theory of justice is critically
examined. It also provides an analytical survey of the foundations of measurement of
inequality and poverty. In order to place these broad issues (as well as further issues
to be discussed in the second volume of the Handbook) in perspective, an extensive
Introduction, written by one of us (Kotaro Suzumura) is presented, discussing the
historical background of social choice theory, the vistas opened by Arrow's Social
Choice and Individual Values, the famous "socialist planning" controversy, and the
theoretical and practical significance of social choice theory.
The primary purpose of this Handbook is to provide an accessible introduction to
the current state of the art in social choice theory and welfare economics. But we also
believe that the expounded theory has a strong and constructive message for pursuing
human well-being and facilitating collective decision-making.

KENNETH ARROW
AMARTYA SEN
KOTARO SUZUMURA
Chapter I

IMPOSSIBILITY THEOREMS IN THE

ARROVIAN FRAMEWORK
DONALD E. CAMPBELL*
The College of William and Mary

JERRY S. KELLY'
Syracuse University

Contents

Abstract 36
Keywords 36
1. Introduction 37
2. Definitions and framework 39
3. Fundamental lemmas and Arrow's theorem 43
4. Relaxing the Pareto criterion 52
5. Relaxing transitivity 57
6. Relaxing the domain condition 64
7. Relaxing independence of irrelevant alternatives 70
8. Modifications of the Arrovian framework 75
9. Concluding remarks 84
References 85

* We thank John Weymark and the Pacific Institute of Mathematics for the opportunity to present an
early draft of this paper at their 1998 workshop at the University of British Columbia. And we are
extremely grateful to the editors of this Handbook for comments on a previous draft.

Handbook of Social Choice and Welfare, Volune 1, Edited by KS Arrow, A.K. Sen and K. Suzumura
©3 2002 Elsevier Science B. V All rights reserved
36 D.E. Campbell and J.S. Kelly

Abstract

Given a set of outcomes that affect the welfare of the members of a group, K.J. Arrow
imposed the following five conditions on the ordering of the outcomes as a function of
the preferences of the individual group members, and then proved that the conditions
are logically inconsistent:
* The social choice rule is defined for a large family of assignments of transitive
orderings to individuals.
* The social ordering itself is always transitive.
* The social choice rule is not dictatorial. (An individual is a dictator if the social
ordering ranks an outcome x strictly above another outcome y whenever that
individual strictly prefers x to y.)
* If everyone in the group strictly prefers outcome x to outcome y, then x should rank
strictly above y in the social ordering.
* The social ordering of any two outcomes depends only on the way that the individuals
in the group order those same two outcomes.
The chapter proves Arrow's theorem and investigates the possibility of uncovering a
satisfactory social choice rule by relaxing the conditions while remaining within the
Arrovian framework, which is identified by the following five characteristics:
· The outcome set is unstructured.
· The society is finite and fixed.
* Only information about the ordering of the outcome set is used to convey information
about individual welfare.
* The output of the social choice process is an ordering of the outcome set.
* Strategic play by individuals is not considered.

Keywords

Arrow, impossibility, information, Pareto, trade-off

JEL classification: D6, D7

Ch. 1: Impossibility Theorems in the Arrovian Framework 37

1. Introduction

Arrow's Theorem [Arrow (1950, 1951, 1963)] on the aggregation of individual

preferences is so startling, and robust, and significant that it spawned a new branch
of social studies, called social choice theory. Given a set of outcomes that affect the
welfare of the members of a group, K.J. Arrow proposed a handful of simple conditions
on the ordering of the outcomes as a function of the preferences of the individual group
members, and then proved that the conditions are logically inconsistent. His theorem is
valid whether the group is large or small - the citizens of a country, or a two-member
committee, at the extremes. In isolation, each criterion seems appropriate, even mild,
in many contexts. (For example, one requires that if everyone in the group prefers
outcome x to outcome y then x should be above y in the social ranking.) However,
Arrow proved that there exists no social (or group) choice procedure that satisfies all
of them. This chapter examines that theorem, and related results. Sen (1986) provides
a thorough review of the first thirty-five years of social choice research, and the other
chapters in this Handbook touch on many of the issues raised by Arrow (1963). Our
aim here is to provide insight into key results within the "Arrovian Framework" which
is identified by the following five characteristics:
* The outcome set is unstructured.
* The society is finite and fixed.
* Only information about an individual's ordering of the outcome set is used to convey
information about that individual's welfare.
* The output of the social choice process is an ordering of the outcome set.
* Strategic play by individuals is not considered.
The remainder of the introduction explains these five properties in more detail.
First, the set X of alternatives is unstructured, to give the results an extremely wide
set of applications. Other important strands of the social choice literature impose
a topological or algebraic structure on X - see Le Breton and Weymark (2002a),
who treat economic environments, and Baigent (2002) on topological social choice -
Chapters 16 and 17, respectively, in Volume 2 of this Handbook.
Second, there is a finite set N of "individuals", which is fixed in this chapter.
Typically, the members of N are different people, who have preferences over the
alternatives in X. Smith (1973) and Young (1974) have some interesting results for
variable electorates, and Blackorby, Bossert and Donaldson (1995) treat social choice
problems for which population size, among other things, varies across states. Fishburn
(1970) showed that Arrow's theorem doesn't hold with infinitely many individuals,
although Kirman and Sondermann (1972) prove something close to Arrow's theorem
for infinite N. There are other interpretations of N. For example, X is a set of
restaurants and N={1,2, 3}, denoting menu, location, and ambience, respectively.
Preference scheme 1 orders the restaurants according to the menus, preference
scheme 2 orders them by location, and the third lists the members of X by ambience.
For each attribute, the ranking reflects the tastes of a single individual. [See May
(1954) and Arrow and Raynaud (1986) on multicriteria decision making]. Dutta and
38 D.E. Campbell and .S. Kelly

Sen (1996) treat the members of N as labels for the different criteria by which one
evaluates opportunity sets. For instance, criterion 1 ranks opportunities in terms of the
highest level of attainable utility, and criterion 2 bases the ranking on the size of the
opportunity set. One then asks if there is a reasonable way of aggregating the different
orderings into a summary ranking of opportunity sets.
Third, social choice is sensitive only to the ordinal properties of individual
preference. Specifically, the informational base for a social choice procedure is a
family of profiles, where a profile is an assignment of a preference over X to each
individual in N. The input to the social choice procedure is a profile. Consideration of
cardinal utility or interpersonal comparisons is omitted here and the reader is referred
to d'Aspremont and Gevers (2002, Chapter 10 in this Volume). More generally,
intensity of preferences is not taken into consideration. Also omitted is the use of
non-preference information like rights. Sen (1970b) introduced this topic into modern
social choice theory with a very striking impossibility theorem of his own. On this see
Suzumura (2002, Chapter 19 in Volume 2 of this Handbook). Sen (1977) pointed out
that when social choice is sensitive only to the ordinal part of individual preference
we are employing a specific invariance condition, drawn from a very wide family of
invariance conditions: specifically, the output of the social choice process is invariant
to transformations of the inputs that do not affect the ordinal part of individual
preferences. See Sen (1999, 2002, Chapter 14 in Volume 2 of this Handbook) for strong
skepticism about this assumption. Bossert (2000) presents a number of theorems on the
algebra of generalized invariance conditions. Roberts (1997) analyzes the problem of
aggregating a set of individual interpersonal comparisons into a summary interpersonal
comparison.
Fourth, the output of a social choice procedure is an ordering of the alternatives
in X, called a social ordering, rather than the selection of one or more members of
X. Moreover, this chapter only offers a few highlights of the literature that allows the
input (the individual preferences) or the output (the social ordering) to be uncertain
or fuzzy. Salles (1998) surveys fuzzy social choice. Various aspects of uncertainty in
social choice are surveyed in three of the chapters in this Handbook: Chapter 10 by
d'Aspremont and Gevers (2002), Chapter 11 by Blackorby, Bossert and Donaldson
(2002), both in this Volume, and Chapter 24 by Coughlin (2002), in Volume 2.
Section 8 below contains a sketch of the literature on rules that select a subset of X
as a function of individual preferences, instead of socially ordering X. Deb (2002,
Chapter 18 in Volume 2 of this Handbook), provides a thorough treatment. See also
Le Breton and Weymark (2002a, Chapter 16 in Volume 2 of this Handbook).
Fifth, preference revelation is non-strategic. We do not discuss what happens when
"the ideals of the just society meet with the play of self interest" [Arrow (1997)].
The potential for individual preference misrepresentation was already recognized by
Arrow in 1951 [p. 7 of Social Choice and Individual Values]. However, the strategic
and non-strategic aspects of social choice can be dealt with separately. We can
discuss various sets of conditions, and deduce how each set constrains the social
choice process - as we do in this chapter. Then we can analyze how the requirement
Ch. 1: Impossibility Theorems in the Arrouian Framework 39

that an individual be unable to profit from strategic misrepresentation of his or her

preference constrains the social choice process - as in BarberA (2002, Chapter 23
in Volume 2 of this Handbook), and Fishburn and Brams (2002, Chapter 4 in this
Volume). Finally, we can identify the social choice rules that come closest - in a
sense that would need to be specified - to satisfying both types of constraints. The
purpose of this chapter is to deduce the implications of a set of conditions that do
not reflect strategic considerations. Sen (1986, 1987), Moulin (1994), and Pattanaik
(1997) also provide surveys of non-strategic social choice in the Arrovian framework.
Thomson (2001) is an excellent general survey and exposition of the axiomatic method
in economics.

2. Definitions and framework

X is the set of all alternatives or outcomes. It has at least two members. We use IX I
to denote the cardinality of X, and if Z is a subset X then X \ Z is the complement
of Z in X. A binary relation ~ on X is a comparison of the members of X two at
a time, thus: x tSy. The statement x t y is read "x is weakly preferred to y", or "x is
preferred or indifferent to y". If x y and y t x both hold we write x -y, and say that
"x is indifferent to y".
A binary relation > on X is complete if for all x, y C X, either x t y or y t x holds.
Note that a complete relation is reflexive, which means that x x holds for each x E X.
The asymmetric part of t is denoted by >-, so x >-y if and only if x t y holds but y t x
does not. When x >-y we often say that x is strictly preferred to y, or that x ranks strictly
above y in . We say that t is transitive if for all x, y, and z in X, if x t y and y z
then x z. We often write x y z to indicate that x y and y z both hold. We
say that a complete and transitive relation t is an ordering. We sometimes use other
symbols, such as R, to denote an ordering. For any Y C X we can define an ordering
on Y in the obvious way, and we let P(Y) denote the set of orderings on Y.
A transitive relation t cannot have x t y t z >-x because z >-x cannot hold ifx i z.
Moreover, if t is complete and we never have x t y t z >-x then:

x >-y z implies x >-z.

x y >-z implies x >-z.
x >-y >-z implies x >-z.
x -y - z implies x z.

We say that the binary relation t on X contains a cycle if we have xl > x2 - .-.
>- XT >- X for some choice of x,x 2 ,... ,XT from X. Why are cycles an issue when we
use a binary relation on X to identify one or more best alternatives in X? Because if
X = Xl, X2 .... , XT} and xl, -X2 > · >- XT X then X contains no "best alternative"
if best alternative means an outcome a in X such that a t b for all b E X. In general,
transitivity is sufficient but not necessary for the existence of a best alternative. See
40 D.E. Campbell and JS. Kelly

Suzumura (1983) for a thorough treatment of binary relations and choice. Transitivity
and related properties can be derived from various formalizations of the desire to avoid
computational complexity or burdensome information processing costs: see Campbell
(1975), Beja (1989), and Johnson (1990, 1995), for instance.
It is often useful to employ a model in which an individual is never indifferent
between distinct alternatives, in which case we say that the preference ordering is
linear. Formally, we say that the complete binary relation Z is antisymmetric if for
all x, y EX, x y and y x imply x=y. A binary relation is a linear ordering if
it is complete, transitive, and antisymmetric. Let L(Y) denote the family of linear
orderings on Y.
The inverse R-l of the binary relation R is defined by setting xR- y if and only if
yRx (for arbitrary x and y). Note that R -t is antisymmetric (or complete, or transitive,
or linear) if and only if R is. The restriction of binary relation R on X to the subset Y
of X is denoted R Y. We define R Y by setting xR Yy if and only if xRy and x and
y belong to Y.
The set N of individuals whose preferences are to be consulted is the finite set
{1,2,. .. , n with n > 1. If J is a subset of N, the set of individuals in N but not in J
is denoted by N \ J. A domain is some non-empty subset P of P(X)N. A member p
of P(X)N is called a profile, and it assigns the ordering p(i) to individual i e N, where
p(i) is interpreted as i's preference ordering at profile p. If S is a subset of P(X), then
a member p of SN is a profile for which p(i) belongs to S for each i E N. A social
welfare function for outcome set X and domain P9 is a functionf from P into the set
of complete binary relations on X. We say that has a full domain if either P = P(X)v
or P=L(X) .
Given social welfare function f on domain P, if f(p) is transitive for each p eP
we say that f is transitive-valued. For each p E P, f(p) is interpreted as the social
ranking of X determined byf when individual preferences are specified by p. We will
often use ti to represent p(i). Then x ij indicates that individual i weakly prefers x
to y at profile p. If there is some danger of confusion with individual preference
at another profile, we will write x iy to indicate that person i weakly prefers x
to y at profile p. Similarly, x -iy means that i strictly prefers x to y, with x >-i Y
if necessary, to indicate that i strictly prefers x to y at profile p. Of course, x -iy
means that individual i is indifferent between x and y, with x -Py if it is necessary
to clarify the fact that p is the profile in question. We will let ~ represent f(p), so
that x y means that x ranks at least as high as y in the social relation determined
by f at p. If we have to distinguish profile p from another profile then we write
x f( p) y, x S>-(p) Y, and x f(p) y for, respectively, x ranking at least as high as y inf(p),
x ranking strictly higher than y in f(p), and neither alternative ranking above the
other in f(p).
We next introduce standard restrictions on the domain P of a social welfare
function f. First, we define p Y, the restriction of profile p to the subset Y of X.
It is the function that assigns the ordering p(i) Y to arbitrary i EN. P(Y) is the
set of profiles p E p(y)N such that there is some r P for which p = r Y. That is,
Ch. 1: Impossibility Theorems in the Arrovian Framework 41

P(Y) = {r IY: r E P}. Given the domain P, we say that a triple {x,y, z} of alternatives
is free if either

(A) P({x, y, z}) = P({x, y, z})N or (B) P({x, y,z)}) = L({x,

L(x,y,y, })N.

In words, statement (A) means that every profile of orderings on {x,y, z} is embedded
in some profile in P. We say that P has the free triple property if X has at least three
alternatives and every triple from X is free. We adopt the convention that whenever
we display a set {x,y,z,... } we mean that the alternatives are distinct. Similarly,
"pair" will mean a set containing two distinct alternatives and "triple" will mean a set
containing three distinct alternatives. All of the theorems to follow that hypothesize
a free triple domain, are valid if either (A) or (B) is used to confirm the free triple
property. The following is an example of a free triple domain that is a proper subset
of L(X)N:
Example 2.1. X= {w,x,y,z} and P7= WN, where W is the set of linear orderings R
on X such that w is not a maximal or minimal element of R. Use condition (B) to
confirm that this domain has the free triple property. ·
The following condition, due to Kalai, Muller and Satterthwaite (1979), is weaker
than the free triple property. We say that P has the chain property if IX I >3
and for every two ordered pairs (x,y) and (w,z) of alternatives in X, there exists
an integer k and a sequence vl,v2,.. , vk such that all of the triples {x,y, vl},
{y, vI,U2},
v2} , v2, v3 },..., {Vk, w,z} are free. Kalai, Muller and Satterthwaite use the
term saturating, which is equivalent to the chain property if every pair of alternatives
is free. (To define a free pair just replace {x,y,z} with {x,y} in definition (A) of a free
triple). But Kalai, Muller and Satterthwaite are concerned with economic domains in
which there are pairs {x,y} such that x gives everyone more of every good than y, and
hence every individual strictly prefers x to y at every profile in the domain. A pair
{x,y} of alternatives is called trivial if p {x,y} = r {x,y} for any two profiles p and r
in the domain. A domain is saturating if there are at least two non-trivial pairs, and for
any two non-trivial pairs {x,y} and {w,z} there is a sequence of free triples connecting
them. The chain property plays a central role in this chapter.
One special consequence of the chain property we will use later is that, given any
pair {x,y} of alternatives, there exists an alternative z such that {x,y,z} is a free triple.
There are domains that have the chain property but not the free triple property:
Example 2.2. X contains more than three alternatives and a is a particular element
of X. Let R be a fixed linear ordering on X \ {a}. The domain P consists of all
profiles r of linear orderings such that each r(i) restricted to X \ {a} is either R or its
inverse R- '. (The position of a in r(i) is unrestricted, except that it cannot be indifferent
to another alternative). P does not have the free triple property, but it does have the
chain property: given (x,y) and (w,z), if {x,y,a}, {y,a,w}, and a,w,z} each have
three members then each is a free triple. (If one or more of the sets contains fewer
than three alternatives then there is a shorter chain). ·
42 D.E. Campbell and JS. Kelly

Our examples will feature domains that are product sets, but this is not required by
the theorems that hypothesize free triples or chains of free triples. For instance, the set
of all profiles in L(X) N such that all individuals have the same top ranked alternative
has the free triple property if X has more than three members, but it is not a product set.
Of special importance are the domains of economic preferences, where the convexity
and monotonicity assumptions are incompatible with the free triple property, but chains
of free triples abound. Le Breton and Weymark (2002a, Chapter 16 in Volume 2 of
this Handbook) explore such preferences in depth.
The key lemmas in the next section use free triples to justify the existence of certain
profiles that are related to other profiles. For example, let q be any profile, let J be
any subset of N, and let p be any profile for which x >- z for all j E J. If {x,y,z} is a
free triple, there exists a profile r such that

r I{x,z} =p {x,z}, and r {y, z} = q {y, z}, and x - y for all J

To verify this, we first create s(i) from p(i) by sliding y up or down so that it is
indifferent to z in s(i) if it is indifferent to z in q(i), andy is strictly above (resp., below)
z in s(i) if it is strictly above (resp., below) z in q(i). Because x >Pz for allj E J, we
can do this while satisfying x y for all j E J. The profile s will have the required
properties, but it does not necessarily belong to the domain. However, if {x,y,z} is free
then there will be some profile r in the domain such that r {x,y,z} =s {x,y,z}. This
means that r I{x,z} =p {x,z}, r I{y,z} = q {y,z}, and for all j C J we have x strictly
preferred to y in r(j).
We next introduce some restrictions on the social welfare function. These elementary
conditions partially describe the way that we want the social ordering to be sensitive
to individual preferences.
Pareto criterion. For every p E P and all x,y EX, we have x >-f(p)y if x >-Py for all
ioN.
In words, if everyone ranks alternative x strictly above alternative y then x must rank
above y in the social ordering. In an economic context, this condition is used to
eliminate every conceivable type of waste. By definition, there is waste somewhere if it
is possible to rearrange production and consumption activities so as to make everyone
better off. Note, however, that the Pareto criterion is well defined for any X and P. It is
a generalized non-wastefulness condition, and is often called the weak Pareto criterion
in the social choice literature.
Later in this chapter we will need to consider relaxing Pareto. For example,
nonimposition requires that there is some profile p at which x ranks at least as high
as y (but not necessarily higher than y) in the social ordering f(p).
Nonimposition. For all x,y EX, there exists some p E P such that x tfj(p)y.
Note we can have both x >f(p)y and y /(p) x, in which case neither y (-/(p)x nor
x >-f(p)y holds. In fact, nonimposition is satisfied by the null rule, which puts every al-
ternative indifferent to every other alternative in the social ordering at each profile. The
Ch. 1: Impossibility Theorems in the Arrovian Framework 43

Pareto criterion implies thatf is not null (unless the domain is severely restricted), but
Pareto is obviously much stronger in general than the requirement thatf is non-null.
Non-null. There exist x,y EX and p E P such that x >f (p)y.
Our next condition, Independence of Irrelevant Alternatives (IIA), is quite different
in spirit from the Pareto, or nonimposition, or non-null criteria, each of which requires
some responsiveness to individual preferences on the part of the social welfare
function. IIA requires the social ordering of x and y to be the same in two situations
if, person by person, the individual ordering of x and y is the same in those two
situations:
IIA. For all p,q E P and all x,y EC X, ifp {x,y} = q I x,y} thenf(p) I x,y} =f(q) I x,y}.
When IIA is satisfied we can define the social welfare functionf I Y, the restriction of
f to an arbitrary non-empty subset Y of X. The domain forf I Y is P(Y), andf IY is
the social welfare function mapping P(Y) into P(Y) and such that the image off I Y
at profile p is f(r) I Y, for any r E P such that r Y =p. IIA is an interprofile condition
because it restricts the social ordering at one profile in a way that is in part conditional
on how it restricts the social ordering at another profile - or profiles. Fishburn (1987)
surveys interprofile conditions and their connection with social choice impossibility
theorems.

3. Fundamental lemmas and Arrow's theorem

The notion of a decisive coalition lies at the heart of the analysis to follow. A coalition
is a subset of N, the society. Informally, J C N is decisive if the members of J
can determine the social ordering whenever they exhibit unanimous strict preference,
whatever the alignment of preferences of the complementary coalition, N \ J. Formally,
coalition J C N is decisive for the ordered pair (x,y) E X xX if, for all p E P, x >-Py
for allj E J implies x >-f(p) y. J is decisive if it is decisive for all pairs. Coalition J C N
is inversely decisive for (x,y)EXxX if, for all pCG , x -y for all j eJ implies
y -f(p)X. J is inversely decisive if it is inversely decisive for all pairs. Of course,
the families of decisive and inversely decisive coalitions depend on f. Individual h
is a direct dictator, and the social welfare function is directly dictatorial, if h} is
decisive. Similarly, individual h is an inverse dictator if h} is inversely decisive.
A social welfare functionf is non-dictatorial if it does not have a direct dictator.
As the next two paragraphs reveal, whether or not a trio of decisive coalitions has a
non-empty intersection determines if an intransitive social ordering can be precipitated.
Hence, the structure of the family of decisive coalitions plays a key role.
Example 3.1. Majority rule: In this case a coalition is decisive if it has more than
n/2 members, where n is the cardinality of N. In the case of majority rule, three
decisive coalitions H, I, and J can have an empty intersection. For instance, if N
has five members, take H={1,2,3}, I = 1,4,5}, and J= 2,4,5}. Then there will be a
44 D.E. Campbell and AS. Kelly

cycle x >-y >-z >-x at a profile p for which everyone in H prefers x to y, everyone in I
prefers y to z, and everyone in J prefers z to x. We have x - y because H is decisive,
y >-z because I is decisive, and z >-x because J is decisive. A
Now, let f be any social welfare function. If there are three nonempty decisive
coalitions H, I, and J with an empty intersection and there is at least one free triple
{x,y,z} thenf (p) will have a cycle for some p. Consider the following profile p:

HnJ H\J N\H

z x y
x y z
y z x

(One of the coalitions H nJ, H \ J, or N \ H might be empty. The table has the
obvious interpretation: every individual in the coalition H n J strictly prefers z to x
and x to y, everyone in H \ J strictly prefers x to y and y to z, and everyone not in H
strictly prefers y to z and z to x. The individual orderings of the pairs not in {x,y,z} are
irrelevant to the argument, and hence are not displayed.) At profile p we have x >-y
because H is decisive and everyone in H strictly prefers x to y. Because (H n J) n I
is empty, we also have y >-z because I is decisive and everyone in I (everyone in
N \ (H J) in fact) strictly prefers y to z. Finally, we have z >-x because J is decisive
and everyone in J strictly prefers z to x at p. Therefore, fJ(p) contains the cycle
x-yb-z>-x.
We will see that transitivity of the social ordering, along with Pareto and IIA, forces
a great deal of structure on the set of decisive coalitions. This structure in turn will
imply that the social welfare functionf is dictatorial. We begin our investigation by
proving that nonimposition, IIA, and transitivity of eachf(p) cause decisiveness over
a single pair of alternatives to spread to all pairs. That puts a severe and regrettable
constraint on the construction of a social choice rule.
The contagion lemma. X is any set with at least three alternatives and P C P(X)N is
a domain with the chainproperty. If social welfarefunctionf on P is transitive-valued
and satisfies IIA and nonimposition, then
(1) any coalition J C N that is decisive for some pair of alternatives is decisive for
every pair of alternatives, and
(2) any coalition J C N that is inversely decisive for some pair of alternatives is
inversely decisive for every pair of alternatives.
Proof: We will prove statement (1). Statement (2) can be proved in the same way. We
first show:
(a) If {x,y,z} is a free triple in X and J is decisive for the pair (x,y) then J is decisive
for (x,z).
Ch. 1: Impossibility Theorems in the Arrovian Framework 45

Let p be any profile for which x -P z for all j E J. We want to show that x (p) Z. By
nonimposition, there is a profile q for which y f( q)z. Since {x,y,z} is a free triple,
there exists a profile r such that

rl {x,z} =pi {x,z}, and r {y,z} = q {y,z}, and x -jy for allj C J

Then we have x >f(r)y by decisiveness of J for (x,y), andy tf (r) z by IIA. Then x >f(r) z
becausef(r) is transitive. Therefore, x >-f(p) z by IIA, and thus J is decisive for (x,z).
For the second step we show:
(b) If {x,y,z} is a free triple and J is decisive for the pair (x,y), then J is
decisive for (z,y).
Suppose z>-Py for all jJ at a profile p. We want to show that Z>f(p)y.
Nonimposition implies that there is a profile q for which z f(q)x. Since {x,y,z} is
a free triple, there is a profile r such that

rl {y,z} =pl {y,z}, and r {x,z} = q x,z}, and x >jy forallj J

Then we have z tf(r) x by IIA and x >-f(,-) y by decisiveness of J for (x,y). Then z Tf(r) y
becausef(r) is transitive. Therefore, z >-(p)y by IIA, and thus J is decisive for (z,y).
Now suppose {x,y,z} is a free triple and J is decisive for the pair (x,y). By (a), J
is decisive for (x,z), and (b) implies that J is decisive for (z,y). Statement (a) and the
decisiveness of J for (z,y) imply that J is decisive for (z,x). Then from J's decisiveness
for (z,x), we can show J is decisive for (y,x) by (b) and from that, J is decisive for
(y,z) by (a). Hence, if J is decisive for one ordered pair from a free triple, it is decisive
for every ordered pair from that triple.
Finally, suppose J is decisive for the pair (x,y) from X, and let (u, v) be any other
pair. By the chain property, there is a sequence vl, v2,... , k such that all of the
triples {x,y, v }, {y, vi,},v} vi, 2, 3}, ... {Vk, , v} are free. Since J is decisive for
(x,y) and {x,y,vl} is free, J is decisive for (y,vo). Since J is decisive for (y,ul)
and {y,vl,v 2} is free, J is decisive for (vl, 2). Continuing along this sequence, we
establish the decisiveness of J for (u, v). 1
Because there are two tests for a free triple domain, depending on whether or
not individual indifference is assumed away, there are two results in the contagion
lemma. One applies to social welfare functions with domain P C L(X)N, which means
that individuals are never indifferent between distinct alternatives; the other allows
individual indifference.
Now we use a series of examples to highlight the role of each of the assumptions
in the hypothesis of the contagion lemma. Each example presents a social welfare
function that satisfies all but one of the assumptions, and it identifies a coalition that
is decisive for at least one pair but not all pairs. First, we will see why the assumption
that X has at least three members is crucial.
Example 3.2. X = {x,y}. Definef on P(X)N by setting x >-y unless y >-i x for all i E N,
in which case y >-x. When IX = 2, transitivity and IIA are satisfied vacuously. This
46 D.E. Campbell and JS. Kellyv

rule f clearly satisfies Pareto, and hence nonimposition. Note that {i} is decisive for
(x,y) for all i e N, but N is the only decisive coalition for (y,x). We don't get contagion
of {i} forX = {x,y} because we don't have a third alternative z to act as a "carrier". Sen
(1976) coined the term "Paretian epidemic" to refer to the transmission of decisiveness
for a pair to global decisiveness. Kelsey (1988) provides considerable insight into the
transmission mechanism. ·
Nonimposition is an important assumption in the contagion lemma, as the next
example shows:
Example 3.3. X has at least three alternatives. Partition X into two nonempty
subsets Y and Z. To define f we let p be an arbitrary profile in P(X)N: set y >- (,) z
for all y in Y and all z in Z, and letf (p) order the members of Y exactly as p(l) does,
with the members of Z ordered according to p(2). The singleton { I } is decisive for
any pair from Y, but no coalition is decisive for (z,y) if z C Z and y C Y. Of course,
f fails to satisfy nonimposition because we never have z l(p) y if y belongs to Y and
z belongs to Z. U
Next we illustrate that IIA is crucial:
Example 3.4. Choose a particular pair of alternatives v and w, and definef on P(X) '
by setting f(r)= r() if v >- w, otherwise f(r)=r(2). Coalition 1} is decisive only
for the pair (v, w). IIA fails because X has more than two alternatives, and we have
v >f(,)z if v >-' w -' z and z>-w >- u, but z >l(p)v if w >-v> z and z >- w - v.
Note that p I {v, z}= r {v, z} ifp(i) = r(i) for all i > 2.
The next example illustrates the need for the assumption that the social ordering is
always transitive:
Example 3.5. Choose a particular pair of alternatives v and w, and define f
on P(X)N by setting f(r)]{v,w}=r(l){v,w}. For all other pairs {x,y}, we set
fJ(r) I x,y} = r(2) I{x,y}. Then coalition {1 } is decisive for (v,w) and (w, v) but for no
other pairs. It is easy to see whyf(r) is not always transitive: if v - I w and w >2 Z -2 U
then we have v >-w >-z >-v.
Finally, we use an important example of Blau (1957) to show why the chain property
is assumed.
x
Example 3.6. X = w,x,y,z}. The domain P is the set of all profiles p in L(X)@ such
2
that w is at the top of p(l) and at the bottom of p( ), and we set f(p) I {x,y,z} =
p(l) I{x,y,z}, with w at the bottom off(p). Person 1 is decisive for any pair from
{x,y,z}, but not for (w,v) for any choice of v {x,y,z}. Note that {x,y,z} is a free triple,
but there is no free triple containing w, so the chain property fails. U
The contagion result places regrettable restrictions onf. For example, a society may
want coalition J to be decisive for the specific pair (x,y) if alternative x can be derived
from y by having the members of J exchange private commodities among themselves.
Ch. 1: Impossibility Theorems in the Arrovian Framework 47

But the hypothesis of the contagion lemma causes J's decisiveness over (x,y) to spread
to all pairs, even those pairs (w,z) such that alternative w can be derived from z by
having the members of N \ J exchange private commodities among themselves. This
may be far from socially acceptable. We will see that Arrow's conditions are even more
demanding - impossibly demanding.
Clearly, from the definitions of decisiveness and inverse decisiveness, if coalition H
is decisive (resp., inversely decisive) and H C J C N, then J is decisive (resp., inversely
decisive). It follows that either the collection of inversely decisive sets is empty, or the
collection of decisive sets is empty; otherwise N is both decisive and inversely decisive,
and that is not possible if there are two alternatives x and y such that everyone strictly
prefers x to y at some profile.
Assume the conditions of the contagion lemma, which gives us many of the details
of the structure of the collection of decisive sets. First, note that the chain property
implies that every distinct pair of alternatives is contained in some free triple. Now we
show that iff is non-null, either N is decisive or N is inversely decisive. Because f
is non-null, there is a profile r and a pair of alternatives, x and y, such that, x >f(,)Y.
There exists an alternative z such that {x,y,z} is a free triple. Let p be a profile such
that
Pi {x,y} = r {x,y}, and for all i C N, both x >-i z and y >-i z hold.
Because IIA is one of the conditions, if x z then coalition N is decisive for (x,z),
zf(p)
and thus for all pairs by the contagion lemma. That is, N is decisive if x -f(p)z.
Now suppose z f(p)x. IIA implies x -f(p)y, and thus transitivity yields z f(p)y.
Therefore, N is inversely decisive for (y,z) by IIA, and thus N is inversely decisive
by the contagion lemma.
After we prove the next lemma, it will take only a few additional lines to complete
the proof of Arrow's Theorem. The new lemma gives considerable detail on the
structure of the set of decisive coalitions, or the set of inversely decisive coalitions,
whichever is not empty. The nonempty collection is an ultrafilter. An ultrafilter on a
given set N is a collection U of subsets of N with the following four properties:
(1) N eU and0 U.
(2) For arbitrary subsets H and J of N, if H E U and H C J then J G U.
(3) For arbitrary members H and J of U, the set H n J belongs to U.
(4) For arbitrary subset H of N, if H U then N \ H E U.
The set of all supersets of {1,2} has the first three properties, but not the fourth because
neither { 1} nor N \ {1} belongs. The family of all supersets of { I is an ultrafilter.
The ultrafilter lemma. Let X be any set with at least three members. Suppose that
p C P(X)N has the chain property and f is a non-null and transitive-valued social
welfare function satisfying IIA and nonimposition. Then the collection of decisive sets
or the collection of inversely decisive sets - whichever is non-empty - is an ultrafilter.
Proof: We will assume U is a nonempty collection of decisive sets; the proof for
inversely decisive sets is similar.
48 D.E. Campbell and AS. Kelly

(1) N e U was established in the remarks just before the statement of the theorem. To
prove 0 C U choose any two distinct alternatives x and y, and any profile r such
that x >-iY for all i in N. (The chain property implies that such a profile exists.)
We have x >-y because NE U. But trivially, y >-ix for all i in 0, and we do not
have y - x, so 0 U.
(2) Suppose that H U, H C J, and x >-iy for all i in J. Then x >-y for all i in H,
and thus x -y because H CU. Therefore, J is also decisive.
(3) Suppose H E U and J E U. We wish to show Hn J E U. By the chain property,
there is a free triple {x,y,z} and a profile p satisfying

HnJ H\J J\H N\(HUJ)

x y z z
y z x y
z x y x

(Some of these sets might be empty.) Atp we have x >-y because J is decisive, and
y >-z because H is decisive. Transitivity of >- implies x >-z. By IIA we have x >-z
at any profile where everyone in H n J prefers x to z and everyone else prefers
z to x.
We have not yet established the decisiveness of H n J because the preferences
of the complementary coalition have been restricted. Accordingly, we let r be any
profile with x >-,y for all i in H n J. Since {x,y,z} is a free triple, there exists a
profile q with
(i) qj {x,y} = rl {x,y}
(ii) x >-I-qz
y for all i in H n J, and
(iii) z >-qy and z >-qx for all i not in HnJ. Then x f(q)z by the previous
paragraph, and z >f(q) y since N is decisive. Therefore x >-f(q) y by transitivity.
Then x >-f(,r)y by IIA. Therefore HnJ is decisive for (x,y) and so by the
contagion lemma, H n J e U.
(4) If H d U, there exists a pair, x and y, and a profile r such that, x >-ry for all i in H
but y ~f(r) x. We will show that N \ H E U. There is a z such that {x,y,z} is a free
triple. Let p be any profile such that y >-P z for all i in N \ H. Because {x,y,z} is
a free triple, there is a profile q such that

q[ {x,y} = rl x,y},ql {y,z} =pl{y,z}, and x >q z for all i in N.

(For iEN \ H, create q(i)l {x,y,z} from r(i) {x,yz} by sliding z down below
x and y. For i E H, slide z below x but duplicate the ordering of y and z in p(i).
Recall that x >- y for all i E H.) We have y f(q) x by IIA and x >-/(q) Z because
N E U. Therefore, y >-f(q) z by transitivity, and then y >-f(p) z by IIA. Then N \ H
is decisive for (y,z), and thus N \ H E U by the contagion lemma. ]
Ch. 1: Impossibility Theorems in the Arrovian Framework 49

The ultrafilter approach was first applied to the study of Arrow's theorem by Hansson
(1972) and Kirman and Sondermann (1972). They independently proved that the
hypothesis of Arrow's theorem implies, for any nonempty set N, that the family of
decisive coalitions is an ultrafilter. Note that the proofs of the contagion and ultrafilter
lemmas do not depend on finiteness of N. For further discussion of ultrafilters in social
choice, see Brown (1974) and Monjardet (1983).
A handful of examples will highlight the role of each of the assumptions in the
hypothesis of the ultrafilter lemma. We could employ Examples 3.2 through 3.6, used
to examine the hypothesis of the contagion lemma, but this new series will provide
more insight into the role of the assumptions in the ultrafilter lemma. Each of the
following social welfare functions satisfies all but one of the assumptions, and none
of them is dictatorial. Two examples are used to show why transitivity is crucial for
the ultrafilter lemma:
Example 3.7. For arbitrary x, y EX and p E P(X)N. Define f by setting x ~y unless
y >-lx and x >-iy for all i 1, in which case x>-y. Then no coalition, including N,
is either decisive or inversely decisive and so (1) and (4) fail. For many profiles p,
the relationf(p) will not be transitive: if x >- z >-l y and z >-ix >-iy for all i X 1, then
x y zx.

Example 3.8. Pareto extension: For arbitrary x, y EX and p E P(X)N, set x -y unless
x >-iy for all i, in which case x >-y. In this case N is the only decisive set, so (4) fails.
Note thatf(p) is not transitive, because if y >-I z >- x and z -i x >-iy for all i 1 we
have x -y z >-x. The idea that unanimity should be reflected in collective decision
making, whenever it occurs, is about a hundred years old, but Sen (1969, 1970a) made
the Pareto extension rule a part of modern social choice theory, by giving it formal
expression and connecting it to Arrow's hypothesis. ·
The next example shows that IIA is crucial for the ultrafilter lemma.
Example 3.9. For convenience, assume that X is a finite set with at least three
members, and choose some v EX. For arbitrary p EL(X)N set f(p)=p(l) if v is at
the top of p(l), and otherwise setf (p)=p(2). Coalition {1,2} is decisive, but neither
{ 1} nor N \ {1} is decisive, so (4) fails. IIA is not satisfied: suppose that v is at the
top of p(l) and at the bottom ofp(2). Then v Sf(p)z for all z X \ {v}. But if profile r
has r(i) =p(i) for all i E N \ { 1}, and w at the top of r(l) with v in second place, then
we have z f(r) for all z EX \ v,w} although r I{v,z} =p {v,z}. U
If we drop nonimposition from the list of requirements of the ultrafilter lemma then
we cannot derive properties (1) or (4), as the next example shows.
Example 3.10. The domain off is P(X)N. Choose some ordering Q C P(X) and set
f(p) = Q for all p E P(X)N. There are no decisive coalitions, althoughf is transitive-
valued and satisfies IIA. ·
50 D.E. Campbell and J.S. Kelly

The next example shows that we cannot establish (3) of the ultrafilter lemma if we
drop the chain property from the hypothesis.
Example 3.11. There are at least three individuals, and the domain forf is the set
of all p E L(X)N such that for some i E N we have p(h) =p(j) for all h, j E N \ {i}.
Set f(p) =p(j) for any j such that p(j) =p(h) for at least one h j. There are no free
triples, so there are no chains of free triples. No singleton coalition {i} is decisive. But
any two-person coalition is decisive, because if x >-iy for two persons i, then either
these two belong to the set of n - 1 individuals with identical preferences, or else every
i N has x >-iy. In either case we get x >-y. Therefore, {1,2} and {2,3} are decisive,
but their intersection is not. U
The assumption that X has at least three members is crucial. Majority rule,
introduced as Example 3.1, satisfies IIA on any domain, and it is non-null and satisfies
nonimposition - unless the domain is severely restricted. If IXI =2 then f is also
transitive-valued. But majority rule does not have the intersection property (3) if there
is a free pair: suppose N = {1,2,3}. Then {1,2} and {2,3} are both decisive coalitions -
they are majority coalitions - although {2} is not decisive.
The assumption thatf is non-null is essential to the ultrafilter lemma because the null
rule satisfies all the other conditions of the lemma, but it has no decisive or inversely
decisive coalitions.
Now that we have the ultrafilter lemma, the proof of Arrow's impossibility theorem
will be brief. Arrow's original proof employs the same ingredients as the contagion and
ultrafilter lemmas, but it organizes them somewhat differently. We emphasize that no
one had conjectured anything like Arrow's theorem when his justly famous monograph
appeared in 1951. For some personal background about Arrow that is related to his
discovery see: Feiwel (1987), Kelly (1987), and Arrow's own comments in Arrow
(1983, pp. 14).
A variety of proofs of the impossibility theorem are available, including: Fishburn
(1970), Blau (1972), Wilson (1975), Barbera and Sonnenschein (1978), Sen (1979,
1986), McLennan (1980), BarberA (1983), Rubinstein and Fishburn (1986), Suzumura
(1988), Blackorby, Donaldson and Weymark (1990), Saari (1994), Krause (1995),
Geanakoplos (1996), Denicol6 (1996) and the correction in Denicol6 (2001), Pouzet
(1998), Reny (2001), and Dardanoni (2001). These proofs, and the one that we're about
to present, imply that a transitive-valued social welfare function on a full domain must
be dictatorial if it satisfies the Pareto criterion and IIA. Because non-dictatorship is one
of Arrow's requirements, his conditions are inconsistent.
Arrow's impossibility theorem. If X has at least three members and P C P(X)N
has the chain property, then there is no transitive-valued social welfare function f
satisfying IIA, the Pareto criterion, and non-dictatorship.
Proof: Suppose f is transitive-valued and satisfies IIA and Pareto. Note that the
Pareto criterion is equivalent to the statement, "N is decisive", which implies that the
collection U of decisive coalitions is non-empty. U is finite because N is finite, and
Ch. 1: Impossibility Theorems in the Arrovian Framework 51

so we can select a coalition H e U of smallest cardinality. We have IHI > 0 because

0 U. Let j be any member of H. If {j} U then N \ j} e U by part (4) of the
ultrafilter lemma. But then H nN \ {j} e U by part (3). But H nN \ {j} has one less
member than H, contradicting our selection criterion for H. Therefore, we must have
{j} E U, and thus f is dictatorial. ]
The argument just presented is basically a demonstration that for any finite set N, and
any ultrafilter U on N, there is some i E N such that U = {J C N: i E J}: if N is finite
then so is U, and thus n U, the intersection of all the members of U, will belong to U
by repeated application of (3). [We say that U is fixed if nU is not empty. Therefore,
every ultrafilter on a finite set is fixed by (1)]. Because nU belongs to U, if i C nU
then N \ {i} C U and thus {i} C U by (4). Finally, {i} C U, i E nU, and (2) imply that
U={JCN: iEJ}.
Some of the alternative proofs that we have cited are short, although they assume a
full domain. (It is easy to extend such arguments to an arbitrary domain with the chain
property by first of all selecting any two pairs of distinct alternatives {x,y} and {w,z}
from X, and then alternatives ul, 2, . ..,k such that {x,y,v }, {, l, v2},... , {k,W,z}
are all free triples. The rulef IY has a full domain for each of these triples Y, and thus
is dictatorial. The same individual must be the dictator on each Y, because each has
two alternatives in common with its successor.) The longer proof used here uncovers
the ultrafilter structure that is needed for the trade-off results of the next section.
Wilson (1975), Rubinstein and Fishburn (1986), and Krause (1995) each obtain
Arrow's theorem as a special case of an original general aggregation theorem. Both
Barbera and Sonnenschein (1978) and McLennan (1980) derive Arrow's theorem as
a corollary of an original theorem on probabilistic social choice. The Krause proof
also generalizes the results of Gevers (1979) and d'Aspremont (1985), which apply to
preference domains that have enough structure to permit interpersonal comparisons
to be made. [See d'Aspremont and Gevers (2002, Chapter 10 in this Volume) on
interpersonal comparability].
Even if individual j is a dictator for f, it may not be the case that f(r)= r(j) for
each profile r if individual indifference is allowed, as we now illustrate:
Example 3.12. Serial dictatorship: With N = {1,2,... , n}, define the social welfare
functionf on P(X)N: for arbitrary x, y EX set x -y if and only if there is some i N
such that x >-i y, and x ah y for all h <i. Individual 1 is a dictator although we do not
havef(p) =p(1) for allp P(X)N. 0
If the domain is full, we can state an interesting special version of Arrow's theorem
by replacing Pareto with a condition that is much weaker in general: social welfare
function f satisfies weak unanimity if whenever r(i) = r(l) for all i E N and there is
an alternative x that is alone at the top of each r(i), then x must rank alone at the
top of f(r). Note that weak unanimity and IIA imply the Pareto criterion on a full
domain, even if X is infinite. (If X is infinite, we can use Zorn's lemma to guarantee
that there exists some linear ordering Q of X. Then for arbitrary x CX, we can create
52 D.E. Campbell and J.S Kelly

a new linear ordering from Q by removing x and placing it at the top). In general -
i.e., without the full domain assumption - weak unanimity is much less restrictive than
Pareto.
Corollary to Arrow's theorem. Suppose that X has at least three members and that
f is transitive-valuedand satisfies IIA and weak unanimity.
(1) If the domain off is P(X)N then f is directly dictatorial.
(2) If the domain off is L(X)N then there is some i N such thatf (p) =p(i)for all
p G L(X) N. That is, there is a dictator and the social ordering is always identical
to the dictatorS preference.
Note that on L(X) N dictatorship implies IIA, but it is not true that dictatorship implies
IIA on P(X)N, as the next example shows.
Example 3.13. P=P(X)N: For every pEP, set f(p)=p(1) unless x-ly for all x,
y eX in which casef(p) =p(2). Person 1 is a dictator forf, but when individual 1 is
indifferent between all the alternatives then person 2's preferences prevail. This rule
satisfies the Pareto criterion because the social ranking is always identical to someone's
preference ordering, andf (p) is always complete and transitive for the same reason.
But IIA is violated if X has at least three members: suppose x ly and x >Az for all
z EX \{x,y}. Let p(2) be any member of L(X) such that z - x for all z CX \ {x}.
Then x -f(p) Z, but if x~'z for all z EX, and r(i)=p(i) for all i 1, then y f-(,.) x,
although r x,y} =p {x,y}.
The assumption that X has three or more members is critical for Arrow's Theorem.
If IX = 2 then both IIA and transitivity are satisfied vacuously. Any non-dictatorialf
consistent with Pareto - majority rule, for instance - will satisfy all of Arrow's
conditions, except for the requirement that X has more than two members.
If IX > 2, but we are willing to relax, or even set aside, one of Arrow's conditions,
what are the prospects for designing an appealing social welfare function that satisfies
the other conditions? Of course, if we drop non-dictatorship from the list of conditions,
the other Arrow conditions are consistent: just choose some i E N and set f(p) =p(i)
for every profile p. This will certainly satisfy IIA, and f(p) will be transitive for all
p P(X)N. This f satisfies the Pareto criterion, because if everyone strictly prefers
x to y then certainly individual i strictly prefers x to y, in which case f sets x -y.
However, there is no social choice context in which dictatorship would be appropriate.
The consequences of relaxing Pareto, transitivity, the domain condition, and
independence of irrelevant alternatives will be taken up in Sections 4, 5, 6 and 7,
respectively.

4. Relaxing the Pareto criterion

If we simply drop Pareto from the list of conditions, the other Arrow conditions are
easily seen to be consistent - by Example 3.10, for instance, which hasf(p) =f(q) for
any two profiles p and q.
Ch. 1: Impossibility Theorems in the Arrovian Framework 53

Example 4.1. Choose some i E N and set x By if and only y i x. That is, f creates
the social ranking by turning person i's preference ordering upside down. ·
Both Examples 3.10 and 4.1 satisfy all of Arrow's conditions, other than Pareto, but
there is no application in which one of them would be appropriate. We might try to
weaken Pareto somewhat; far enough to give us compatibility with the other conditions,
but still with enough force to disqualify an unresponsive rule like Example 3.10, and
with the responsiveness in the right direction, unlike Example 4.1. Nonimposition, for
instance, is much weaker than the Pareto criterion, but Wilson (1972) proved that it is
strong enough to force rules to be quite undesirable. Iff satisfies nonimposition and
is non-null, then there is either a direct dictator or an inverse dictator, provided that
Arrow's conditions, other than Pareto, are also satisfied.
Wilson's Theorem. Suppose X has at least three members, and P C P(X)N has the
chainproperty. Iff is transitive-valued and satisfies IIA and nonimposition, then f is
null, or directly dictatorial, or inversely dictatorial.
Proof: If N is directly decisive then f is dictatorial by Arrow's Theorem. If N is
inversely decisive, then the ultrafilter lemma implies that the set U of inversely decisive
coalitions has properties (1)-(4). The proof of Arrow's theorem can be adapted to show
that U contains a singleton coalition {j}. Person j must be an inverse dictator. If N
is neither directly nor inversely decisive, thenf is null by the ultrafilter lemma. E
As in the results of the previous section, because there are two tests for a free triple
domain, depending on whether or not individual indifference is assumed away, there
are two results embodied in this theorem. One applies to social welfare functions with
domain P C L(X)N; the other allows individual indifference. Theorem 6.2 in Murakami
(1968, p. 103) is in the same vein as Wilson's Theorem, but not as strong. See Malawski
and Zhou (1994) for an interesting perspective on Wilson's Theorem.
We now employ a trade-off approach to learn more about the nature of admissible
social welfare functions when the Pareto criterion is relaxed. For a general introduction
to social choice trade-off theory, see Campbell and Kelly (1994a, 1997); for a view
of this as part of a general axiomatics program, see Thomson (2001). Here we show
how the scope of an individual's power and the degree of satisfaction of the Pareto
condition can each be measured, allowing us to determine if one can go a long way
towards avoiding dictatorship without departing too much from the spirit of Pareto.
Arrow's theorem implies that every non-dictatorial and transitive-valued social
welfare functionf on L(X)N that satisfies IIA will violate the Pareto criterion. In fact,
if X has a finite number m of members, then there will be at least m - 1 ordered pairs
(x,y) such that y t x for every profile at which everyone strictly prefers x to y. Here's
the proof: iff is a transitive-valued and non-dictatorial rule on L(X)N satisfying IIA,
then there is a profile p and at least one pair {x,y} such that x >-py for all i E N, but
y tf(p)x. Let r be another profile for which x is alone at the top of each r(i) and
y is alone at the bottom of each r(i). Then y f(,) x (by IIA). Let z be an arbitrary
member of X \ {x,y}. If z tf(r) x then we have another violation of Pareto. If x >f(,-) z
54 D.E. Camnpbell and AS. Kelly

then y > (r)z, by transitivity off(r), and that is violation of the Pareto criterion. We
already know that the pair (x,y) violates Pareto, and for every z e X \ {x,y} either (x,z)
or (z,y) will violate Pareto. Counting (x,y), there are at least m- 1 violations. For some
rules there will be exactly m - 1 violations. The argument of this paragraph proves that
if X is infinite then there will be an infinite number of violations of Pareto.
Suppose we look for a non-dictatorialf with a relatively small number of violations
of the Pareto criterion. If thisf satisfies nonimposition then, according to the contagion
lemma, if N were decisive for even one pair, it would be decisive for all pairs. So if
we want to have just a small number of violations of Pareto, there will also have to
be violations of nonimposition. If we allow just a few imposed pairs, can we move
far from dictatorship? We relate this to the question: can we confine the scope of any
individual's decisiveness to a small subset of X? We will say that individual i dictates
on Y C X if individual i is a direct or inverse dictator forf I Y. We illustrate what can
happen with the following example:
Example 4.2. X =X 1 UX 2, withX 1 CX 2 = 0 and X 1 I = IX 2 1= 10. Definef by having
individual 1 dictate on X 1, individual 2 dictating on X 2, and for all x CXI and all
y CX 2, setting x >-y at all profiles in L(X)N. No individual dictates on more than
half of X. But the social ordering of a majority of ordered pairs is fixed in the
sense of being the same at every profile. Of the 20 19 = 380 ordered pairs of distinct
alternatives, only 10 9 + 10. 9 = 180 are not fixed. ·
In general, iff is transitive-valued and satisfies IIA then either some individual is
a dictator (or an inverse dictator) on a set containing at least half the members of X,
or else over half of the pairs of alternatives are fixed. To prove a general trade-off
theorem relating the size of largest sets on which an individual dictates to the number
of fixed pairs, we need a preliminary result that shows that the kind of partitioning of
X into X 1 and X 2 displayed in Example 4.2 is inevitable, unlessf is null or dictatorial
or inversely dictatorial. We begin by defining the binary relation >> on the family of
nonempty subsets of X, for arbitrary social welfare function f: set Y >>f Z if and only
if y >-f(p) z for all y E Y, z E Z, and p E P. Most of the following lemma was proved
in Wilson (1972). Note that it only assumes IIA and transitivity of the social ordering
in addition to the free triple property. Therefore, it is valid regardless of the number
of violations of the Pareto criterion.
Wilson's partition lemma. Let X be any set of alternatives. Suppose that P C p(X) N
has the free triple property, and f satisfies IIA and is transitive-valued. Then > is
transitive, and there is a unique partition C of X such that either Y Z or Z of Y
for any two distinct members Y and Z of C. Moreover for all Y E C containing more
than two members, f Y is null, or directly dictatorial, or inversely dictatorial.
Each member Y of the partition C is called a component off. Iff Y is either directly
or inversely dictatorial for a member Y of C, we refer to Y as a dictatorial component.
Iff I Y is null we refer to Y as a null component. Proof of the above version of the
lemma can be found in Campbell and Kelly (1993). The key is showing that f Y
Ch. 1: Impossibility Theorems in the Arrovian Framework 55

satisfies nonimposition if Y is a component off. Then Wilson's Theorem is applied

to f I Y if Y has three or more members.
Notice that for the partition lemma we strengthened the chain property to the free
triple property. The chain property is not sufficient here:
Example 4.3. X contains at least four alternatives. Choose an element a of X, and
let R be a fixed linear ordering on X \ {a}, and let O be the ordering on X \ {a} for
which each alternative is indifferent to every other alternative. The domain P consists
of all profiles r such that, for each i N, the restriction of r(i) to X \ {a} is R or
R - 1 or O. The position of a in r(i) is unrestricted. P has the chain property: given
distinct pairs (x,y) and (w,z) of alternatives from X \ {a}, the triples {x,y,a}, y,a,w},
and {a,w,z} are free. Letf be the rule that has alternative a at the top of everyf(r),
andf(r) (X \ {a})= O if that is the ordering of X \ {a} for even a single individual.
Otherwise f(r) I(X \ {a})=R, unless r(i) (X \ {a})=R' for a strict majority of the
individuals i, in which casef(r) (X \ {a}) =R- l . Thenf satisfies IIA and eachf(r) is
transitive. The components off are {a} and X \ {a}, butf I(X \ {a}) is not inversely
or directly dictatorial, nor is f I(X \ {a}) null. ·
We say that the pair (x,y) is fixed by f if f(p) {x,y} =f(r) {x,y} for all profiles
p and r in the domain off. Hence, the social ordering of fixed pairs is completely
unresponsive to individual preferences. The partition lemma reveals that iff is neither
inversely nor directly dictatorial then some pairs will be fixed, and that fixed pairs have
their social ordering determined independently of individual preferences. The lemma
also enables us to establish a lower bound on the number of fixed pairs as a function
of the scope of dictatorial power.
Consider first the case of a finite outcome set X, with m members. If we don't want
individuals having dictatorial power (direct or inverse) over large subsets of X then
there will have to be a lot of components or else one or more large components that are
null. Either of these cases leads to many fixed pairs. Assuming a free-triple domain,
if f is transitive-valued and satisfies IIA we can prove that either some individual
dictates over more than half the outcome set or at least half of the pairs of outcomes
have their social ranking determined without consulting anyone's preferences. In fact,
for any fraction t > 2, either there will be some individual who dictates on a subset
containing more than the fraction t of outcomes, or at least the fraction 1 - t of the pairs
of outcomes have their social ranking fixed independently of individual preference.
This tells us that even if we do not insist on satisfying all of Arrow s criteria in
a strictly logical sense, we may not be able to come close to satisfying them all
in spirit: every transitive-valued social welfare function satisfying IIA will either
violate the Pareto criterion at many points, or there will be an individual with a lot of
power.
The trade-off theorem. Let m denote IXI. Assume that < t < 1, and 7 C p(X) N
has the free triple property. If f satisfies HA and is transitive-valued then either f
56 D.E. Campbell and AS. Kelly

has a dictatorialcomponent with more than tm alternatives, or else the fraction oJ

ordered pairs of distinct alternatives that arefixed by f is at least (1- t).
We will see why t < 2 has to be excluded from the hypothesis.
Example 4.4. X= {Xl,X 2 , ... ,xm}, where m =2k, an even number. Each {xi,xi }
is a component off for i= 1,3,5,..., m-1. The alternatives in each component are
ordered by majority rule, and for every profile we set xi - xj if xi and x belong to
different components and i <j. Note that each f(p) is transitive for every profile p in
P(X)N, because each component contains only two alternatives. If we set t = ± then
we can say that no dictatorial component has more than mt members, because there
are no dictatorial components. There is a total of m(m - 1) pairs in X. If t = ± then
(1 - t) m(m - 1) = (m - 1)2. Are there this many fixed pairs? The only pairs that are not
fixed are those contained in a component. Then there are m pairs that are not fixed, and
hence m(m - 1)- m = m(m -2) fixed pairs. This number is smaller than (m - 1)2. [
We will look at an example that does have some dictatorial components.
Example 4.5. n>4, m=12, and X={xI,x 2 ,...,x 2 }. We define f on L(X) by
first identifying its components: They are {x,x 2,x 3}, {x4,x 5 ,x6}, {x 7,x 8,x 9 }, and
{xio,xl, xl 2 }. Let person i dictate within the ith component, with xh, >-xj ifxh and xj
belong to different components and h <j. There are 132 pairs in X. Each of the four
components has 6 ordered pairs of distinct alternatives, so there are 24 pairs that are not
fixed. Then there are 108 fixed pairs. If t= I then no dictatorial component has more
than mt = 3 alternatives. It can be verified thatf(p) is transitive for every p in P(X)N.
Clearly,f satisfies IIA. The theorem says that there will be at least ()(132)= 99 fixed
pairs, and the actual number is even greater. U
In Example 4.5 slightly over 80% of the ordered pairs are fixed. This means that over
80% of the pairs are socially ordered without consulting anyone's preferences. The
6 ordered pairs in the component {xI,x2,x 3 } are socially ordered without consulting
the preferences of any member of the coalition N \ {1}. Because the 108 pairs drawn
from different components are ranked independently of any individual's ordering, we
have 132- or 86%, of the ordered pairs from X socially ordered independently of
the preferences of anyone in N \ {1. In addition, the six pairs from {x 4 ,X5 ,X6 } are
socially ordered without consulting the preferences of anyone in the coalition N \ {2}.
Therefore, 12, or about 91%, of the ordered pairs are socially ranked independently
of the preferences of anyone in N \ {1,2}, and 126,
132' or 95%,
5%, of
of the
the pairs from X
pairs from X are
are
socially ordered independently of the preferences of anyone in N \ {1,2,3}. In fact, for
any transitive-valued social welfare function satisfying IIA, if the integer : is small
then there is some subset S C N consisting of all but persons such that at least the
fraction 1 - of outcome pairs have their social ranking determined independently
of the members of S. We say that the pair (x,y) is independent of S C N if for all
profiles p and r in the domain we have f(p) {x,y} =f(r) I{x,y} whenever p(i) =r(i)
for all iEN\S.
Ch. 1: Impossibility Theorems in the Arrovian Framework 57

The independence theorem. XI = m. Let B3 be any positive integer such that

[3 < (m - /m)2/(m - 1) and [3 <n. If P C L(X)N has the free triple property, and f
is transitive-valued and satisfies IIA, then there exists a subset S of N such that
IS >n n-3 and the fraction of pairs (x,y) that are independent of S is at least
1 - (m + 3)2/413m 2 .
For example, if there are 100 individuals (n = 100, the order of magnitude of a
legislature) then 90% of the members of N are not consulted in determining the social
ranking of over 97% of the pairs of alternatives, whatever transitive-valued social
welfare function we employ, if it also satisfies IIA. (Set 3 = 10). 3 has to be small
relative to the total number of alternatives for this to hold. In that case, 1- 1 is a
good approximation to the lower bound of the theorem if there is a reasonably large
number of outcomes. The trade-off and independence theorems were first stated and
proved in Campbell and Kelly (1993).
When X is a measurable subset of Euclidean space we face virtually the same
dilemma as in the trade-off theorem, and the same disenfranchisement problem as in
the independence theorem, but with 1 - I itself as the bound for the second theorem
[Campbell and Kelly (1995c)]. For countable X, an asymptotic density version of the
trade-off theorem is presented in Campbell and Kelly (1995d). When X is the space
of allocations of public goods, and the domain is restricted to individual preferences
with classical economic properties, there is a counterexample to the conclusion of the
partition lemma [see Campbell and Kelly (1995c)]. However, it is not known if there
is a counterpart to the trade-off theorem in that case.
A binary relation t on X is continuous if for all x EX, the sets {z EX: z >-x} and
{z EX: x >-z} are open. If X is a connected T 1 space, then we face an extreme trade-
off within the family of rules satisfying IIA and continuity of the social ordering:
The extreme trade-off theorem. Assume that X is a connected T 1 space and P has
the free triple property. Iff satisfies IIA, andf (p) is a continuous orderingfor each
p E P, then f is constant or dictatorialor inversely dictatorial [see Campbell (1992a,
Theorem 2)]. Campbell (1992c, Chapter 8) also contains a proof

5. Relaxing transitivity

Sen (1969) began the formal study of the consequences of relaxing the transitivity
requirement. If we simply drop transitive-valuedness from the list of conditions, the
other Arrow conditions are easily seen to be consistent:
Example 5.1. Paradox of voting [Condorcet (1 785)]: Simple majority rule is defined
on P(X)N by setting xtf(r)y if and only if {iEN: x y} has at least as many
members as {i E N: y tr x}. It is easy to confirm that majority rule satisfies the Pareto
criterion, IIA, and non-dictatorship. But transitivity does not hold in general: suppose
there are three individuals and X = {x,y,z}. Consider the profile
58 D.E. Campbell and JS. Kelly

1 2 3

z x y
x y z
y z x

Then x-y>-z>-x. Nurmi (1999) presents a large variety of voting paradoxes,

accompanied by insightful discussions. U
McGarvey (1953) was the first to prove that, given any complete binary relation >
on X, no matter how many intricate cycles are embedded in it, for n sufficiently large
there exists a profile such that the social ranking generated by simple majority rule is A.
McGarvey's bound on the number of individuals needed to replicate an arbitrary binary
relation was substantially reduced by Stearns (1959). See also Erd6s and Moser (1964).
Now, suppose that there are k bills under consideration, and that each will either be
adopted or rejected. In that case, the members of X are k-vectors of zeros and ones.
An individual's preference ordering on X is separableif her preference concerning any
given bill is independent of the fate of any other bill. Hollard and Le Breton (1996)
establish that for every separable binary relation t on X there is a profile of separable
individual preferences for which the majority rule relation is >. This result is extended
by Vidu (1999).
Profiles that yield majority rule cycles are not rare. As early as Black (1948,
1958), it was shown that for the small numbers case, the proportion of profiles at
which intransitivities are generated by the simple majority rule increases with the
number of alternatives and with the number of individuals (taking some parity issues
into account). A general theorem is provided by Kelly (1974), which also presents a
conjecture regarding a similar monotonicity for the proportion of profiles at which a
majority winner fails to exist [see Kelly (1994c)]. There are many sampling results
supporting this conjecture [see Gehrlein (1997)] and some theoretical support [see
Fishburn, Gehrlein and Maskin (1979a,b)].
A super majority rule is determined by a fraction a > , with x -y if and only if
at least an individuals strictly prefer x to y. (Assume linear individual preferences.)
Balasko and Crbs (1997) show that for a > 0.53 super majority cycles become rare
events as the number of alternatives increases. However, Tovey (1997) demonstrates
that the paucity of cycles for a > 0.53 is a consequence of the preponderance of
ties. Mala (1999) demonstrates that McGarvey's theorem does not hold for any super
majority rule.
Alternative x is a majority winner at profile p if x ty for all y, where is the
majority rule relation atp. McKelvey (1979) has shown that in a spatial voting context,
if there is no majority winner, i.e., an alternative that beats or ties all others, then
there is a voting cycle over all of X. The seminal paper is Plott (1967). See also
Schofield (1985), Enelow (1997), and Banks (1996). The spatial model uses two-
Ch. 1: Impossibility Theorems in the Arrovian Framework 59

dimensional (or higher) Euclidean space and a special family of preferences, each of
which is characterized by a bliss point and a distance function. We say that bi in X is
individual i's bliss point at profile p if bi is the unique most-preferred alternative for
p(i). We move down the preference ordering as alternatives get more and more distant
from the bliss point.
The rest of this section explores the implications of weakening the transitivity
requirement. We say that a binary relation is quasitransitive if for all x, y,
and z in X, x >-y - z implies x - z. Note that transitivity implies quasitransitivity.
Quasitransitivity preserves transitivity of strict preference but, unlike full transitivity,
does not impose the problematic transitive indifference property. In general, if X is
finite and is quasitransitive then there exists an x E X such that x t y for all y E X.
If f(r) is quasitransitive for every profile r in the domain off we say that the social
welfare functionf is quasitransitive-valued.
Example 5.2. Recall the Pareto extension rule, for which x-y holds unless x >-iy
for all i C N, in which case x >-y. This rule satisfies all of Arrow's conditions except
transitivity, but f is quasitransitive-valued: if r is an arbitrary profile in P(X)N, and
x >-y >-z then everyone strictly prefers x to y and everyone strictly prefers y to z.
Therefore, everyone strictly prefers x to z, by transitivity of individual preference. Then
x >-z, and hence f(r) is quasitransitive. ·
The Pareto extension ranking establishes that the Arrow conditions are consistent
provided that we substitute quasitransitivity off(p) for transitivity. But for domains
within which each individual preference is a linear ordering, any social welfare
function satisfying this new set of conditions either gives each individual the power
to prevent an alternative y from socially ranking above any x, simply by declaring
a strict preference for x over y, or else there are one or more individuals who have
no influence onf at any profile. To prove this we need some new definitions. Social
welfare functionf on domain P gives coalition J C N veto power, if for all p C P, and
all x,y EX, x iy for all i EJ implies x y. In words, y cannot rank above x in the
social preference relation if every member of a coalition with veto power expresses a
strict preference for x over y. We say that individual i has veto power if coalition {i}
has veto power, and we sometimes say that individual i is a vetoer in that case. What
if persons 1 and 2 both have veto power? If x >-ly and y >-2x then we must have
x -y. Unless both individuals declare at least a weak preference for x over y, we can't
have x >-y. In fact, Arrow's conditions, with quasitransitivity in place of transitivity,
imply that there is a set of individuals each with veto power, and that set is a decisive
coalition: the coalition J C N is called an oligarchy for social welfare functionf if J
is decisive forf and every member of J has veto power. In that case, we say thatf is
oligarchical.
Gibbard (1969), Guha (1972), and Mas-Colell and Sonnenschein (1972) indepen-
dently proved the following analog of Arrow's Theorem.
Oligarchy theorem. Suppose that X has at least three members and the domain
60 D.E. Campbell and .S. Kelly

p C p(X) N has the chain property. Iff is quasitransitive-valuedand satisfies IIA and
the Pareto criterion, then f is oligarchical.
Proof: We can use part (3) of the ultrafilter lemma because the proof of (3) uses only
transitivity of the strict part off(p), not transitivity off(p) itself. It does appeal to
the contagion lemma to establish that every coalition J C N that is decisive for some
pair (x,y) of distinct alternatives is decisive for every pair of alternatives. However,
whenf is quasitransitive-valued, this follows in part from the fact that Pareto implies
nonimposition. In addition, the Pareto criterion allows us to replace y f(q)z and
z f (q)x in steps (a) and (b) in the proof of the contagion lemma with y f/ (q)z and
Z >f (q) x, and that in turn allows us to use transitivity of >f (r) instead of transitivity of
tf(r) in the proof.
Now, N is decisive by Pareto, so the collection U of decisive coalitions is not empty.
It is finite, because N is finite. Because eachf(p) is quasitransitive, the intersection
of two decisive coalitions is decisive by our adaptation of part (3) of the proof of
the ultrafilter lemma. Because U is finite, the intersection property implies that the
intersection of all members of U is decisive. Let J denote that coalition. We show
thatf is oligarchical by proving that every member of J has veto power.
Suppose that j belongs to J, but individual j does not have veto power. Then there
is a profile p and two alternatives x and y such that x >-Py but y Zf(p) X. Let {x,y,z}
be a free triple. Choose any profile r such that r I{x,y} =p {x,y} and x >- z >-y, with
x - z and y >- z for all i X j. Then y >f(,) x by IIA, and x >f(r)z by Pareto. Therefore,
y >Sf () z, by quasitransitivity off(r). Note that everyone but individual j strictly prefers
y to z at r. The argument of the second paragraph of the proof of part (3) of the
ultrafilter lemma shows that this implies that N \ {j} is a decisive coalition.
We have a contradiction: individual j belongs to the intersection of all decisive
coalitions, and N \ {j} is decisive. We have to drop the supposition that J contains an
individual without veto power. Therefore, f is oligarchical. l
This theorem presents us with an unfortunate dilemma: because x -y must hold if
x -iy and y >-ix for two members i and j of the oligarchy, we will typically have
lots of social indifference if the oligarchy is large. If we don't want a lot of social
indifference between very different alternatives, we can employ a rule with an oligarchy
that has very few members. But if the oligarchy is small and individuals are not
indifferent between distinct alternatives, then the preferences of many citizens will
not be consulted in determining the social ranking at those profiles: if everyone
in the oligarchy strictly prefers x to y, then the social preference relation has x
strictly preferred to y by decisiveness. If every member of the oligarchy strictly
prefers y to x, then y is strictly preferred to x socially. The only other possibility
has someone in the oligarchy strictly preferring x to y and another strictly preferring
y to x, in which case x and y are socially indifferent by veto power. In other words,
if there is no individual indifference then the social ranking of an arbitrary pair of
alternatives is determined without consulting the preferences of non-members of the
oligarchy. Moreover, a consequence of Theorem 1 in Fountain and Suzumura (1982)
Ch. 1: Impossibility Theorems in the Arrovian Framework 61

is that the trade-off dilemma is inevitable even if we replace the Pareto criterion
with strict nonimposition, which requires (for arbitrary but distinct x and y) that
there be some profile p at which x ranks strictly above y inf(p). Schwartz (2001)
replaces Arrow's transitivity requirement with a restriction on the length of a
e
sequence x' -f(p) 2 f(p) 3 ~f(p) *~'' x, and proves that this limit, along with
,(p)
the other Arrow conditions, implies that there will be profiles p at which every pair
of alternatives belongs to some cycle x1 >-f(p) x2 f(p) ... f(p) XT f(p) x 1.
If there is an oligarchy, then of course every member of the oligarchy will have
veto power, and thus N \ {i} is not decisive for any pair, for any individual i in the
oligarchy. What if we are willing to relax the Pareto criterion? No matter how we do
this, there will still be some individual i with substantial power, in the sense that the
coalition N \ {i} is not decisive over more than half of the pairs of alternatives in X.
Barrett, Pattanaik and Salles (1990, 1992) establish counterparts to this claim for
fuzzy aggregation rules and for fuzzy individual preferences, respectively. We shouldn't
actually measure the power of a coalition by counting the number of pairs for which
that coalition is decisive because that can be misleading. If x - y for all profiles, then
every coalition is decisive for (x,y). We say that coalition J is significantly decisive
for (x,y) if it is decisive for that pair, and y t x holds at some profile.
The quasitransitivity trade-off theorem. Suppose that X has at least three members
and P C L(X)N has the free triple property. Iff is quasitransitive-valuedand satisfies
[IA, then there is some individual i such that N \ {i} fails to be significantly decisive
for at least half of the pairsfrom X [Campbell and Kelly (1998)].
The bound of one half is tight, as we now demonstrate.
Example 5.3. X = {xl,X 2,... ,Xm} where m < n. Define f:
(1) If h < i, then Xh xi for every profile p.
(2) Set Xh >- Xi if and only if h <i and there is some J C N such that Xh >-j xi for all
j J, and IJI > n-(i-h).
IIA is certainly satisfied, andf(p) is quasitransitive for all p, as we now demonstrate.
Suppose xh -xi and xi >-Xk; then h < i < k. At least n - (i - h) persons j have xh j Xi,
and at least n- (k-i) persons j have xi >-j Xk. Then the number of individuals j
for whom h -j xi >-j xk is at least n - [(i - h) + (k- i)] = n - (k - h). Therefore, by
quasitransitivity of individual preference, there are at least n - (k - h) persons j with
xh >j Xk, and thus h >- Xk by definition off. Vacuously, every individual has veto
power for every pair (xh,xi) such that h < i. No individual has veto power for any
pair (xh, xi) such that h > i. Therefore, each individual has veto power over exactly half
of the pairs of alternatives. Finally, for arbitrary i E N, coalition N \ i} is significantly
decisive for exactly half of the pairs, as we now show. Suppose i >h. Then xh -xi
if xi -i h for all i e N. But for arbitrary i, if every j in N\ {i} has h >-jxi, then
IN \ {i}l = n - > n - (i - h), and hence x >-xi. ·
The binary relation t is acyclic if for every positive integer T > 1 and every choice
of T alternatives, x l , x 2,... ,x r in X, if x 1 >x 2 >. ..- >xT then xT •x I. Acyclicity is
62 D.E. Campbell and JS. Kelly

a necessary and sufficient condition for the existence of maximal elements from finite
subsets of X. See von Neumann and Morgenstern (1944, p. 597) and Sen (1970a,
p. 16). Note that quasitransitivity implies acyclicity. A social choice rulef is acyclic-
valued if f(p) is acyclic for all p in the domain off. What happens if we retain
Arrow's other conditions but further weaken the condition of transitive-valuedness, and
require only acyclic-valuedness of the social ranking? (We mean the version of Arrow's
theorem that assumes a full domain, not one based on the free triple assumption. The
test for acyclicity applies to all T > 3, not just to T = 3 - i.e., not just to triples). We
first observe that with each f(p) merely required to be acyclic, it is possible that the
intersection of the decisive coalitions will be empty, even if Arrow's other axioms are
satisfied:
Example 5.4. Assume that IX I < n, which means that there are more individuals than
alternatives, and a full domain. Let x be preferred to y if and only if either n or
n - 1 individuals strictly prefer x to y. This rule satisfies Pareto and IIA and is acyclic-
valued. All coalitions with exactly n- 1 members are decisive, and the intersection of
those coalitions is empty. And, of course, no individual is a vetoer. I
This example illustrates that results will depend on the number of alternatives relative
to the number of individuals: we will see that if X I > n, the intersection of all decisive
sets will be non-empty and constitute a minimal decisive set. However, f need not be
oligarchical because not all individuals in the minimal decisive coalition have to be
vetoers.
Example 5.5. x>y if and only if x>-ly and x>-iy for at least one il 1. The
intersection of the decisive sets is { 1}, and 1 has veto power, but { 1} is not an oligarchy
because it is not decisive; person 1 needs the support of one other person. No proper
superset of {1} is an oligarchy because person 1 is the only individual with veto
power. X
The set {1} in Example 5.5 is a collegium, which is a non-empty intersection of all
the decisive coalitions.
The Brown-Banks acyclicity theorem. Suppose that X has at least three members,
and there are at least as many alternatives as individuals. Iff is an acyclic-valued
social welfare function on a fJidl domain and it satisfies the Pareto criterion then it
has a collegium.
Brown (1975) first conjectured and proved this result, but with IIA added to the
hypothesis. Banks (1995) pointed out that Brown's proof does not depend on IIA,
because decisiveness itself embodies a lot of independence. Note that the Borda rule
meets all of the conditions of the Brown-Banks Theorem, except perhaps the one
relating m = IXI to n= IN[. Suppose that all but one individual ranks y in last place
and x in second last place at p. If the remaining person ranks y first and x last, then the
Borda score for x will be n - 1 and the Borda score for y will be m - 1. (See Example
7.1 below, p. 70.) In that case, the Borda rule yields y t x if and only if m > n. Because
Ch. 1: Impossibility Theorems in the Arrovian Framework 63

all but one individual ranks x above y, we conclude that N is the only decisive coalition
when m > n. This agrees with the implication of the theorem.
Even where n =m and a collegium exists, it may be that no individual has veto
power:
Example 5.6. X={x,y,z}, n=3, and P=L(X)N. The social rankings on {x,y} and
{y,z} are determined by simple majority voting, while the social ranking has x strictly
above z (resp., z strictly above x) if and only if every individual strictly prefers x to z
(resp., z to x). This rule satisfies Pareto and IIA and yields an acyclic ranking at all
profiles. The only decisive set is N (which then is the collegium); no one is a vetoer
on all pairs of alternatives. As we shall soon see, this is due to the fact that the social
choice rule is not neutral in its treatment of alternatives. ·
In the case of Example 5.6, while no individual is a vetoer on all pairs of alternatives,
every individual is a vetoer on two pairs, (x,z) and (z,x), while all coalitions of two
individuals have veto power (in fact are decisive) over all pairs of alternatives. Much
research on acyclic social choice either identifies large collections of pairs on which
at least one individual is a vetoer, or finds small coalitions (but with more than one
individual) that have veto power over all pairs.
Most analyses of the existence of coalition veto power have assumed some degree of
neutrality as suggested by Example 5.6. A rule satisfies NIM (neutrality, independence,
and monotonicity) if and only if for all x,y,z,w in X, whenever profiles p and q satisfy
the condition that for all i,

(x >-P y implies z t>q w) and (w >-q z implies y >-P x),

then x -f(p)y implies w f(q)Z. Blau and Deb (1977) proved the following:
The acyclicity theorem for NIM rules. Assume that X has at least three members.
Let t be an integer no greaterthan IXI and let {N 1, N2, ... , Nt} be any partition of N
into disjoint non-empty coalitions. Iff is acyclic-valued on P = L(X)N andf satisfies
NIM, then at least one of the Ni has veto power over all pairs of alternatives.
In particular, if X is finite and IXI =m then some coalition as small as [II,] individuals
must have such veto power, where I[ Imis the largest integer not exceeding . It should
be noticed that the acyclicity theorem for NIM rules establishes that many (small)
coalitions have veto power since N can be partitioned in many different ways [Kelsey
(1985)]. In Example 5.5, any of the n - 1 coalitions C with 1 C and IC = 2 will have
veto power (in fact will be decisive).
Kelsey's results are extended and sharpened in Le Breton and Truchon (1995). They
give the size of the smallest coalitions that must have veto power. Moreover, they show
that if that minimum is achieved by some rule, then that rule must give any larger
coalition veto power over all pairs of alternatives.
For how many pairs might a single individual be a vetoer? In an important early
paper, Blair and Pollak (1982) showed that there is an individual who has veto power
64 D.E. Campbell and JS. Kelly

over at least (m - n + 1)(m - 1) pairs if m > n > 2 and m > 4, and f is acyclic-valued.
Le Breton and Truchon (1995) give a simple proof of this theorem, based on a lemma
in Ferejohn and Fishbumrn (1979). For the case of rules satisfying NIM, Kelsey (1985)
extends the Blair-Pollak analysis to veto by groups.
We conclude this section with a brief review of three results that assume positive
responsiveness and Pareto decisiveness. Strong positive responsiveness requires that
x f(p y holds if x f(r)y and profile p is the same as r except that some i for
whom x -'y has x -py or some i for whom y >- x has x tpy [May (1952)]. Pareto
decisiveness requires either x -f (p)y or y >-f(p) x at arbitrary profile p if x >-py for all
i. Strong positive responsiveness tends to eliminate indifference in the social ranking.
When there is no indifference at all, transitivity, quasitransitivity, and acyclicity are
equivalent. That observation is not meant to be taken as a starting point for a proof of
any of the next three theorems, but rather an attempt to take some of the mystery out
of the results. The three theorems concern the presence of a quasi-dictator: individual i
is a quasi-dictator if he has veto power and i,j} is decisive for eachj # i.
Strong positive responsiveness is difficult to defend, but the following results
are interesting: Mas-Colell and Sonnenschein (1972) show that f must have a
quasi-dictator if it is acyclic-valued and satisfies IIA, Pareto, and strong positive
responsiveness. Fountain and Suzumura (1982) obtain a partition theorem under the
Mas-Colell and Sonnenschein assumptions, with Pareto decisiveness in place of the
Pareto criterion. They prove that there is a partition of X such that: (l)f I Y is quasi-
dictatorial or inversely quasi-dictatorial for each component Y of the partition; and
(2) if x and y belong to different components thenf {x,y} is imposed. Nagahisa (1991)
generalizes both the Mas-Colell and Sonnenschein and the Fountain and Suzumura
results and also proves a new one: if X is a separable and connected T space,
then for any product set domain with the free triple property, there exists an acyclic-
valued social welfare function satisfying IIA, Pareto decisiveness, and strong positive
responsiveness if and only if X is homeomorphic to an interval in the real line.
(Nagahisa's proof only uses profiles of continuous individual orderings in the domain.
The other two papers assume a full domain.)

6. Relaxing the domain condition

The proof that Arrow's conditions are incompatible does not depend on the domain
of a social welfare function being all of P(X) N, or even all of L(X)N. The free triple
property suffices, and even that was weakened to the chain property by Kalai, Muller
and Satterthwaite (1979). A quite different weakening of the free triple condition that
is still sufficient for an impossibility theorem is given in Kelly (1994a). Fishburn
and Kelly (1997) demonstrate that there is substantial scope for additional domain
reductions. Redekop (1991) shows how robust Arrow's theorem is from a topological
perspective. Domains on which Arrow's conditions are consistent are extremely
small, and Redekop (1993) demonstrates that this remains true for domains in which
Ch. 1: Impossibility Theorems in the Arrovian Framework 65

individual preference is characterized by a single parameter. This work is discussed

by Le Breton and Weymark (2002a, Chapter 16 in Volume 2 of this Handbook). For a
general discussion of domain conditions, see Gaertner (2001, 2002, Chapter 3 in this
Volume). Here is a simple example of a domain on which the Arrow conditions (other
than the chain property) are consistent:
Example 6.1. P is the set of all p C L(X)N such that there exist i and j E N with
p(i) =p(j) . There are no free triples. The Pareto criterion will be satisfied vacuously
by anyf. Now, choose a fixed Q E L(X) and setf(p) = Q for all p in the domain. IIA
and non-dictatorship are satisfied by f, which is transitive-valued. ·
Surprisingly, there is a domain lying between L(X)N and P(X)N on which the Arrow
conditions (other than the chain property) are consistent, as the next example, due to
Bordes and Le Breton (1990), demonstrates.
Example 6.2. Let O be the null ordering, for which x y for all x and y. Let
P = (L(X) U {O})N . There are no free triples, so there can be no chains of free triples.
Now definef with domain P: Ifp E L(X)N thenf(p) =p(l), but if p(i) =O for some
i N, thenf(p) =p(2). This rule satisfies Pareto and yields transitive social preference.
It is non-dictatorial and satisfies IIA. To establish IIA, suppose p {x,y} = r {x,y). Then
x Py for some i implies both x~ y and p(i)=O= r(i), in which case f(p) I{x,y}
and f(r)l {x,y} are the same as p(2)l {x,y}, and thus f(p) {x,y} =f(r) {x,y}. If
Pl {x,y} =rl {x,y} and x Py does not hold for any i then p and r both belong to
L(X)N, in which case f(p) {x,y} =f(r) {x,y} because f(p) I{x,y} andf(r)I{x,y are
both identical to p(l) I x,y}. See Kelly (1994b) for more details about the possible
variety of Bordes-Le Breton type examples. ·
Most of the work on the role of domain conditions focuses on majority rule.
Transitivity of the majority ranking calls for certain patterns of preferences. Consider
Example 5.1 (the paradox of voting) once again: each alternative is first in someone's
preference ordering, second in someone else's, and last in another person's preference.
Value restriction was introduced in Sen (1966) to eliminate such profiles. Sen gave a
general definition and theorem, but for expositional purposes we will confine attention
to the case of linear preferences. Profile p satisfies value restriction if for every three-
alternative subset {x,y,z} of X there is one member of {x,y,z} that is not below the other
two in any p(i), or is not above the other two in any p(i), or is not in between the other
two in any p(i). Sen showed that when individual preference is linear and the number
of individuals is odd, value restriction implies that majority rule is transitive. Sen also
proved that if individual preferences are linear, then for any n, value restriction implies
that the majority rule ranking is quasitransitive. Value restriction gives us a domain
on which the Arrow conditions (excluding the chain property) are consistent - and
compatible with many other criteria such as neutrality, strong positive responsiveness,
and symmetric treatment of individuals. The family P of all value restricted profiles
does not have any free triples (and hence does not have the chain property), because,
for arbitrary x, y and z, it excludes profiles for which the three orderings of {x,y,z} of
66 D.E. Campbell and JS. Kelly

Example 5.1 are embedded, and also profiles in which the three orders of that example
are turned upside down.
Sen's majority decision theorem. If n is odd andp is value restricted then majority
rule is transitive at p.
Proof: Let t be the simple majority relation determined by p. If a >-b - c then at
least (n+1)/2 individuals prefer a to b and at least (n+ 1)/2 prefer b to c. Thus,
there must be at least one individual i such that a -i b -i c. Now, suppose there is
a cycle, x -y >-z - x. Then there must be at least one individual i with x >-iy -i z, at
least one j with y >-j z -jx and at least one individual k with z -k x >-k y. Therefore,
x >-y >-z s-x implies that each member of x,y,z} ranks above the other two in at
least one individual's preference ordering, each member of {x,y,z} ranks below the
other two in at least one individual's preference ordering, and each member of {x,y,z}
ranks between the other two in at least one individual preference ordering. Therefore,
profile p is not value restricted. []
There is a partial converse to this theorem. If n =3 or n >4 and the domain P is
equal to SN for some set S of linear orderings on X, and there is a profile in S N
that is not value restricted, then there is a majority rule cycle for some profile in that
domain. To prove this, assume that n >4, and let k be the smallest integer that is
not less than n. (Example 5.1 takes care of the case n=3, and it is easy to show
that x >-y z >- x cannot hold at any profile in L(X)( 1' 2' 3'4 } , or in L(X) 1'2 }.) If some
profile r in SN does not satisfy value restriction, then there are three alternatives x,
y and z and a profile r such that {r(i) I {x,y,z}: i N} contains the three orderings
of the profile displayed in Example 5.1. Because the domain is a product set, we can
define the profile p by assigning the ordering of column 1 of the profile in Example 5.1
to exactly k of the individuals in N, assign the second column to another k members
of N, and let the remaining n - 2k members of N have the ordering from column 3
of Example 5.1. One can show that n <k < n, and those inequalities imply that a
coalition with 2k members or n-k members is a majority. Then we have a majority
voting cycle x - y >-z S-x at profile p. See Sen (1966, 1970a) for the general treatment.
Cantillon and Rangel (2001) use the geometric tools devised by Saari (1994) to analyze
majority rule and its relatives in a new and insightful way.
A tournament is simply a binary relation - such that x •y implies either x >-y
or y x, but not both. The statement x >-y represents the defeat of alternative y by
alternative x. Even when cycles are present, the mathematics of tournaments can be
used to select an outcome ("winner") in a systematic way. For instance, graph theoretic
techniques can be applied to the tournament for which x >-y means that some majority
prefers x to y. There is a wide variety of solution techniques available - including
Markov methods - and most are analyzed in Laslier (1997), which reports original
work by the author, much of it in collaboration with Gilbert Laffond, Jean Lain6, and
Michel Le Breton.
Levchenkov (1999a,b) proposes a new tournament solution concept that initially
Ch. 1: Impossibility Theorems in the Arrovian Framework 67

assigns a score to alternative x that reflects the number of alternatives that x defeats in
a majority comparison. Then an alternative's score is adjusted to reflect the scores
of the alternatives that it defeats. When this adjustment process reaches a steady
state, the scores are used to compute the solution. Levchenkov proves that, under his
assumptions, his is the only method with this consistency property.
Without some common thread such as value restriction running through individual
preferences, there does not even exist a super majority rule that guarantees the
existence of a winner in all cases, as we demonstrate with the next example.
Example 6.3. There are n individuals and n alternatives, x l,x 2 ,... ,xn. Consider the
profile

1 2 ... i ... n

XI X2 Xi XI

X2 X3 Xi+ Xl

Xn-1 Xn Xi-2 Xn 2
Xn XI il X,n 1

Note that n- 1 individuals prefer xl to x 2, n- 1 individuals prefer x2 to X3, and in

general n - 1 persons prefer xi to xi + 1 for 1 < i < n - 1, and n - 1 individuals prefer x,
to xl. For any given fraction 2, however small, we can find a value of n such that each
alternative fails to get even the fraction ;i of the votes in a contest with at least one
other alternative. ·
The profile of Example 6.3 exhibits a high degree of diversity of individual preference:
each alternative is first in someone's ordering and last in someone else's ordering. In
fact, each alternative is in the jth position of someone's ordering, for arbitrary j such
that 1 j < n. Sen's majority decision theorem proves that value restriction imposes
enough coherence of individual preferences to guarantee the existence of a majority
winner.
Welfare economics frequently requires a social ranking of alternatives that belong
to some Euclidean space. Thus, we now consider the question of imposing restrictions
on preferences to assure a majority winner when X is a subset of -dimensional
Euclidean space. In that setting Rubinstein (1979) showed that without restrictions,
virtually all profiles fail to yield a majority winner. He assumed that X is a compact and
convex set with a non-empty interior, and that individual preferences are continuous
orderings onX. The seminal paper is Plott (1967). Moreover, when there is no majority
winner, between any two distinct members x andy of X one can find a finite number of
additional alternatives x 1, x 2 ,... ,xk in X such that x >-xl >- x 2 .. >- xk >- y, where
>- denotes the strict simple majority rule relation [McKelvey (1979)].
68 D.E. Campbell and JS. Kelly

Consider an arbitrary set X, not necessarily a subset of Euclidean space. Suppose

that at profile p each individual i N has a bliss point bi. In addition, suppose that
we can map the members of X into the real line in such a way that for each i and
each x and y in X on the same side of bi, individual i's preference p(i) has x strictly
preferred to y if and only if x is closer to bi than y. Note that these two suppositions -
called the single-peaked preferences assumption - restrict both the distribution of
bliss points and the nature of individual preferences. When working with a set X that
is not given any structure, the single-peakedness assumption is generalized, and one
merely requires that every triple {x,y,z} from X contains an alternative that is not
below the other two alternatives in the preference ordering of any individual.
Single-peakedness is a special case of Sen's value restriction. To see why, begin by
locating the members of X on the real line and representing individual preferences by
means of a utility function. Under single peakedness, the graph of the utility function
will be A-shaped, possibly with ties at the top or with the right or left arm missing.
Note that if we discard all but three alternatives x, y and z, we will still have the
A shape. Therefore, whichever of the three alternatives is between the other two on
the real line will not be at the bottom for any of the other individual orderings restricted
to {x,y,z}, and hence value restriction holds. Although value restriction yields the more
general existence theorem, it is uery restrictive in Euclidean space of dimension two
or higher [Kramer (1973)].
If the number of individuals is odd, the median voter is the person whose bliss
point bmed sits in the middle of the array of bliss points on the real line. In an important
and influential paper, Black (1948) showed that with single-peaked preferences bed
is a majority winner: if n is odd, then the median voter is person n + 1/2. If x is to
the left (resp., right) of b,,ed then over half of the voters will have their bliss points to
the right (resp., left) of x. If x is to the left (resp., right) of bmed then the median voter
and all those whose bliss point is to the right (resp., left) of bmed will prefer bmed to x.
(If n is even there will be two median voters and perhaps two majority winners.)
When X is a compact and convex subset of E e (-dimensional Euclidean space)
Black's approach does not work for > 1. Assuming that individual preferences are
continuous and convex orderings on X, Greenberg (1979) showed that even super
majority rule will not precipitate an undefeated alternative in X unless the fraction
of the voters required for a majority is at least l/((+ 1).
For arbitrary , a preference ordering on ES is said to be Euclidean if x is preferred
to y if and only if x is closer to the bliss point than y as measured by Euclidean
distance. To get the existence of a majority winner in compact subsets of f-dimensional
Euclidean space one needs a substantial restriction on the distribution of bliss points
and on the form of individual preferences. Grandmont (1978) proved an existence
theorem for the family of domains that properly includes the class of profiles of
Euclidean preferences. Grandmont's condition on the form of individual preference,
while quite restrictive, admits a wide variety of profiles. Caplin and Nalebuff (1991)
obtain a majority winner existence theorem under the Grandmont condition by
assuming only that the probability density of bliss points has a very general concavity
Ch. 1: Impossibility Theorems in the Arrovian Framework 69

property. This represents a big improvement on the other existence theorems, including
Arrow (1969), Davis, De Groot and Hinich (1972), Grandmont (1978), Caplin and
Nalebuff (1988), and Tullock (1967), the seminal paper in this series. Caplin and
Nalebuff (1991) prove that for any compact subset X of e-dimensional Euclidean space,
a lower bound of 1- [£l(C + l)]t on the proportion of voters needed for a majority
guarantees the existence of an undefeated alternative. As increases, the bound
increases monotonically to 1- lie, which is almost 64%. The undefeated alternative
is the mean voter's bliss point. Ma and Weiss (1995) demonstrate that this outcome
is not always invariant to transformations of the parameters of the individual utility
functions, even when the transformations do not change the individual's underlying
preference ordering.
It is possible to reduce the domain to a single profile without being able to break
away from dictatorship, although the single-profile impossibility theorems substitute
two new criteria for Arrow's domain assumption. Without any new conditions we are
in the clear, if the domain P = {p} is a singleton: we can let f(p) be any member of
P(X) consistent with the Pareto criterion. IIA will be satisfied vacuously. (We won't
worry about dictatorship if p(l) =p(2) = ... =p(n), in which case Pareto implies that
everyone is a dictator.) However, the addition of a neutrality condition and a preference
diversity assumption to Arrow's list can precipitate an impossibility theorem for many
of these singleton domains. For any domain 2, we say thatf on P satisfies neutrality
if for arbitrary x and y in X and arbitrary profile p E 2, if x f(p) y then we must have
w >-f(q) z for any w and z and any q E P such that, for arbitrary i N, w qz holds if
and only if x ,Py, and z tq w holds if and only if y Px. Note that this includes the
case q =p, and hence we can test singleton domains for neutrality.
Suppose once again that P = {p} and that f is transitive-valued and satisfies the
Pareto criterion and neutrality. Consider the proof of Arrow's Theorem after the
ultrafilter lemma has been obtained. No part of the former requires P to have more
than one member. But can we prove that the collection U of decisive sets forf is an
ultrafilter? We have N E U by Pareto, and this gives us (1), which requires N to be
decisive (but not the empty set). By definition, any superset of a decisive coalition is
decisive, so we have (2). Neutrality implies that if coalition H is decisive for a pair then
it is decisive for all pairs. Therefore, we don't need the contagion lemma. Property (3)
requires the intersection of two decisive sets to be decisive. Note that the proof of (3)
in the ultrafilter lemma can be accomplished with a single profile, provided that there
is sufficient diversity of individual preference. Property (4) requires the complement
of a coalition to be decisive if the coalition itself is not decisive. The proof of (4)
employs three profiles, but we can get away with a single profile if we use additional
alternatives and the neutrality property. Hence, if we add to the Arrow requirements,
not only neutrality, but also a condition ensuring that p(l), p(2 ), p(3 ), etc., are so
related that we can establish (3) and (4), we can show thatf is dictatorial. The first
single-profile impossibility theorems are due to Parks (1976), Kemp and Ng (1976)
and Hammond (1976). Pollak (1979), Roberts (1980) and Rubinstein (1984) are also
important contributions to this literature.
70 D.E. Campbell and S. Kelly

The single profile theorem of Dutta and Sen (1996) concerns the properties of a
ranking of opportunity sets obtained by aggregating one ranking based on the highest
level of attainable utility and another based on the size of an opportunity set. They
obtain a dictatorship result: one of the criteria will be ignored. Section 8 contains a
discussion of a very recent impossibility theorem, Kaplow and Shavell (2001), that
employs only one profile and a non-Arrovian framework.

7. Relaxing independence of irrelevant alternatives

Weak IIA requires x (p)y if x ¥/(q) Y andp x,y} = q I {x,y}. Baigent (1987) showed
that if XI >4 and weak IIA is substituted for IIA, while retaining the other Arrow
conditions, then some individual has veto power. See also Campbell and Kelly (2000b),
which corrects a defect in Baigent's proof and extends the theorem to economic
environments.
Rather than assuming a modified version of IIA, we can look at an arbitrary
transitive-valued social welfare function and count the number of pairs of alternatives
{x,y} for which p {x,y} q {x,y} implies f(p) l {x,y} =f(q) {x,y} for all profiles p
and q, and f I x,y} is a non-dictatorial social welfare function satisfying the Pareto
criterion. Powers (2001) shows that this number cannot exceed 1/ + 1/(m-1),
where m is the cardinality of X. Powers assumes that m > 3, n > 3, and the domain
off has the free triple property. His proof is based on the one in Campbell and
Kelly (1995a), which establishes that if f is transitive-valued and satisfies Pareto
then f I x,y} satisfies non-dictatorship and IIA for at most the fraction 2/m of the
pairs x,y}.
For the rest of this section, there will be pairs {x,y} such that f(p)l x,y} is
conditional on p IY for some proper superset of {x,y}, and not just on p I x,y}. If we
simply drop IIA from the list of conditions, the other Arrow conditions are easily seen
to be consistent - and compatible with many other criteria such as neutrality, strong
positive responsiveness, and symmetric treatment of individuals. Here is a standard
example of a rule that violates IIA but has many desirable properties.
Example 7.1. Global Borda rule [Borda 1781)]: Assume that X is finite. For
simplicity, we define this rule on L(X)", so each of the n individuals has a linear
ordering over X. For each such individual, allocate m - 1 points to the alternative that
is at the top of his preference order, m -2 points for the alternative in second place,
and in general, m -j points for the alternative injth position in his preference ranking.
Then the social ordering is constructed by ranking x over y if and only if x's total
score (added over all n individuals) is greater than y's total. This gives us a transitive
social ordering because the ordering "greater than" on the real numbers is transitive. It
satisfies the Pareto criterion because if every individual strictly prefers x to y, then the
total score for x will exceed y's total score by at least n. The Borda rule is obviously
not dictatorial. [
Ch. 1: Impossibility Theorems in the Arrovian Framework 71

The Borda rule is given an axiomatic characterization in Young (1974). IIA is not
one of the axioms, of course. Debord (1992) gives an axiomatic characterization of a
generalized Borda-type rule that is used to select k > 1 alternatives from X. The Borda
rule is one member of the family of positional rules, a family whose properties are
thoroughly analyzed in Saari (1994, 1996). See also the discussion of positional rules
in Pattanaik (2002, Chapter 7 in this Volume), Fishburn and Brams (2002, Chapter 4
in this Volume) and Saari (2002, Chapter 25 in Volume 2 of this Handbook).
The Borda ranking of x and y can differ in two situations even though the individual
orderings of x and y are the same in those two situations. Suppose n = 5 and m = 3.
Consider a simple profile:

1 2 3 4 5

x x x y y
y y y x x

The total score for x is 8 = 2 + 2 +2+ 1+ 1 and for y it is 7 = 1 + 1 + 1 + 2 + 2. Therefore

x ranks above y in the social ordering determined by the Borda rule. But consider a
different situation, with different preference orderings for persons 4 and 5:

1 2 3 4 5

x x x y y
y y y z z

The total score for x is 6 and the total score for y is 7. The social ordering of x and y
has reversed, even though each individual has the same ordering of x and y in this new
configuration as he did in the first situation. This is clearly a violation of IIA. Saari
(1994) has many more examples of this sort; many of his examples and theorems are
very surprising. See also Saari (1989, 2000a,b, 2002, Chapter 25 in Volume 2 of this
Handbook).
It has long been known that abandoning Arrow's (1951) IIA condition opens the
door to rules that are far from dictatorial. The Borda rule is a good example. With the
exception of Fleurbaey and Maniquet (2001) and the material discussed in the rest of
this section, the only departures from IIA in the welfarist literature - in which social
choice depends only on individual preferences for outcomes - are rules for which the
social ranking of x and y depends on the individual orderings over all of X, as in the
case of the global Borda rule. Identification of the entire feasible set before rejecting
even a single alternative is far too costly, however - we would not even expect this of
72 D.E. Campbell and J.S. Kelly

a single consumer choosing from a budget set in the absence of uncertainty. Bordes
and Tideman (1991, p. 184) put it more forcefully: the set of all candidates "is often
not really defined". We can still allow the social ordering of x and y to depend on at
least a few additional alternatives. But Fishburn (1973, p. 6) asks, on which additional
alternatives should we condition the social ordering of x and y? If for a given profile,
the social ordering of x and y depends on individual preferences on some superset
of {x,y}, then the disagreement among individuals that is supposed to be resolved by
the adoption of a social welfare function re-emerges in the form of conflict over the
selection of that superset. We can use the global Borda rule (Example 7.1) to show
how the social ordering of a pair of alternatives can depend on our choice of X. Let
the second table be our profile p. We wish to determine the social ordering of x and y.
If X = {x,y} then x ranks above y, but if X = {x,y,z} then y will rank above x.
Nevertheless, it is noteworthy that even a small departure from IIA allows the
construction of social choice rules that are far from dictatorial, and even allow majority
rule to play a key role - without spoiling transitivity or violating Pareto. Much of the
material in the rest of this section is from Campbell and Kelly (2000a).
A simple case is our earlier Example 3.4: select two alternatives v and w in advance.
If person 's ordering has v strictly preferred to w then the social ordering is the same as
person l's; otherwise the social ordering is the same as person 2's preference ordering.
Then to socially order x and y we only need to know the configuration of individual
preferences on the set {x,y,v,w}. This is a modest additional information requirement,
but it doesn't get us far enough from dictatorship, because no one other than individuals
1 or 2 has any influence on the social choice.
Now we define a family of rules - gteau rules - that can be substantially non-
dictatorial, in addition to satisfying Pareto and transitivity of the social ordering on
a full domain. Moreover, they constitute a minimal departure from IIA, in the sense
that the social ordering of an arbitrary pair only depends on one additional alternative
that is the same for all pairs. Assume X > 3. We define our general gteau rule f
by first choosing a distinguished element v in X, two individuals i andj, and a social
welfare functionf* that satisfies Pareto and IIA. (We could letf* be majority rule, for
example.) To define f(r) for arbitrary profile r in L(X) ' we let >- denote the strict
preference part off(r), and we let >-* be the strict preference part of f*(r). (For
expositional convenience, we assume that either x >-* y or y >-* x holds for any two
distinct members of X.) Let T(r), the top layer of f(r), be the set of alternatives
x in X such that x >-*v. B(r), the bottom layer off(r), is the set of alternatives x in X
such that v >-*x. To define f(r) we set

x - v >-y (and x >-y) if x belongs to T(r) and y belongs to B(r);

f(r) T(r) = r(i)l T(r) andf(r) B(r) = r(j)l B(r).

Clearly, f(r) is transitive and the Pareto criterion is satisfied. Note that will be non-
dictatorial iff* is, or if i andj are not the same.
Ch. 1: Impossibility Theorems in the Arrovian Framework 73

Iff* is majority rule then we put all the alternatives that defeat v by a majority in the
top layer T(r) and then order T(r) internally according to person i's preferences. All
the alternatives that are ranked below v by majority rule go in the bottom layer, B(r),
which is then internally ordered by individual j's preferences.
We can modify the definition so that individuals i andj have less power. For example,
we could choose two individuals h and k, and split T(r) into S(r) = {x E T(r): x >-h v}
and T(r) \S(r). Then we could order S(r) according to i's preferences and order
T(r) \ S(r) according to individual k's preferences. However, Arrow's Theorem places
restrictions on the social ordering of the top layer. Consider the subfamily of profiles r
for which v is at the bottom of each individual's preference ordering. Because f*
satisfies Pareto, T(r) =X \ {v} for any such profile. If the social ordering of x andy can
depend only on individual preferences restricted to {v,x,y} then IIA will hold within
this special family of profiles. Arrow's Theorem tells us that the ordering of T(r) within
this family is dictated by some individual i. Say that person i is this local dictator.
Now, let p be any profile for which everyone ranks x and y above v. Then x and y
belong to T(p). There is a profile r for which v ranks at the bottom of each r(h)
and r I{v,x,y} =p I{v,x,y}. Because the social ordering of x and y can depend only on
{v,x,y}, we must have x and y socially ordered according to p(i) at profile p. A similar
restriction applies to B(p).
The gaiteau family takes us a small step away from IIA; but as we have just seen,
Arrow's Theorem still gives some special power to some individual. In fact, even if
we modify the Arrovian framework by replacing IIA with the far milder requirement
that for any two alternatives x and y, at least one member of X is irrelevant to the
social ordering of x and y, if the domain is full we can prove that some individual will
have the power to prevail against unanimous opposition in some situations. The new
condition is called independence of some alternative (ISA): for any two alternatives
x and y there is a proper subset Y of X such that for any two profiles p and q in the
domain, if p Y = q lY then f(p) {x,y} =f(q) {x,y}.
Information restriction theorem [Campbell and Kelly (2000a)]. Suppose that X
has at least three members andf is a social welfare function with domain L(X)N or
P(X)N. Iff is transitive-valuedand satisfies the Pareto criterion and independence of
some alternatives, then there is an individual i c N, two distinct alternatives x and y,
and a profile r such that

x -i y,y -j x for allj N\ {i}, and x >-y.

Given a rule f, say that Y is sufficient for {x,y} if for any two profiles p and q
in the domain,f(p) I{x,y } =f(q) I{x,y} if p IY = q Y. (Y can be empty when f {x,y}
is constant - a possibility for some x and y in economic environments). Note that
ISA is equivalent to the following: for any two (distinct) x,y EX there is some
z EX such that X \ {z} is sufficient for {x,y}. However, there can be more than one
74 D.E. Campbell and AS. Kelly

sufficient set for {x,y}, and when the domain is full, the family of sufficient sets can
embody substantial restrictions on the possible departures from IIA, as the following
intersection principle shows. Note that it does not assume the Pareto criterion or any
type of transitivity property forf(p).
Intersection principle. Iff has a full domain, and Y and Z are each sufficient for
{x,y} then Y n Z is sufficient for x,y}.
Proof: Suppose that p I(Y n Z) = q I(Y n Z). Then there exists a profile r in the domain
such that r I Y =p Y and r IZ = q Z. We have f(p) l {x,y} =f(r) {x,y} because Y is
sufficient for {x,y}, and f(r) I{x,y} =f(q) {x,y} because Z is sufficient for {x,y}.
Therefore, f(p) {x,y} =f(q) I x,y}. l
If {x,y} is sufficient for {x,y} we say that the pair is self-sufficient. IIA can
be expressed by saying that {x,y} is self-sufficient for arbitrary x,y EX. We can
generalize IIA by specifying a family S{xy} of subsets of X that are sufficient for
arbitrary {x,y}. For finite X, the intersection principle implies that if for all x,y EX
the intersection of the members of S{,Xy equals {x,y} thenf satisfies IIA. This was first
proved by Blau (1971) for the special case S{xy} = {X \ {z}: z EX \ {x,y}}, although
Blau required f(p) (X \ {z}) =f(q) (X \ {z}), not just f(p) {x,y} =f(q) I{x,y}, if
p (X \ z}) = q (X \ z}). [Something very similar to the proof of the intersection
principle is a key step in the proof of the Kalai and Schmeidler (1977) impossibility
theorem for cardinalpreferences].
We conclude this section by reiterating that almost all of the social welfare functions
defined and discussed in the literature either satisfy IIA or violate both IIA and ISA.
There has been little research on the middle ground. One could argue that the set X
used to define the Borda rule (Example 7.1) is not the entire set of logically possible
alternatives, but is simply the current agenda - in other words, the set of feasible
alternatives. We could have X = {x,y,z} in one situation and X = {x,y} in another. Of
course, as we have already pointed out, the profile

1 2 3 4 5

x x x y y
y y y z z

leads to y >-x if we apply the Borda rule to X = {x,y,z}, but to x >-y if we apply
the Borda rule to X = {x,y}. However, for economic problems we can't use individual
preferences over the entire feasible set to socially order a particular pair of alternatives
because it is exceedingly costly, if not impossible, to identify the feasible set with
precision.
Ch. 1: Impossibility Theorems in the Arrovian Framework 75

8. Modifications of the Arrovian framework

Close examination of the proof of Arrow's theorem reveals that the full strength of
each condition is not employed in the proof. For instance, Wilson's Theorem allows
us to replace the Pareto criterion with the assumption that f is nonimposed and
f l{x,y} satisfies Pareto for some choice of distinct x and y from X. Transitivity
can be relaxed: Blair and Pollak (1979) and Blau (1979) show that dictatorship is
still implied if each f(p) is merely assumed to be a semi-order, which is between
quasitransitivity and full transitivity in strength. Both results are generalized by
Fountain and Suzumura (1982). In fact, Wilson (1975) implies that there is a range
of intermediate transitivity properties that lead to Arrow's conclusion in the presence
of the other axioms. Similar surgery can be performed on IIA and on the domain
assumption. (However, we can't expect to be able to operate successfully on all the
conditions simultaneously.) Instead of pursuing this theme - an inspection of the proof
in Section 3 would be more profitable - we turn now to a brief discussion of several
departures from the full Arrovian framework. We consider in order: social choice
correspondences, probabilistic social choice, fuzzy social choice, a theorem on the
conflict between Pareto and the desire to use non-utility information about individual
welfare, consensus functions, social choice functions that select themselves, and finally
infinite societies.
We could attempt to obtain a satisfactory social choice rule by abandoning the
requirement that every profile map into an ordering on X, and instead ask for a
selection of alternatives from an agenda Y C X as a function of individual preferences
over the alternatives in X. The selection rule is called a social choice correspondence.
But we now have to specify the domain F of agendas (or feasible sets), in addition
to the domain P of preference profiles. (F is a collection of nonempty subsets of X.)
For each profile p e 7P and each Y E F the social choice correspondence C specifies
a nonempty subset C(p, Y) of Y. We say that C is rationalized by the social welfare
function f if for all Y E F and p 7'P,

C(p, Y) = {x G Y : x tf(p) y for all y E Y}

Not every social choice correspondence can be rationalized by a transitive-valued

social welfare function, so the correspondence approach to social choice offers
hope that we can satisfy all of Arrow's conditions - after translating them to the
correspondence framework. Deb (2002, Chapter 18 in Volume 2 of this Handbook),
provides a thorough treatment of social choice correspondences.
IIA is converted to independence of infeasible alternatives (IIF), which requires
C(p, Y) = C(q, Y) whenever p I Y = q I Y. IIF applied to social choice correspondences is
easier to defend than IIA because the former is implied by incentive compatibility: if
C(p, Y) is the set of equilibrium outcomes for some mechanism, and p Y = q I Y then
the set of equilibria for q and Y must be identical to the set of equilibria for p and Y.
Therefore, C(q,Y)= C(p,y). See Campbell (1992b,c) for details.
76 D.E. Campbell and J.S. Kelly

The Pareto criterion is converted to Pareto optimality: we have y C(p, Y) if x >-i y

for all i N and some x E Y. And C is non-dictatorial if for all i N there is some
p E P and Y 5 such that there exist x and y in Y with x >-iY and y C(p, Y).
Transitivity-valuedness off is replaced by Arrow's choice axiom (ACA), which
requires, for arbitrary profile p EP, C(p,Y)= Y n C(p,Z) whenever YZ F, Y C Z,
and Y n C(p,Z) X 0. If C(p,Y) = {x C Y: y -i x for all i E N implies y Y}, the set of
Pareto optimal alternatives in Y at p, then C satisfies IIF, Pareto optimality, and non-
dictatorship (even when F is the collection of all nonempty and finite subsets of X and
P = P(X)N) but not ACA. The addition of ACA to the list of requirements precipitates
an impossibility theorem - the conditions are inconsistent [Sen (1969, 1970a)]. That's
because ACA implies the existence of a transitive-valued social welfare functionf that
rationalizes C [Arrow (1959)].
When F contains all two-element subsets of X one can distill from C what
Herzberger (1973) terms the base relationf(p), for arbitrary p P, by setting x bf(p)y
if and only if x C(p,{x,y}). However, an advantage of working with C itself is
that internal consistency conditions, connecting C(p,Y) to C(p,Z) for subsets and
supersets Z of Y, are often easier to motivate and evaluate than transitivity-type
restrictions onf(p). Moreover, when C is rationalized by a social welfare functionf,
the transitivity properties of the f(p) can be derived from the postulated consistency
conditions for C. For instance, beginning with Plott (1973), a number of papers require
independence of C(p, Y) from the path taken to that set via choices made from two-
element (and other) subsets of Y. Various strengths of this condition lead to a variety
of near-transitivity conditions, lying between quasitransitivity and full transitivity, that
still imply dictatorship in the presence of IIF and Pareto optimality, although some
of the theorems employ strong Pareto optimality along with a somewhat weaker
path independence requirement. (C satisfies strong Pareto optimality if y ~ C(p, Y)
whenever there exists an x in Y such that x iy for all i EN and x -iy for some
i EN.) For details see Blair, Bordes, Kelly and Suzumura (1976) and Bandyopadhyay
(1984-1986, 1990). In a different vein, Nermuth (1992) proves an impossibility
theorem for multistage aggregation in the abstract algebraic aggregation model of
Wilson (1975).
Economic models often deal only with feasible sets that have far more than two
members. Grether and Plott (1982) take a small step in this direction by dropping
the assumption that every nonempty and finite subset of X belongs to . They
assume that a positive integer m is given, and F includes all finite subsets of X
with m or m+ 1 members, but no set with less than m members. They prove an
impossibility theorem when C satisfies ACA, IIF, Pareto, and non-dictatorship on
P(X)N. Campbell and Kelly (1994b) use the Grether-Plott specification of F to prove
Wilson's Theorem and also the Trade-Off Theorem for social choice correspondences.
Campbell (1995) treats economic environments, and Campbell and Kelly (1995b)
prove a correspondence counterpart to the extreme trade-off theorem presented at the
end of Section 4.
Gibbard, Hylland and Weymark (1987) place an interesting restriction on F: there is
Ch. I: Impossibility Theorems in the Arrovian Framework 77

assumed to be a status quo alternative x° in X that belongs to every member of F. In

that case there is a non-dictatorial social choice correspondence on P(X)N satisfying
ACA, IIF, and Pareto. However, the Gibbard-Hylland-Weymark example gives one of
the individuals enormous power, as they acknowledge. They also prove that unequal
treatment of individuals is inevitable in their framework. See also Yanovskaya (1991).
Feasible sets in economic models are seldom discrete. By taking F to be the family
of compact subsets (with nonempty interior) of some f-dimensional Euclidean space,
Le Breton and Weymark (2002b) exhibit a family of social choice correspondences
satisfying ACA, IIF, Pareto, and non-dictatorship on two domains on which Arrow's
impossibility theorem goes through for social welfare functions: the domain of
profiles of Euclidean preferences, and the domain of profiles of monotonic and
analytic preferences. Their rules are based on utility representations of individual
preferences. It has long been recognized that transitive-valuedness, Pareto, and non-
dictatorship can be satisfied on virtually any domain P by assigning to each profile p
in P an n-tuple ul,u2,....,u, of real-valued functions on X such that ui is a
utility representation of p(i) for i = 1,2, .. ., n. Then for any real-valued function W
on n-dimensional Euclidean space, we can define f by setting x y if and only
if W(ul(x), U2 (),,..., Un(x))> W(Ul(y), U2(y),..., u,(y)). Obviously, f is transitive-
valued. If W is monotonically increasing in its arguments thenf will satisfy Pareto. If
W is symmetric in its arguments thenf will be non-dictatorial, and will, in fact, treat
individuals symmetrically. However, IIF will typically fail with this technique. Because
each X E F is compact and has a nonempty interior, for the Le Breton and Weymark
domains, two orderings that agree on X must be identical over the entire space,
and hence have the same utility representation. That implies that the correspondence
C(p,X)={x EX: x f(p)y for all y cX}, based on f defined above, satisfies IIF, the
correspondence version of IIA.
The fixed agenda approach has F= {X}. In that case we can write C(p) instead
of C(p,X). We interpret X as the feasible set, and we select some alternatives
in X as a function of the profile. IIF and ACA are satisfied vacuously when
F= {X}, but Hansson (1969) shows that an impossibility theorem is precipitated if
the following Hansson independence condition is satisfied in addition to Pareto and
non-dictatorship: if x E C(p), y d C(p), and q I x,y} =p I x,y} then y C C(q). That is,
if alternative y is not chosen at profile p but x is chosen, then y cannot be chosen at
profile q if the individual orderings of x and y are the same at q as they are at p. There
exist social choice correspondences that violate Hansson independence, but which are
rationalized by a social welfare function satisfying IIA, as we show with the Pareto
optimal correspondence and the following two profiles p and q:

p(l) p(i) for i > q(l) q(i) for i >

x z x y
z y y x
Y x z
78 D.E. Campbell and JS. Kelly

The Pareto optimal correspondence has C(p,{x,y,z})= {x,z} and C(q,{x,y,z})= {x,y}
in violation of Hansson independence. Hansson independence (or some close relative)
is used in a number of papers to prove counterparts to the theorems of Arrow and
Wilson. See Denicol6 (1985, 1993), Campbell and Kelly (1996b), and Peris and
Sanchez (1997). The key step is showing that, under a domain assumption that
is stronger than free triple, every social choice correspondence satisfying Hansson
independence is rationalized by a transitive-valued social welfare function satisfying
IIA. [See Campbell and Kelly (1996b)]. The oligarchy theorem (and related results)
for fixed agenda social choice are obtained by Peris and Sanchez (1998) and Sanchez
and Peris (1999) with a condition that is milder than Hansson independence. Peris
and Sanchez (2001) go beyond establishing the existence of dictatorship or oligarchy
or veto power, and investigate the relationship between C(p) and individuals' sets of
most-preferred alternatives at p.
Sen (1993) proposes a rejection decisiveness condition called independent deci-
siveness: for any H C N, if for every profile p such that x >-Py for all i E H there is a
profile s such that s {x,y} =p I {x,y} and y d C(s) then y ~ C(r) for any profile r such
that x >- y for all i e H. Independent decisiveness is violated by the Pareto optimality
correspondence. (For the two profiles p and q displayed in the table above, take
H = 1t}, s =p, and r = q.) Sen proves that independent decisiveness, Pareto optimality,
and non-dictatorship are incompatible if P equals P(X)N or L(X) N. Sen's Theorem
is intended to demonstrate that impossibility results can be obtained in the Arrovian
framework without any rationality condition; social choices need not be intermediated
by a social ranking. Denicol6 (1998) translates Sen's condition back into the social
welfare function framework, and then provides an example to show that the translated
independent decisiveness condition is weaker than IIA. Denicol6 also proves Arrow's
theorem with this weaker condition substituting for IIA, but states that within this
social welfare function framework, he can find no rationale for the weaker condition
that would not also justify IIA.
If we allow C(p) to be empty for some p and Y then it is possible to satisfy Hansson
independence, Pareto, and non-dictatorship on F = {X} and P =P(X)`'. In fact, each
condition is satisfied vacuously by letting C(p) be empty. Duggan (1997) gets an
impossibility theorem, strengthening Hansson (1969), by allowing C(p)=0, but only
for profiles p for which some close relative of p has a nonempty choice set.
Aizerman (1985) and Aizerman and Aleskerov (1995) review an alternative
approach to choice correspondences, developed primarily at the Institute for Control
Sciences in Moscow over two decades. Choice functions are not assumed to be
generated by an underlying binary relation - although that case is not ruled out. A key
role is played by the algebraic structure imposed on some special families of choice
functions, although some of these families are characterized by conditions that are used
in the conventional literature on the rationalization of a choice function by a binary
relation. Aleskerov (2002, Chapter 2 in this Volume), shows how this approach can
be extended to social choice, with an n-tuple of individual choice functions as input,
instead of an n-tuple of individual preference orderings.
Ch. 1: Impossibility Theorems in the Arrovian Framework 79

To introduce the probabilistic approach to preference aggregation, we assume briefly

that N= {1,2,3} is a set of three partners in a restaurant and each member of X is a
restaurant decor. One solution to the problem of selecting a decor for the restaurant is
to choose one of the partners randomly, and let that partner decide on the decor - the
random dictator rule.
In general, a probabilistic aggregation rule is a function rT that maps each
profile p G L(X)N into a probability measure Jr(p) on L(X). In this context the Pareto
criterion requires that if x >-i y for all i E N at profile p, then the lottery Zr(p) only
assigns positive probability to members of L(X) that rank x strictly above y. The
probabilistic version of IIA requires the probability of x ranking socially above y to
be the same at p as it is at q if p {x,y} = q {x,y}. Of the following three rules, only
the third satisfies both IIA and Pareto.
Example 8.1. The domain of zr is L(X)N and .r(p) is the uniform probability
distribution at any profile p. That is, r selects a member of L(X) at random according
to the uniform distribution. For any distinct x and y in X, and any profile p, the
probability that x is ranked above y at p is . Therefore, IIA is satisfied, but for the
same reason Pareto is not. ·

Example 8.2. The domain of 3r is L(X)N, there are n >3 individuals, and X
has at least three members. At any profile p the rule T selects a member of
{p(l),p(2 ),... ,p(n)} at random according to the uniform probability distribution on
the set {p(l),p(2),... ,p(n)}. For any distinct x and y in X, let p be a profile for which
x -I y, p(1)=p(2)=p(3)=... =p(n - 1), andy >-, x. Because {p(l),p(2),.. . ,p(n)} is a
two-element set, the probability that x is ranked above y at p is . If q is another profile
such that q I{x,y} =p I x,y} but q(l) X q(2 ) = q(3) ... = q(n - 1), then the probability
that x is ranked above y at q is because q(1), q(2 ),. . , q(n)} is a three-element set.
This rule violates IIA, but it satisfies Pareto because we always have ar(p)=p(i) for
some i. ·

Example 8.3. The domain of r is VN for any subset V of L(X). The rule r
selects a member i of N = 1,2,.. ., n} at random according to the uniform probability
distribution and then sets nr(p)=p(i). IIA and Pareto are both satisfied. ·
BarberA and Sonnenschein (1978) prove that for every :r satisfying Pareto and IIA
there is a probability measure yu on the family of subsets of N such that for each
profile p, and for arbitrary x, y EX, the probability of x socially ranking above y is
yi({i E N: x >-iy}). (Because Mu is a probability measure it is subadditive, which means
that t(H UJ) < 1 (H)+ t(J) for all subsets H and J of N.) McLennan (1980) shows
that t has to be additive, not merely subadditive, if X has at least six alternatives.
Subadditivity disqualifies rules that select the majority rule relation with probability 1
if it is transitive, and which select a member of {p(i): i E N} randomly when majority
rule is not transitive at p. The BarberA-Sonnenschein-McLennan result precipitates
Arrow's theorem as a corollary. Bandyopadhyay, Deb and Pattanaik (1982) prove the
80 D.E. Campbell and 1S. Kelly

Oligarchy Theorem for probabilistic social choice. BarberA and Valenciano (1983)
extend the Bandyopadhyay-Deb-Pattanaik theorem by considering functions :r that
directly determine, for arbitrary p, x and y, the probability that x will be socially
ranked above y. Chapter 24 in Volume 2 of this Handbook [Coughlin (2002)] provides
a comprehensive treatment of the probabilistic approach.
Random dictatorship may be a satisfactory form of conflict resolution for some
problems, but it is inappropriate for many standard models of resource allocation.
Suppose, for example, that X is the set of feasible allocations in an n-person exchange
economy, and P is the set of profiles of economic preferences. In particular, p(i) is
selfish and monotonic for all p E P and all i C N. If a member p(j) of {p(i): i N} is
chosen randomly and we setf(p) =p(j) then we will have x >-f() y for all y E X {x},
where x is the allocation that assigns everything to individual j.
Lain6, Le Breton and Trannoy (1986) and Weymark (1997) do not so much depart
from the Arrovian framework as provide a new and interesting interpretation. Let X
be a set of events. For arbitrary p E P, x >-iy represents the statement "Individual i
believes that event x is more likely than event y". Then f(p) is interpreted as the
consensus probability. The free triple property cannot be assumed: if z is the event "x
and y" then z cannot be less likely than x or than y, and thus z cannot be positioned
below x or y in p(i). In that case, {x,y,z} is not a free triple. But there are enough
chains of free triples to yield a version of Arrow's theorem for which the dictatorship
is slightly - but only slightly - qualified.
The "fuzzy social choice" approach requires the selection of a fuzzy social
ordering as a function of the profile. Fuzziness might be expected to side-step logical
impossibility because of its smoothing properties, but the results are mixed. The
conclusions of theorems depend on how one formulates the analogue of transitivity
and whether one works with fuzzy strict preference or fuzzy weak preference. That is
itself a negative result: a satisfactory social choice process is one that is robust with
respect to small changes in the way that the criteria are formalized. The key papers in
the literature include: Barrett, Pattanaik and Salles (1986, 1990, 1992), Dutta (1987)
and Banerjee (1994). Assuming a fuzzy or uncertain outcome set X, Campbell and
Kelly (1996a) prove a trade-off theorem. Candeal and Indurdin (1995) and Garcia-
Lapresta and Llamazares (2000) depart even farther from the Arrovian framework,
to characterize families of fuzzy social choice rules satisfying one or two simple
conditions - neutrality, for example. See Salles (1998) for a survey of the subject.
Kaplow and Shavell (2001) prove a striking new impossibility theorem, employing
only a single profile: a social utility function F maps each outcome x into a real
number. It is said to be individualistic if F(x) depends only on individual utility levels
at x. Assuming that there is at least one private good k in which F is continuous and
which each individual prefers to have more of, ceteris paribus, Kaplow and Shavell
prove that F must violate the Pareto criterion if it is not individualistic. In fact, their
proof never uses the fact that F is a function, and hence generates a complete and
transitive binary relation. One could replace F with a binary relation ~ that ranks
outcomes as a function of individual utilities and perhaps other non-utility information
Ch. 1: Impossibility Theorems in the Arrovian Framework 81

as well. The relation t need not be transitive; the Kaplow-Shavell proof goes through
without modification. Perhaps more significantly, they assume full continuity of F(x) in
private good k, but lower continuity suffices - i.e., if x >-y then there is a neighborhood
V of y such that x > z for all z in V. The significance of this weaker condition arises
from the fact that lower continuity and acyclicity of t are sufficient for the existence
of a maximal alternative on a nonempty compact set [Walker (1977)].
Inspired by Arrow's work, a number of scholars have obtained impossibility
theorems in contexts arising from issues in biology and chemistry. N represents a set
of n different types of observations on an object or a family of objects. A profile p
identifies a specific observation p(i) for each i N. For example, if the objective is
to draw a tree diagram that is an estimate of the evolutionary history of a particular
species, p(i) could be a (partial) tree diagram based on data from region i. If T is the
family of all trees that have the properties necessary for representing an evolutionary
history, then the function mapping TN into T gives the estimatef(p) for eachp E TN.
In this literature the rules f are called consensus functions, and they have been used
to analyze molecular sequences, as in Day and McMorris (1992)], and evolutionary
histories as in McMorris and Neumann (1983). See Barth6lemy, Leclerc and Monjardet
(1986) for an introduction and survey. The results cannot be obtained directly from
Arrow's theorem or its relatives, even when T is isomorphic to a superset of L(X)
because there is an additional restriction. The function must map TN into T, not into
the set of trees that is isomorphic to L(X).
Why would IIA, the Pareto criterion, and non-dictatorship be imported into the
analysis of consensus functions? A dictatorial rule is one that ignores the data in all
categories but one, and that seems to be unscientific. The Pareto criterion takes the
form of a reasonable unanimity property: If p(l) =p(2)=... =p(n) then f(p) =p(l).
Notable contributions to the consensus function literature are: McGuire and Thompson
(1978), McMorris and Neumann (1983), Barth6lemy, McMorris and Powers (1991,
1992, 1995), McMorris and Powers (1993), and Powers (1996). These papers differ
with respect to the assumption about the structure of the members of T, and in some
cases with respect to the way that IIA is incorporated. Under the supposition that T
is a set of equivalence relations, Mirkin (1975) proved an oligarchy theorem: there
is a nonempty subset J of N such that, for arbitrary p E TN, attributes x and y are
equivalent according to f(p) if and only if they are equivalent according to p(i) for
each i E J. This result was generalized to a class of value relations by Leclerc (1984).
A number of contributions to the consensus function literature relax the Arrovian
conditions and add new ones, to derive counterparts to majority rule and the median
rule. See for example: Margush and McMorris (1981), Day and McMorris (1985,
1992), McMorris and Powers (1991), Barth6lemy and Janowitz (1991) and McMorris
and Powers (1995).
Koray (2000) reaches the conclusion of Arrow's theorem by a much different route
than any other treatment of social choice. He begins with a social choice function,
which selects a member of the outcome set as a function of individual preferences,
which are assumed to be linear. The rule is unanimous if it selects the alternative that
82 D.E. Campbell and JS. Kelly

is at the top of everyone's preference ordering whenever that happens. Koray seeks a
member of the family of unanimous rules which, when used to select a social choice
function for society, will select itself. Attention is also restricted to rules with a strong
neutrality property, which allows the social choice functions under consideration to
be defined on many outcome sets; without this property one can't even address the
question, "Does the rule select itself"?
Consider a three-person society with x, y and z as the feasible outcomes. Let profile p
be as follows:

1 2 3

x y z
y z x
z x y

We will apply the following two social choice functions g and h:

g selects the majority winner if there is one; and otherwise selects the majority
winner from among the top two alternatives in person 2's ordering.
h selects the majority winner if there is one; and otherwise selects the top ranked
alternative of person 1.
Note that there is a voting cycle x >-y >-z >-x at p. There is no majority winner, so
g(p)=y and h(p)=x. Now, we can use that information to see which social choice
rule would be selected by g at p, and which would be selected by h at p: because
persons 1 and 3 both strictly prefer x to y at p, each would strictly prefer h to g at
p. Because person 2 strictly prefers y to x she would strictly prefer g to h at p. The
resulting profile is

1 2 3

h g h
g h g

Two of the three individuals prefer h to g, so both of these rules would select h. Note
that h selects itself at profile p, but g does not. Koray proved that a unanimous and
neutral social choice function selects itself at every profile if and only if it is dictatorial.
(Inverse dictatorship is neutral and self-selecting, but it violates unanimity). It is easy to
see why a dictatorial rule is self-selecting: suppose that k always chooses the alternative
that is at the top of person l's ordering. Then at any profile, there will be no social
choice function that person 1 prefers to k. Therefore, k will select itself. Barbera
and Jackson (2000) use self-selection to single out majority rule in a model with a
different structure than the Arrovian framework and with a less demanding notion of
self-selection than Koray's.
Ch. 1: Impossibility Theorems in the Arrovian Framework 83

If N is infinite, there exists a free ultrafilter on N. An ultrafilter U is free if nu, the

intersection of all the members of U, is empty. Fishburn (1970) was the first to show
that a free ultrafilter can be used to define a social welfare function satisfying all of
Arrow's conditions for any infinite society N: set x By if and only if the set {i E N:
x tiy} belongs to U. The resulting rule obviously satisfies IIA, and it satisfies Pareto
as a consequence of parts (1) and (4) of the definition of an ultrafilter. For any ultrafilter,
it is easy to show that individual i E N is a dictator for the rule f just defined if and
only if {i} e U. Hencef is non-dictatorial when U is free. (If {i} E U then i J c N
implies N \J e U by (2) because {i} c N \J. Therefore, every member of U contains
i by (1), contradicting the fact that U is free.) Finally, to prove thatf(p) is transitive,
suppose that x y t z. Then I = {i E N: x iy} and J={i C N: y ti z} both belong
to U, and thus InJ belongs to U by (3). Therefore, {iEN: xtiz}, a superset of
I n J, belongs to U by (2). We then have x z by definition. [See also Lauwers and
Van Liederkerke (1995)]. The technique that we have just described won't work with
finite N because no finite set admits a free ultrafilter, as we show in the paragraph
following the proof of Arrow's impossibility theorem (see p. 51).
Suppose that N is an interval in the real line. Kirman and Sondermann (1972)
show that iff is a transitive-valued social welfare function on a full domain, and
it satisfies IIA and Pareto, then for any positive number 6, however small, there is a
decisive coalition of Lebesgue measure less than 6. There may not be a dictator, but
there are arbitrarily small decisive coalitions, and that seems almost as unacceptable
as dictatorship. (In fact, Kirman and Sondermann prove their theorem for atomless
measure spaces in general.)
A free ultrafilter is a highly nonconstructive mathematical object, and hence it puts
substantial limits on any social welfare function satisfying Arrow's conditions for an
infinite society. Building on Armstrong (1980, 1985, 1992), and assuming an arbitrary
(possibly infinite) society, Mihara (1997a) proves that a transitive-valued social welfare
function cannot satisfy IIA, Pareto, and anonymity - i.e., symmetric treatment of
individuals - if the domain satisfies a modest richness condition.
Suppose that N is countable. Mihara (1997b) proves that a transitive-valued social
welfare function on a "rich" domain is dictatorial if it satisfies Pareto and IIA and
is computable. Lewis (1988) was the first to use computability to extend Arrow's
theorem to infinite societies. [See also Kelly (1988) for a discussion of computational
complexity in social choice.] Mihara (1999) relaxes the notion of computability, and
exhibits a social welfare function satisfying Arrow's conditions that in fact can be
computed. Mihara also gives an interesting justification for employing an infinite
society model: if there is a finite number of agents but an infinite number of possible
states of the world, yet to be realized, each member of N can be interpreted as a
particular individual in a particular state. In that context, a small decisive coalition
need not be an affront to democracy. For countable N and a connected T 1 space X
that is also locally connected, Campbell (1992c, p. 110) proves the following: iff is
a non-constant and transitive-valued social welfare function satisfying IIA then there
is some individual i E N such that eitherf(p) =p(i) for every profile p in the domain,
84 D.E. Campbell and JS. Kelly

or else f(p)=p(i)- 1 at every profile. (A full domain is not required, but something
stronger than free triple is assumed).
Diamond (1965) considered the problem of ordering the space of infinite utility
streams. This becomes a social choice problem when we view N, the set of positive
integers, as the society, with i N denoting generation i, and ui as the utility of that
generation. Diamond proved that a complete and transitive binary relation cannot treat
generations equally if it is continuous in the product topology and monotonic, which
requires u=(ul,u2 ,... ) to rank strictly above vu=(ul, 2 ,...) if Ui > Ui for all i and
ui > ui for some i. Shinotsuka (1998) showed that both monotonicity and transitivity
can be dispensed with. He proves that the null ordering (universal indifference) is
the only continuous binary relation that treats individuals symmetrically. Moreover,
Shinotsuka only requires continuity in the Mackey topology, which is much less
demanding than the product topology.

9. Concluding remarks

Logical inconsistency is a feature of even ery short lists of conditions - conditions

that capture only a small part of the intuitive properties that we would like a rule to
have. For example, in many situations, for most alternatives and nearly all individuals,
we would want at least some rough equality of treatment of individuals, but we often
only ask for non-dictatorship, or the absence of an oligarchy, or something similar
and very weak. [Young (1994) abounds with applications in which fairness is one of
the key considerations.] There appears to be no chance of finding rules that come
remotely "close" to embodying the spirit of all the appropriate criteria. The evidence
of fifty years of research since Arrow discovered his theorem is that there is no ideal
aggregation procedure that can be used for all social choice problems.
For a specific social choice problem, one must assess the appropriateness of each
aggregation criterion, by taking into consideration the context in which the group
choice is to be made. [See Kelly (1978, pp. 159-161).] Arrow's formal system has
many different possible interpretations of the primitive notions, like X and N. Whether
or not transitivity or IIA is desirable is very sensitive to the interpretation of these
primitives. For most social choice problems, at least one of Arrow's conditions will be
inappropriately strong. Therefore, the response to Arrow's theorem could be to give up
the attempt to find one framework that covers every application. Instead, the approach
would be ad hoc, using a different set of conditions for different social choice contexts.
This is in fact the path that many social choice theorists have taken in recent years.
Notable examples are: fair allocation [Thomson (2002, Chapter 26 in Volume 2 of this
Handbook)]; serial cost sharing [Moulin (2002, Chapter 6 in this Volume)]; inequality
and poverty measurement [Dutta (2002, Chapter 12 in this Volume)]; and variable
population problems [Blackorby, Bossert and Donaldson (1995, 2002, Chapter 11 in
this Volume)]. In fact, almost all of the other chapters in the two volumes of this
Handbook could be mentioned here.
Ch. 1: Impossibility Theorems in the Arrovian Framework 85

The literature cited in the previous paragraph embraces a very large number of
applications, and most of them identify a social choice procedure satisfying a list
of conditions considered appropriate for the context at hand. The value of these
specialized approaches stems from the fact that each assumes a lot of structure for
the feasible set X and for individual preferences. This structure allows the researcher
to express criteria that are somewhat different from Arrow's. But much of this literature
presents (1) impossibility theorems; or (2) characterization theorems, which are then
just one additional criterion away from an impossibility result; or (3) demonstrations
that the selection of choice procedures that satisfy criteria lists are extremely sensitive
to small details in the specification of the context and the criteria. One of the authors
of this chapter is more sanguine about the value of these exercises. He feels that the
researcher will develop an intuition that will enable her to recommend good, if not
best, procedures for particular social choice problems. The other author believes that
even in these restricted contexts there will always be reasonable criteria violated by
every rule; we will not be able to identify "good" outcomes as the result of applying
"good" social choice procedures.

References

Aizerman, M. (1985), "New problems in the general choice theory. Review of a research trend", Social
Choice and Welfare 2:235-282.
Aizerman, M., and E Aleskerov (1995), Theory of Choice (North-Holland, Amsterdam).
Aleskerov, E (2002), "Categories of Arrovian voting schemes", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. I (Elsevier, Amsterdam) Chapter 2, this volume.
Armstrong, TE. (1980), "Arrow's theorem with restricted coalition algebras", Journal of Mathematical
Economics 7:55-75.
Armstrong, T.E. (1985), "Precisely dictatorial social welfare functions: Erratum and addendum to 'Arrow's
theorem with restricted coalition algebras' ", Journal of Mathematical Economics 14:57-59.
Armstrong, T.E. (1992), "Hierarchical Arrow social welfare functions", Economic Theory 2:27-41.
Arrow, K.J. (1950), "A difficulty in the concept of social welfare", Journal of Political Economy
58:328-246.
Arrow, K.J. (1951), Social Choice and Individual Values, 1st edition (Wiley, New York).
Arrow, K.J. (1959), "Rational choice functions and orderings", Economica 26:121-127.
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd. Edition (Wiley, New York).
Arrow, K.J. (1969), "Tullock and an existence theorem", Public Choice 6:105-111.
Arrow, K.J. (1983), "A difficulty in the concept of social welfare", in: Collected Papers of Kenneth J.
Arrow, Vol. 1: Social Choice and Justice (Harvard University Press, Cambridge, MA) pp. 1-29.
Arrow, K.J. (1997), "Introduction", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Social Choice
Re-examined (Macmillan, London) pp. xvi-xvii.
Arrow, K.J., and H. Raynaud (1986), Social Choice and Multicriterion Decision Making (MIT Press,
Cambridge, MA).
Baigent, N. (1987), "Twitching weak dictators", Journal of Economics 47:407-411.
Baigent, N. (2002), "Topological theories of social choice", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 17, forthcoming.
Balasko, Y, and C.A. Crbs (1997), "The probability of Condorcet cycles and super majority rules",
Journal of Economic Theory 75:237-270.
86 D.E. Campbell and J.S. Kelly

Bandyopadhyay, T. (1984), "On the frontier between possibility and impossibility theorems in social
choice", Journal of Economic Theory 32:52-66.
Bandyopadhyay, T. (1985), "Pareto optimality and the decisive power structure with expansion consistency
conditions", Journal of Economic Theory 35:366-378.
Bandyopadhyay, T. (1986), "Rationality, path independence, and power structure", Journal of Economic
Theory 40:338-348.
Bandyopadhyay, T. (1990), "Sequential path independence and social choice", Social Choice and Welfare
7:209-220.
Bandyopadhyay, T., R. Deb and PK. Pattanaik (1982), "The structure of coalitional power under
probabilistic group decision rules", Journal of Economic Theory 27:366-375.
Banerjee, A. (1994), "Fuzzy preferences and Arrow-type problems in social choice", Social Choice and
Welfare 11:121-130.
Banks, J.S. (1995), "Acyclic social choice from finite sets", Social Choice and Welfare 12:293-310.
Banks, J.S. (1996), "Singularity theory and core existence in the spatial model", Journal of Mathematical
Economics 24:523-536.
Barberl, S. (1983), "Pivotal voters: A new proof of Arrow's theorem", Economics Letters 6:13-16.
BarberA, S. (2002), "Strategy proofness", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook
of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 23, forthcoming.
BarberA, S., and M.O. Jackson (2000), "Choosing how to choose: Self-stable majority rules", Mimeo
(Division of Humanities and Social Sciences, California Institute of Technology).
BarberA, S., and H. Sonnenschein (1978), "Preference aggregation with randomized social orderings",
Journal of Economic Theory 18:244-254.
BarberA, S., and F Valenciano (1983), "Collective probabilistic judgements", Econometrica
51:1033-1046.
Barrett, C.R., PK. Pattanaik and M. Salles (1986), "On the structure of fuzzy social welfare functions",
Fuzzy Sets and Systems 19:1-10.
Barrett, C.R., P.K. Pattanaik and M. Salles (1990), "On choosing rationally when preferences are fuzzy",
Fuzzy Sets and Systems 34:197-212.
Barrett, C.R., PK. Pattanaik and M. Salles (1992), "Rationality and aggregation of preferences in an
ordinally fuzzy framework", Fuzzy Sets and Systems 49:9-13.
Barthelemy, J.P., and M.E Janowitz (1991), "A formal theory of consensus", SIAM Journal on Discrete
Mathematics 4:305-322.
Barth6lemy, J.P, B. Leclerc and B. Monjardet (1986), "On the use of ordered sets in problems of
comparison and consensus of classifications", Journal of Classification 3:187-224.
Barthblemy, J.P., F.R. McMorris and R.C. Powers (1991), "Independence conditions for consensus n-trees
revisited", Applied Mathematics Letters 4:43-46.
Barth6lemy, J.P, ER. McMorris and R.C. Powers (1992), "Dictatorial consensus functions on n-trees",
Mathematical Social Sciences 25:59-64.
Barth6lemy, JP., FR. McMorris and R.C. Powers (1995), "Stability conditions for concensus functions
defined on n-trees", Mathematical Computer Modelling 22:79-87.
Beja, A. (1989), "Finite and infinite complexity in axioms of rational choice or Sen's characterization
of preference-compatibility cannot be improved", Journal of Economic Theory 49:339-346.
Black, D. (1948), "On the rationale of group decision making", Journal of Political Economy 56:23-34.
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, London/
New York).
Blackorby, C., D. Donaldson and J.A. Weymark (1990), "A welfarist proof of Arrow's theorem",
Recherches Economiques de Louvain 56:259-286.
Blackorby, C., W. Bossert and D. Donaldson (1995), "Intertemporal population ethics: critical-level
utilitarian principles", Econometrica 63:1303-1320.
Blackorby, C., W. Bossert and D. Donaldson (2002), "Utilitarianism and the theory of justice", in:
Ch. 1: Impossibility Theorems in the Arrovian Framework 87

K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 1
(Elsevier, Amsterdam) Chapter 11, this volume.
Blair, D.H., and R.A. Pollak (1979), "Collective rationality and dictatorship: The scope of the Arrow
theorem", Journal of Economic Theory 21:186-194.
Blair, D.H., and R.A. Pollak (1982), "Acyclic collective choice rules", Econometrica 50:931-943.
Blair, D.H., G.A. Bordes, J.S. Kelly and K. Suzumura (1976), "Impossibility theorems without collective
rationality", Journal of Economic Theory 13:361-376.
Blau, J.H. (1957), "The existence of social welfare functions", Econometrica 25:302-313.
Blau, J.H. (1971), "Arrow's theorem with weak independence", Economica 38:413-420.
Blau, J.H. (1972), "A direct proof of the Arrow theorem", Econometrica 40:61-67.
Blau, J.H. (1979), "Semiorders and collective choice", Journal of Economic Theory 21:195-206.
Blau, J.H., and R. Deb (1977), "Social decision fimnctions and the veto", Econometrica 45:871-879.
Borda (J.-C. de Borda) (1781), "M6moire sur les elections par scrutin", M6moires de l'Acad6mie Royale
des Sciences ann6e 1781, pp. 657 665. Translated in English in 1953 by A. de Grazia: "Mathematical
derivation of an election system", Isis 44:42-51.
Bordes, G.A., and M. Le Breton (1990), "Arrovian theorems for economic domains: The case where
there are simultaneously private and public goods", Social Choice and Welfare 7:1-18.
Bordes, G.A., and T.N. Tideman (1991), "Independence of irrelevant alternatives in the theory of voting",
Theory and Decision 30:163-186.
Bossert, W. (2000), "Welfarism and information invariance", Social Choice and Welfare 17:321-336.
Brown, D.J. (1974), "An approximate solution to Arrow's problem", Journal of Economic Theory
9:375-383.
Brown, D.J. (1975), "Aggregation of preferences", Quarterly Journal of Economics 89:456-469.
Campbell, D.E. (1975), "Realization of choice functions", Econometrica 46:171-180.
Campbell, D.E. (1992a), "Transitive social choice in economic environments", International Economic
Review 33:341-352.
Campbell, D.E. (1992b), "Implementation of social welfare functions", International Economic Review
33:525-533.
Campbell, D.E. (1992c), Equity, Efficiency, and Social Choice (Clarendon Press, Oxford).
Campbell, D.E. (1995), "Nonbinary social choice for economic environments", Social Choice and Welfare
12:245-254.
Campbell, D.E., and J.S. Kelly (1993), "t or 1 - t. That is the trade-off", Econometrica 61:1355-1365.
Campbell, D.E., and J.S. Kelly (1994a), "Trade-off theory", American Economic Review, Papers and
Proceedings 84:422-426.
Campbell, D.E., and J.S. Kelly (1994b), "Arrovian social choice correspondences", International
Economic Review 37:803-824.
Campbell, D.E., and J.S. Kelly (1995a), "Nondictatorially independent pairs", Social Choice and Welfare
12:75-86.
Campbell, D.E., and J.S. Kelly (1995b), "Continuous-valued social choice", Journal of Mathematical
Economics 25:195-211.
Campbell, D.E., and J.S. Kelly (1995c), "Lebesgue measure and social choice trade-offs", Economic
Theory 5:445-459.
Campbell, D.E., and J.S. Kelly (1995d), "Asymptotic density and social choice trade-offs", Mathematical
Social Sciences 29:181-194.
Campbell, D.E., and J.S. Kelly (1996a), "Social choice trade-offs for an arbitrary measure: with application
to uncertain or fuzzy agenda", Economics Letters 50:99-104.
Campbell, D.E., and J.S. Kelly (1996b), "Independent social choice correspondences", Theory and
Decision 41:1-11.
Campbell, D.E., and J.S. Kelly (1997), "The possibility-impossibility boundary in social choice", in:
K.J. Arrow, A.K. Sen and K. Suzumura, eds., Social Choice Re-examined (Macmillan, London)
pp. 179-204.
88 D.E. Campbell and JS. Kelly

Campbell, D.E., and J.S. Kelly (1998), "Quasitransitive social preference: why some very large coalitions
have very little power", Economic Theory 12:147-162.
Campbell, D.E., and J.S. Kelly (2000a), "Information and preference aggregation", Social Choice and
Welfare 17:85-93.
Campbell, D.E., and J.S. Kelly (2000b), "Weak independence and veto power", Economics Letters
66:183-189.
Candeal, J.C., and E. Indurin (1995), "Aggregation of preferences from algebraic models on groups",
Social Choice and Welfare 12:165-173.
Cantillon, E., and A. Rangel (2001), "A graphical analysis of some basic results in social choice", Social
Choice and Welfare, forthcoming.
Caplin, A., and B. Nalebuff (1988), "On 64%-majority rule", Econometrica 56:787-814.
Caplin, A., and B. Nalebuff(1991), "Aggregation and social choice: A mean voter theorem", Econometrica
59:1-24.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probabilit des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Coughlin, P.J. (2002), "Probabilistic and spatial models of voting", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 24,
forthcoming.
Dardanoni, V (2001), "A pedagogical proof of Arrow's impossibility theorem", Social Choice and
Welfare 18:107-112.
d'Aspremont, C. (1985), "Axioms for social welfare orderings", in: L. Hurwicz, D. Schmeidler and
H. Sonnenschein, eds., Social Goals and Social Organization (Cambridge University Press, Cambridge)
pp. 19-76.
d'Aspremont, C., and L. Gevers (2002), "Social welfare functionals and interpersonal comparability",
in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 1
(Elsevier, Amsterdam) Chapter 10, this volume.
Davis, O.A., M.H. De Groot and M.J. Hinich (1972), "Social preference orderings and majority rule",
Econometrica 40:147-157.
Day, H.E., and ER. McMorris (1985), "A formalization of consensus index methods", Bulletin of
Mathematical Biology 47:215-229.
Day, H.E., and ER. McMorris (1992), "Critical comparison of consensus methods for molecular
sequences", Nucleic Acids Research 20:1093-1099.
Deb, R. (2002), "Nonbinary social choice theory", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds.,
Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 18, forthcoming.
Debord, B. (1992), "An axiomatic characterization of Borda's k-choice function", Social Choice and
Welfare 9:337-344.
Denicol6, V (1985), "Independent social choice correspondences are dictatorial", Economics Letters
19:9-12.
Denicol6, V (1993), "Fixed agenda social choice theory: Correspondence theorems for social choice
correspondences and social decision functions", Journal of Economic Theory 59:324-332.
Denicol6, V (1996), "An elementary proof of Arrow's impossibility theorem", The Japanese Economic
Review 47:432-435.
Denicolb, V (1998), "Independent decisiveness and the Arrow theorem", Social Choice and Welfare
15:563-566.
Denicol6, V (2001), "An elementary proof of Arrow's impossibility theorem: correction", The Japanese
Economic Review 52:134-135.
Diamond, P. (1965), "The evaluation of infinite utility streams", Econometrica 33:170-177.
Duggan, J. (1997), "Hansson's theorem for generalized social welfare functions: an extension", Social
Choice and Welfare 14:471-478.
Dutta, B. (1987), "Fuzzy preferences and social choice", Mathematical Social Sciences 13:215-229.
Ch. 1: Impossibility Theorems in the Arrovian Framework 89

Dutta, B. (2002), "Inequality, poverty, and welfare", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds.,
Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 12, this volume.
Dutta, B., and Arunava Sen (1996), "Ranking opportunity sets and Arrow impossibility theorems:
Correspondence results", Journal of Economic Theory 71:90-101.
Enelow, J.M. (1997), "Cycling and majority rule", in: D. Mueller, ed., Perspectives on Public Choice:
A Handbook (Cambridge University Press, Cambridge) pp. 149-162.
Erdfs, P., and L. Moser (1964), "On the representation of directed graphs as the unions of orderings",
Magyar Tud. Akad. Mat. Kutat6 Int. Kzl. [Publication of the Mathematical Institute of the Hungarian
Academy of Sciences] 9:125-132.
Feiwel, G.R. (1987), "The many dimensions of Kenneth J. Arrow", in: G.R. Feiwel, ed., Arrow and the
Foundations of the Theory of Economic Policy (New York University Press, New York) pp. 1-11.
Ferejohn, J.A., and P.C. Fishburn (1979), "Representation of binary decision rules by generalized
decisiveness structures", Journal of Economic Theory 21:28-45.
Fishburn, P.C. (1970), "Arrow's impossibility theorem: Concise proof and infinite voters", Journal of
Economic Theory 2:103-106.
Fishburn, PC. (1973), The Theory of Social Choice (Princeton University Press, Princeton, NJ).
Fishburn, P.C. (1987), Interprofile Conditions and Impossibility (Harwood, Chur, Switzerland).
Fishburn, PC., and S.J. Brams (2002), "Voting procedures", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 4, this volume.
Fishburn, P.C., and J.S. Kelly (1997), "Super-Arrovian domains with strict preferences", SIAM Journal
of Discrete Mathematics 10:83-95.
Fishburn, P.C., WV Gehrlein and E. Maskin (1979a), "A progress report on Kelly's majority conjectures",
Economics Letters 2:313-314.
Fishburn, P.C., W.V Gehrlein and E. Maskin (1979b), "Condorcet proportions and Kelly's conjectures",
Discrete Applied Mathematics 1:229-252.
Fleurbaey, M., and E Maniquet (2001), "New possibilities for the concept of social welfare", Mimeo
(CATT, Universit6 de Pau).
Fountain, J., and K. Suzumura (1982), "Collective choice rules without the Pareto principle", International
Economic Review 23:299-308.
Gaertner, W (2001), Domain Conditions in Social Choice Theory (Cambridge University Press, New
York).
Gaertner, W (2002), "Domain restrictions", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook
of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 3, this volume.
Garcia-Lapresta, J.L., and B. Llamazares (2000), "Aggregation of fuzzy preferences: some rules of the
mean", Social Choice and Welfare 17:673-690.
Geanakoplos, J. (1996), "Three brief proofs of Arrow's impossibility theorem", http://cowles.econ.yale.edu/
faculty/geanakoplos.htm.
Gehrlein, W.V (1997), "Condorcet's paradox and the Condorcet efficiency of voting rules", Mathematica
Japonica 45:173 199.
Gevers, L. (1979), "On interpersonal comparability and social welfare orderings", Econometrica 47:
75-89.
Gibbard, A.F (1969), "Social choice and the Arrow condition", Mimeograph (Harvard University).
Gibbard, A.EF, A. Hylland and J.A. Weymark (1987), "Arrow's theorem with a fixed feasible alternative",
Social Choice and Welfare 4:105-115.
Grandmont, J.M. (1978), "Intermediate preferences and the majority rule", Econometrica 46:317-330.
Greenberg, J. (1979), "Consistent majority rules over compact sets of alternatives", Econometrica
47:627-636.
Grether, D.M., and C.R. Plott (1982), "Nonbinary social choice: An impossibility theorem", Review of
Economic Studies 49:143-149.
Guha, A.S. (1972), "Neutrality, monotonicity, and the right of veto", Econometrica 40:821-826.
90 D.E. Campbell and .S. Kelly

Hammond, PJ. (1976), "Why ethical measures of inequality need interpersonal comparisons", Theory
and Decision 7:263-274.
Hansson, B. (1969), "Voting and group decision functions", Synthese 20:526-537.
Hansson, B. (1972), "The existence of group preferences", Working Paper No. 3 (The Mattias Fremling
Society, Lund, Sweden).
Herzberger, H. (1973), "Ordinal preference and rational choice", Econometrica 41:187-237.
Hollard, G., and M. Le Breton (1996), "Logrolling and a McGarvey theorem for separable tournaments",
Social Choice and Welfare 13:451-455.
Johnson, M.R. (1990), "Information, associativity, and choice requirements", Journal of Economic Theory
52:440 452.
Johnson, M.R. (1995), "Ideal structures of path independent social choice functions", Journal of Economic
Theory 65:468-504.
Kalai, E., and D. Schmeidler (1977), "Aggregation procedure for cardinal preferences: a formulation
and proof of Samuelson's impossibility conjecture", Econometrica 45:1431-1438.
Kalai, E., E. Muller and M.A. Satterthwaite (1979), "Social welfare functions when preferences are
convex, strictly monotonic and continuous", Public Choice 34:87-97.
Kaplow, L., and S. Shavell (2001), "Any non-welfarist method of policy assessment violates the Pareto
principle", Journal of Political Economy 109:281-286.
Kelly, J.S. (1974), "Voting anomalies, the number of voters and the number of alternatives", Econometrica
42:239-251.
Kelly, J.S. (1978), Arrow Impossibility Theorems (Academic Press, New York).
Kelly, J.S. (1987), "An interview with Kenneth J. Arrow", Social Choice and Welfare 4:43-62.
Kelly, J.S. (1988), "Social choice and computational complexity", Journal of Mathematical Economics
17:1-8.
Kelly, J.S. (1994a), "The free triple assumption", Social Choice and Welfare 11:97-101.
Kelly, J.S. (1994b), "The Bordes-Le Breton exceptional case", Social Choice and Welfare 11:273-281.
Kelly, J.S. (1994c), "Conjectures and unsolved problems: 1. Condorcet proportions", Social Choice and
Welfare 3:311-314.
Kelsey, D. (1985), "Acyclic choice and group veto", Social Choice and Welfare 1:131-137.
Kelsey, D. (1988), "What is responsible for the 'Paretian epidemic'?", Social Choice and Welfare
5:303-306.
Kemp, M.C., and Y.-K. Ng (1976), "On the existence of social welfare functions, social orderings and
social decision functions", Economica 43:59-66.
Kirman, A.P., and D. Sondermann (1972), "Arrow's theorem, many agents, and invisible dictators",
Journal of Economic Theory 5:267-277.
Koray, S. (2000), "Self-selective social choice functions verify Arrow and Gibbard-Satterthwaite
theorems", Econometrica 68:981-995..
Kramer, G.H. (1973), "On a class of equilibrium conditions for majority rule", Econometrica 41:
285-297.
Krause, U. (1995), "Essentially lexicographic aggregation", Social Choice and Welfare 12:233-244.
Lain, J., M. Le Breton and A. Trannoy (1986), "Group decision making under uncertainty: a note on
the aggregation of 'ordinal probabilities'-", Theory and Decision 11:155-161.
Laslier, J.-E (1997), Tournament Solutions and Majority Voting (Springer, Berlin).
Lauwers, L., and L. Van Liederkerke (1995), "Ultraproducts and aggregation", Journal of Mathematical
Economics 24:217-234.
Le Breton, M., and M. Truchon (1995), "Acyclicity and the dispersion of veto power", Social Choice
and Welfare 12:43-58.
Le Breton, M., and J.A. Weymark (2002a), "Arrovian social choice theory on economic domains",
in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2
(Elsevier, Amsterdam) Chapter 16, forthcoming.
Ch. 1: Impossibility Theorems in the Arrovian Framework 91

Le Breton, M., and J.A. Weymark (2002b), "Social choice with analytic preferences", Social Choice and
Welfare, forthcoming.
Leclerc, B. (1984), "Efficient and binary consensus functions on transitive valued relations", Mathematical
Social Sciences 8:45 61.
Levchenkov, VS. (1999a), "Self-consistent rule for group choice I: Axiomatic approach", working paper
(Moscow State University).
Levchenkov, VS. (1999b), "Self-consistent rule for group choice II: Dynamic approach", working paper
(Moscow State University).
Lewis, A.A. (1988), "An infinite version of Arrow's theorem in the effective setting", Mathematical
Social Sciences 16:41-48.
Ma, B.K., and J.H. Weiss (1995), "On the invariance of a mean voter theorem", Journal of Economic
Theory 66:264-274.
Mala, J. (1999), "On -majority voting paradoxes", Mathematical Social Sciences 37:39-44.
Malawski, M., and L. Zhou (1994), "A note on social choice without the Pareto principle", Social Choice
and Welfare 11:103-108.
Margush, T., and ER. McMorris (1981), "Consensus n-trees", Bulletin of Mathematical Biology 43:
239-234.
Mas-Colell, A., and H. Sonnenschein (1972), "General possibility theorems for group decisions", Review
of Economic Studies 39:185 192.
May, K.O. (1952), "A set of independent necessary and sufficient conditions for majority decision",
Econometrica 20:680-684.
May, K.O. (1954), "Intransitivity, utility and the aggregation of preference patterns", Econometrica
21:1 13.
McGarvey, D.C. (1953), "A theorem on the construction of voting paradoxes", Econometrica
11:608-610.
McGuire, J.B., and C.J. Thompson (1978), "On the reconstruction of an evolutionary order", Journal of
Theoretical Biology 75:141-147.
McKelvey, R.D. (1979), "General conditions for global intransitivities in formal voting models",
Econometrica 47:1085-1111.
McLennan, A. (1980), "Randomized preference aggregation: additivity of power and strategy proofness",
Journal of Economic Theory 12:1-11.
McMorris, ER., and D. Neumann (1983), "Consensus functions defined on trees", Mathematical Social
Sciences 4:131 136.
McMorris, ER., and R.C. Powers (1991), "Consensus weak hierarchies", Bulletin of Mathematical
Biology 53:679-684.
McMorris, ER., and R.C. Powers (1993), "Consensus functions on trees that satisfy an independence
axiom", Discrete and Applied Mathematics 47:47-55.
McMorris, ER., and R.C. Powers (1995), "The median procedure in a formal theory of consensus",
SIAM Journal on Discrete and Applied Mathematics 8:507-516.
Mihara, H.R. (1997a), "Anonymity and neutrality in Arrow's theorem with restricted coalition algebras",
Social Choice and Welfare 14:503-512.
Mihara, H.R. (1997b), "Arrow's theorem and Turing computability", Economic Theory 10:257-276.
Mihara, H.R. (1999), "Arrow's theorem, countably many agents, and more visible dictators", Journal of
Mathematical Economics 32:267-287.
Mirkin, B. (1975), "On the problem of reconciling partitions", in: H.M. Blalock, A. Aganbegian,
EM. Borodkin, R. Boudin and V Capecchi, eds., Quantitative Sociology. International Perspectives
on Mathematical and Statistical Modelling (Academic Press, New York) pp. 441-449.
Monjardet, B. (1983), "On the use of ultrafilters in social choice theory", in: P.K. Pattanaik and M. Salles,
eds., Social Choice and Welfare (North-Holland, Amsterdam) pp. 73-78.
Moulin, H. (1994), "Social choice", in: R.J. Aumann and S. Hart, eds., Handbook of Game Theory with
Economic Applications, Vol. 2 (North-Holland, Amsterdam) pp. 1091-1125.
92 D.E. Canmpbell and JS. Kelly

Moulin, H. (2002), "Axiomatic cost and surplus-sharing", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 6, this volume.
Murakami, Y. (1968), Logic and Social Choice (Routledge & Kegan Paul, London).
Nagahisa, R. (1991), "Acyclic and continuous social choice in T connected spaces including its
application to economic environments", Social Choice and Welfare 8:319-332.
Nermuth, M. (1992), "Two-stage aggregation: the Ostrogorski paradox and related phenomena", Social
Choice and Welfare 9:99-116.
Nurmi, H. (1999), Voting Paradoxes and How to Deal with Them (Springer, Berlin).
Parks, R.P. (1976), "An impossibility theorem for fixed preferences: A dictatorial Bergson-Samuelson
welfare function", Review of Economic Studies 43:447-450.
Pattanaik, PK. (1997), "Some paradoxes of preference aggregation", in: D. Mueller, ed., Perspectives on
Public Choice: A Handbook (Cambridge University Press, Cambridge) pp. 201-225.
Pattanaik, PK. (2002), "Positional rules of collective decision-making", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 7,
this volume.
Peris, J.E., and M.C. Sanchez (1997), "Fixed agenda social choice correspondences", Mimeo (Universidad
de Alicante).
Peris, J.E., and M.C. SBnchez (1998), "An oligarchy theorem in fixed agenda without Pareto conditions",
Mimeo (Universidad de Alicante).
Peris, J.E., and M.C. Snchez (2001), "Characterization of social choice sets in terms of individuals'
maximal sets: the fixed agenda framework", Social Choice and Welfare 18:113-127.
Plott, C.R. (1967), "A notion of equilibrium and its possibility under majority rule", American Economic
Review 57:788-806.
Plott, C.R. (1973), "Path independence, rationality, and social choice", Econometrica 41:1075-1091.
Pollak, R.A. (1979), "Bergson-Samuelson social welfare functions and the theory of social choice",
Quarterly Journal of Economics 93:73-90.
Pouzet, M. (1998), "A projection property and Arrow's impossibility theorem", Discrete Mathematics
192:293-308.
Powers, R.C. (1996), "Arrow's theorem for closed weak hierarchies", Discrete Applied Mathematics
66:271-278.
Powers, R.C. (2001), "Nondictatorially independent pairs and Pareto", Social Choice and Welfare,
forthcoming.
Redekop, J. (1991), "Social welfare functions on restricted economic domains", Journal of Economic
Theory 53:396-427.
Redekop, J. (1993), "Social welfare functions on parametric domains", Social Choice and Welfare
10:127-148.
Reny, P.J. (2001), "Arrow's theorem and the Gibbard-Satterthwaite theorem: a unified approach",
Economic Letters 70:99-105.
Roberts, K.W.S. (1980), "Social choice theory: single-profile and multi-profile approaches", Review of
Economic Studies 47:441-450.
Roberts, K.WS. (1997), "Objective interpersonal comparisons of utility", Social Choice and Welfare
14:791-796.
Rubinstein, A. (1979), "A note about the "nowhere denseness" of societies having an equilibrium under
majority rule", Econometrica 47:511-514.
Rubinstein, A. (1984), "The single profile analogs to multiprofile theorems: mathematical logic's
approach", International Economic Review 15:719-730.
Rubinstein, A., and P.C. Fishburn (1986), "Algebraic aggregation theory", Journal of Economic Theory
38:63-77.
Saari, D.G. (1989), "A dictionary for voting paradoxes", Journal of Economic Theory 48:443-475.
Saari, D.G. (1994), Geometry of Voting (Springer, Berlin).
Saari, D.G. (1996), "The mathematical symmetry of choosing", Mathematica Japonica 44:183-200.
Ch. 1: Impossibility Theorems in the Arrovian Framework 93

Saari, D.G. (2000a), "Mathematical structure of voting paradoxes I: Pairwise vote", Economic Theory
15:1-53.
Saari, D.G. (2000b), "Mathematical structure of voting paradoxes II: Positional voting", Economic Theory
15:55-101.
Saari, D.G. (2002), "Geometry of voting", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook
of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 25, forthcoming.
Salles, M. (1998), "Fuzzy utility", in: S. BarberA, P.J. Hammond and C. Seidl, eds., Handbook of Utility
Theory, Vol. 1: Principles (Kluwer, Dordrecht) pp. 321-344.
Sanchez, M.C., and J.E. Peris (1999), "Veto in fixed agenda social choice correspondences", Social
Choice and Welfare 16:81-97.
Schofield, N. (1985), Social Choice and Democracy (Springer, Berlin).
Schwartz, T. (2001), "From Arrow to cycles, instability, and chaos by untying alternatives", Social
Choice and Welfare 18:1-22.
Sen, A.K. (1966), "A possibility theorem on majority decisions", Econometrica 34:491-499.
Sen, A.K. (1969), "Quasi-transitivity, rational choice, and collective decisions", Review of Economic
Studies 36:381-393.
Sen, A.K. (1970a), Collective Choice and Social Welfare (Holden-Day, San Francisco, CA).
Sen, A.K. (1970b), "The impossibility of a Paretian liberal", Journal of Political Economy 78:152-157.
Sen, A.K. (1976), "Liberty, unanimity, and rights", Economica 43:217-243.
Sen, A.K. (1977), "On weight and measures: informational constraints in social welfare analysis",
Econometrica 45:1539-1572.
Sen, A.K. (1979), "Personal utilities and public judgements: or, what's wrong with welfare economics",
Economic Journal 89:537-558.
Sen, A.K. (1986), "Social choice theory", in: K.J. Arrow and M.D. Intrilligator, eds., Handbook of
Mathematical Economics, Vol. III (North-Holland, Amsterdam) pp. 1073-1181.
Sen, A.K. (1987), "Social choice", in: J. Eatwell, M. Milgate and P. Newman, eds., The New Palgrave
Dictionary of Economics, Vol. 4 (Macmillan, London) pp. 382-393.
Sen, A.K. (1993), "Internal consistency of choice", Econometrica 61:495-521.
Sen, A.K. (1999), "The possibility of social choice", American Economic Review 89:349-378.
Sen, A.K. (2002), "Informational basis of social choice theory", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 14,
forthcoming.
Shinotsuka, T. (1998), "Equity, continuity, and myopia: a generalization of Diamond's impossibility
theorem", Social Choice and Welfare 15:21-30.
Smith, J.H. (1973), "Aggregation of preferences with variable electorate", Econometrica 41:1027-1041.
Stearns, R. (1959), "The voting problem", American Mathematical Monthly 66:761-763.
Suzumura, K. (1983), Rational Choice, Collective Decisions, and Social Welfare (Cambridge University
Press, Cambridge).
Suzumura, K. (1988), "Reduction of social choice problems: a simple proof of Arrow's general possibility
theorem", Hitotsubashi Journal of Economics 19:219-221.
Suzumura, K. (2002), "Welfare, rights, and social choice procedures", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 19,
forthcoming.
Thomson, W (2001), "On the axiomatic method and its recent applications to game theory and resource
allocation", Social Choice and Welfare 18:327-386.
Thomson, W (2002), "Fair allocations", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of
Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 26, forthcoming.
Tovey, C.A. (1997), "Probabilities of preferences and cycles with super majority rules", Journal of
Economic Theory 75:271-279.
Tullock, G. (1967), "The general irrelevance of the general impossibility theorem", Quarterly Journal of
Economics 81:256-270.
94 D.E. Campbell and JS. Kelly

Vidu, L. (1999), "An extension of a theorem on the aggregation of separable preferences", Social Choice
and Welfare 16:159-167.
von Neumann, J., and 0. Morgenstern (1944), Theory of Games and Economic Behavior (Princeton
University Press, Princeton, NJ).
Walker, M. (1977), "On the existence of maximal elements", Journal of Economic Theory 16:470 474.
Weymark, J.A. (1997), "Aggregating ordinal probabilities on finite sets", Journal of Economic Theory
75:407-432.
Wilson, R.B. (1972), "Social choice without the Pareto principle", Journal of Economic Theory 5:14-20.
Wilson, R.B. (1975), "On the theory of aggregation", Journal of Economic Theory 10:89-99.
Yanovskaya, Y. (1991), "Social choice problems with fixed sets of alternatives", Mathematical Social
Sciences 11:129-152.
Young, H.P. (1974), "An axiomatization of Borda's rule", Journal of Economic Theory 9:43-52.
Young, H.P (1994), Equity in Theory and Practice (Princeton University Press, Princeton, NJ).
Chapter 2

CATEGORIES OF ARROVIAN VOTING SCHEMES

FUAD ALESKEROV*
Russian Academy of Sciences, Institute of Control Sciences, Moscow, Russia

Contents
Abstract 97
Keywords 97
1. Introduction 98
2. Voting: a general description 99
3. Rationality of individual opinions and collective decisions 100
3.1. Binary relations and choice functions 100
3.2. Expansion-contraction axioms 101
4. Social decision rules 103
4.1. Locality 103
4.2. Normative conditions 104
4.3. Rules from the Central Class 105
4.4. Rationality constraints and rules from the Central Class 108
4.5. Remarks and guide to the literature 109
5. Functional voting rules 112
5.1. Locality 112
5.2. Normative conditions 113
5.3. Rules from the Central Class 113
5.4. Rationality constraints and rules from the Central Class 114
5.5. Remarks and guide to the literature 115
6. Social choice correspondences 116
6.1. Locality 116
6.2. Normative conditions 117

* I would like to express my deep gratitude to Professors Kenneth Arrow, Salvador BarberA, Michel Le
Breton, Charles Plott, Norman Schofield, Amartya Sen and Kotaro Suzumura for their permanent interest
in and support for this work. I also thank colleagues from Turkey for the help and support: Professors
Selguk Erez, Hasan Ersel, evket Pamuk. I appreciate the editorial comments made by Professor Kenneth
Arrow and an anonymous referee. My thanks are also due to Professors Pavel Chebotarev, Betina Klaus,
Ismail Saglam and Mr. Vyacheslav Yakuba. Their suggestions helped me to improve the text. Naturally,
all shortcomings are solely my responsibility.
This work has been partially supported by a grant of the European Community (INTAS Project
"Measurement and Aggregation of Preferences"). This support is gratefully acknowledged.

Handbook of Social Choice and Welfare, Volume 1, Edited by KJ Arrow, A.K. Sen and K. Suzumura
© 2002 Elsevier Science B. V All rights reserved
96 F Aleskerou

6.3. Rules from the Central Class 118

6.4. Rules from the Symmetrically Central Class 120
6.5. Rationality constraints and rules from the Central Class 122
6.6. Remarks and guide to the literature 123
7. Conclusion 125
References 126
Ch. 2: Categories of Arrovian Voting Schemes 97

Abstract

Within the framework of the axiomatic approach three types of voting schemes
are investigated according to the form in which the individual opinions about the
alternatives are defined, as well as to the form of desired social decision. These
types of rules are Social Decision Rules, Functional Voting Rules, and Social Choice
Correspondences.
Consideration is given to local rules, i.e., to the rules which satisfy some analogue
of Arrow's Independence of Irrelevant Alternatives condition.
A general description of the problem of axiomatic synthesis of local rules, and
various formalizations of voting schemes are given.
The notion of "rationality" of individual opinions and social decision is described.
Various types of binary relations (preferences) are introduced. The characteristic
conditions (Expansion-Contraction Axioms) on choice functions are defined, and the
interrelations between them are established.
Two types of Social Decision Rules (transforming individual preferences to social
ones) are studied. The explicit forms of those rules are investigated. The rules restricted
by rationality constraints, i.e., by the constraints on domains and ranges of the rules,
are studied as well.
Functional Voting Rules are investigated which transform individual opinions
defined as choice functions into a social choice function. In doing so, a rationalizability
of those choice functions is not assumed. The explicit form of these rules is obtained,
and the rules which satisfy different rationality constraints are studied.
Social Choice Correspondences deal with the case when the individual opinions are
formalized as binary relations, and the collective decision that we look for is a choice
function. The explicit form of rules is studied. The obtained classes comprise the rules
such as the generalized Pareto rules. Several new classes of the rules are introduced
and analyzed. The explicit form of the Nash-implementable rules is found.
The analysis of publications on the axiomatic synthesis of the local aggregation rules
is made.

Keywords

voting schemes, independence of irrelevant alternatives, locality, social decision

rules, functional voting rules, social choice correspondences

JEL classification:D7
98 F Aleskerov

1. Introduction

Descriptions of voting procedures date from antiquity. Plutarch gave two examples
of voting procedures which were used in the ninth century B.C.
The origin of voting theory traditionally dates back to the end of the eighteenth
century when two members of l'Acadmie Francaise, J.C. Borda and J.M. Condorcet,
proposed procedures of collective choice [Borda (1781), Condorcet (1785)]. Further-
more, Condorcet constructed the following remarkable example demonstrating that the
well-known majority rule could give rise to unsolvable paradoxes.
Let a group of three individuals (electors) evaluate three alternatives - x, y, and z.
The first elector prefers x to y and y to z, (and hence x to z), which is designated
conventionally as x >- y >- z. The preferences of the second and third electors are,
respectively, z >- x - y and y - z - x. If we assume that one alternative is
preferable to another for the group as a whole if it is preferred by at least two (simple
majority) members, we obtain the following collective decision based on the individual
preferences: x >-y >- z >-x, that is, for the group x is better than y, y is better than z,
and z is better than x.
This result (called Condorcet Paradox or Paradox of Voting) gave rise to numerous
studies that continued till the middle of this century and were aimed at avoiding such
paradoxes by using other, more sophisticated procedures. Among the scientists who
proposed their own aggregation rules were Rev. C.L. Dodgson [Dodgson (1873)] better
known as Lewis Carroll and E. Nanson [Nanson (1882)]. The reader interested in the
background of voting theory is referred to Black (1958), Nurmi (1987, 1999) and
Vol'skiy and Lezina (1991).
The next important milestone in the development of voting theory was reached
in 1951 when Kenneth Arrow formulated the problem of aggregation in different
terms and solved it [Arrow (1951, 1963)]. Disregarding any particular rule, he
formulated some conditions to be satisfied by any reasonable aggregation procedure,
and his attempt to describe explicitly a procedure meeting these conditions led to an
absolutely unexpected conclusion - the conditions proved to be incompatible. This
result, often called Arrow S Paradox or the GeneralImpossibility Theorem, has become
a cornerstone of social choice theory, a vast and ever increasing scientific field. In
contrast to the earlier stage in the development of the theory, these works focus on
finding a solution to the problem within the framework of the axiomatic approach
proposed by K. Arrow.
The main condition used by Arrow was called Independence of Irrelevant Alter-
natives. This condition pre-defines the 'local' treatment of alternatives (or pairs of
alternatives, or sets of alternatives, etc.) in voting procedures. The counterpart of his
condition is called below a 'Locality' condition.

i Plutarch's Lives in eleven volumes, 1959 (Harvard University Press, Cambridge, MA) Vol. 1, pp. 239
241, 285.
Ch. 2: Categories of Arrovian Voting Schemes 99

Remaining within the framework of the axiomatic approach and based on the
consideration of local rules, we investigate three formulations of the voting problem
according to the form in which the individual opinions about the alternatives
are defined, as well as to the form of desired social decision. In other words,
we study three models. What is common among them is that in all models
some analogue of the Independence of Irrelevant Alternatives condition is used.
That is why we call the procedures obtained in these models Arrovian voting
schemes.

2. Voting: a general description

The theory of voting studies the following problem: a group N consisting of a finite
number n of voters (n > 2) considers a finite set A of m alternatives (m > 3);
each voter under constraints, the same for all voters, has a sovereign right to have
her own opinion about alternatives from the set A. The problem is how to transform
generally inconsistent voters' opinions to the collective decision which satisfies the
same constraints.
This problem can be studied in two ways. First, one can fix the voting pro-
cedure and study which constraints the procedure under question satisfies. The
other approach suggested by K. Arrow is rather different, and based on the idea
of axiomatic synthesis. In this approach let us introduce three spaces of rules,
elements of which transform individual opinions into collective decisions. The
first space HI contains procedures which transform individual binary relations into
social binary relations. The second space Hj1 consists of procedures which deal
with individual and social choice functions, and the third space H 1 1 , contains
rules which transform individual binary relations into social choice functions.
The rules from H 1 are called Social Decision Rules, the rules from H 11 are
called Functional Voting Rules, and the rules from Hm are called Social Choice
Correspondences.
The rules from these spaces also satisfy some additional constraints which can be
formulated in all spaces in similar terms.
The main condition is that of locality. It is so important and specific to voting
systems that we include this condition in the definition of these spaces. It will be
discussed for each space below, but just to give an idea about the locality condition
let us mention the well-known condition of binariness for the space HI.
In each space some classes can be defined by other so-called normative conditions.
As an example of such a condition let us mention Anonymity.
Let us consider now the other way of determining the classes in the spaces of
voting rules. If voters are restricted to some specific type of binary relations or
choice functions when representing their opinions (e.g., to acyclic relations), then it is
reasonable to constrain the social decision to some specific type of binary relation or
choice function (e.g., the social decision has to be acyclic). Constraints of these types
100 F Aleskerou

are usually called rationality constraints; they lead to the restrictions of the domain
and range of the rules under study.
Now let us formulate the problem of axiomatic synthesis of Arrovian voting
schemes. Let some of the spaces be fixed. It is necessary to find a rule n, or a class
of rules, which satisfies some pre-defined normative conditions and some rationality
constraints.
Remark. This view on the problem of synthesis of voting schemes in different
spaces hi1-H1 m, and the relations among these models are given in Aleskerov (1984,
1997, 1999) and Aizerman and Aleskerov (1987).

3. Rationality of individual opinions and collective decisions

3. 1. Binary relations and choice functions

The problem of aggregation cannot be formulated in precise terms until the

notion of "individual opinion" is formalized. For example, in Arrow's General
Impossibility Theorem individual opinions are represented by linear or weak orders.
The individual opinions can be described either by indicating some relations 2 between
the alternatives from a finite set A (either between their subsets, or between subsets
of alternatives and individual alternatives), or by indicating the alternatives chosen
by an agent from each nonempty subset X C A. The first approach formalizes
an agent's opinion about alternatives as a binary relation P on the set A, i.e.,
P C A x A. The second approach to describe an agent's opinion indicates
directly the subset Y X chosen from each non-empty subset X from A.
The set of all pairs {(X, Y)} defines a choice function denoted by C(.) in what
follows.
The following types of binary relations P are often considered:
* acyclic, i.e., the relation does not contain a cycle xlPx2P PxsPxl of any length s,
1 < s < card(A);
* strict partial orders, i.e., irreflexive (xPx, where P = A x A\P) and transitive (V x,y, z
xPyPz X xPz) relations;
* weak orders (strict partial orders which satisfy the additional condition of negative
transitivity V x,y, z xPy and yPz xPz);
* linear orders (weak orders which satisfy the condition of linearity V x,y either xPy
or yPx).
Hereafter, the set of all binary relations on A will be denoted by B, the set of all
acyclic relations will be denoted by AR, all strict partial orders by SPO, all weak
orders by WO, all linear orders by L(.

2 The concept of relations is based on preference of one alternative to another, on similarity or difference
between alternatives, etc.
Ch. 2: Categories of Arrovian Voting Schemes 101

In what follows we consider the case where a choice function is defined on the
family A of all non-empty subsets of A, i.e., A = 2A \ {0} and

C(.): A 2A with C(X) C X for all X C A.

In some special cases we will consider a choice function to be single-valued, i.e.,

C(.): A - A.

The set (space) of all choice functions on A will be denoted hereafter by C. Generally,
we do not restrict a choice function to be non-empty, i.e., we admit C(X) = 0 for
some 3 X. The subspace of C which consists of all non-empty choice functions will
be denoted by C+. Finally, the subspace of C+ which contains all single-valued choice
functions will be denoted by C.
The use of binary relations in classic choice theory is based on the pre-assumption
of pair-dominance, i.e., by defining the choice of best alternatives as those elements
which "pass" through pair comparisons with respect to some binary relation P. In other
words, one can define a choice function as

C(X) = {y X I x EX: xPy}.

The choice function which can be represented in this form using some P is said to
be rationalizable 4 by P. It will also be called a pair-dominant choice function.

3.2. Expansion-contractionaxioms
A different way to postulate the properties of choice functions is to prescribe how the
chosen set C(-) changes when the set X is varied.
Definition 1. The choice function C(-) is said to satisfy the Heredity condition (H),
if for all X,X'

x' c x = C(X) c(x)

C(X) ';

the Concordance condition (C), if for all X',X"

C(X' UX") D C(X')n C(X");

3 The fact that for some X, C(X) = 0 can be interpreted as refusal from choice.
4 Rationalizability of choice function can be defined by two different (both in sense and use) basic
relations - those of strict preference ("better than"), P, and of nonstrict preference ("at least as good
as"), R. We will use the first type of relations. For more details the reader is referred to Aizerman and
Malishevski (1981) and Aizerman and Aleskerov (1995).
102 F Aleskerow

the Outcast condition (O), if for all X,X'

C(X) C X' C X C(X') = C(X);

Arrow's Choice Axiom (ACA), if

X CX if C(X) = 0, then C(X') = 0,

-if C(X) n X -0, then C(X') = C(X) n;

Path Independence (PI), or Plott's condition, if

C(X' UX") = C(C(X') U C(X")).

The first three conditions are denoted by a, y and in Sen (1970, 1993);
condition ACA was introduced in Arrow (1959) and is equivalent to Weak Axiom
of Revealed Preferences [Samuelson (1938)]; condition PI was introduced in Plott
(1973).
Theorem 1: Generalized Sen's Theorem. There exists a one-to-one mapping between
the classes of choicefunctions pointed out in thefirst and second columns of Table 1.

Table 1
Choice function rationalizable by Domains in C

Linear order ACA in C

Weak order ACA in C-
Strict partial order H n C n O in C+
Acyclic relation H n C in C+
Arbitrary binary relation H n C in C

Theorem 2. The domain of functions satisfying the Path Independence condition

coincides in C with the intersection of H and 0, i.e., PI = H n 0.
Remark. Studies on the problem of rationalizability of choice functions began with the
famous work by Samuelson (1938); a general view on these problems can be found in
Sen (1970, 1993), Suzumura (1974, 1983), Aizerman and Malishevski (1981), Moulin
(1985), Aumann (1986), Green and Shapiro (1994) and Malishevski (1998). For the
complete proof of Theorem 1 and a detailed study of Conditions H, C, 0, and ACA
the reader is referred to Aizerman and Aleskerov (1995). For the case of non-empty
choice, this Theorem was given in Sen (1970), the first two statements are due to Arrow
(1959). The proof of Theorem 2 can be found in Aizerman and Aleskerov (1995).
Ch. 2: Categories of Arrovian Voting Schemes 103

4. Social decision rules

4.1. Locality
Let individual opinions be represented by the totality of binary relations - the voter's
(preference) profile f = {Pl, ..., P,}, and the social decision we look for is a binary
relation P as well.
For simplicity, the relations Pi E f are hereafter assumed to be weak orders.
Let two profiles Vf and Pi be given, and let

V (x,y; ) = {i E N (x,y) E Pi}.

Definition 2. A rule F in II is said to be local (or, satisfies the binariness condition)

if for every pair (x,y) E A x A for which

V (x,y; )= V (x,y; P') and V (y,x; ) = V (y,x; P),

it is true that

(x,y) E P iff (x,y) E P',

where P = F(f), P' = F(P).

Remark. Let a logic variable li(x,y), i E N, be given, taking on one of three values
at each pair (x,y) E A x A:

1, if(x,y) Pi;
li(x,y) = 0, if (x,y) f Pi and (y,x) d Pi;
-1, if (y,x) E Pi.

It is easy to see that the locality of a rule is equivalent to the existence of a boolean
function of the three-valued variables

f (x,y; T(x,y)) =f (l(x,y)., (xy))

taking on the value 1 for (x,y) E P and 0 for (x,y) P.

Representing the rule F in terms of the functionf(x,y; T(x,y)), one may introduce
an Q2-representation (or, list representation) of this rule. For a certain profile {Pi},
that is, for a certain set r(x,y), let the value of the function f(x,y; (x,y)) be 1.
Consider the set V(x,y; P) = {i E N I li(x,y) = 1} of those indices from N for which
the variable li(x,y) equals 1, and the set V(y,x; = {ii)E N I li(x,y) = -1} of those
indices for which li(x,y) equals -1. By definition the sets V(x,y; P) and V(y,x; P)
satisfy the condition V(x,y; A) n V(y,x; P) = 0.
104 F Aleskerou

Let us now put each pair (x,y) into correspondence with a family 2(x,y) of pairs
(V(x,y; P), V(y,x; P)) defined for all feasible profiles Pl (for all sets I (x,y)) by the
conditionf(x,y; I (x,y)) = 1. It is easy to see that the following is true for the local
rule F:

(x,y) F (P) (x,y; P), V (y,x; P) 2(x,y).

e 2) (4.1)

The set of families {Q(x,y)} for various pairs (x,y) together with the above rule,
forms the list representation of a local rule. It follows directly from the definition of
a list representation that each local rule has its own list representation 5 . A stronger
assertion may be made. Let some rule be given in its list form, i.e., let a set of families
y
2(x,y) = {(o Y, WY)} and the rule (4.1) be specified. Since the profile contains only
weak order relations Pi, the sets waxy, c2, , and rz satisfy
<, ' <,tand satisfy somemutual
some mutual
relations of the form

(W2 U ,") n wo u ca c wjy u as.

(02y u ) n ofx u wy-' c c u w2z.

It may be shown that each rule specified in the list form which satisfies these
conditions does satisfy the condition of locality.
Remark. A strong locality condition (often called "Strong Binariness") used in the
literature takes into account only decision of voters about the pair (x,y), and does not
use the information about the pair (y,x). It can be easily seen that in this case list
representation of the rules consists of winning coalitions w.

4.2. Normative conditions

Let us introduce some normative conditions for a rule F:
- Non-Imposition (NI) (Sovereignty): for each (x,y) A x A there exist profiles 5
and P such that (a) (x,y) F(CP); (b) (x,y) ~ F(P').
- Monotonicity (M): let two profiles fP and P be given, and (x,y) F(F). If
V(x,y; P) C V(x,y; '), and V(y,x; P) D V(y,x; A), then (x,y) C F(P').
- Neutrality (Ne): V (x,y), (z, w) [V(x,y; P) = V(z,w; F), V(y,x; 7) = V(w,z; P)]
[(x,y) F(-) X (z, w) F(P)].
- Anonymity (An): for every one-to-one mapping nt: N N, F({P}) = F({P,(i)}).
- Positive Pareto principle (P+): if V i N, (y,x) d P and io: (x,y) Pi, then
(x,y) P.
- Negative Pareto principle (P-): V (x,y), V i C N, (x,y) Pi > (x,y) P.

5 The term "winning coalition" instead of "list representations" is often used. In the description of local
rules in the list form given in Ferejohn and Fishburn (1979), the term "decisive structure" was used.
Ch. 2: Categories of Arrovian Voting Schemes 105

- Positive Unanimity (U+): V (x,y) Vi C N, (x,y) E Pi = (x,y) E P.

- Negative Unanimity (U-): V i G N, V (x,y), (y,x) C Pi = (x,y) P.
Definition 3. The class of rules satisfying the conditions (a) NI n M n Ne n P- and
(b) NI n M nNe n An P, will be called (a) Central Class and (b) Symmetrically
Central Class, in the space HI of all local Social Decision Rules.
These classes will be denoted by AC and Asc, respectively. Two special rules from
Ac are:
- Unanimity rule U: P = i CNPi;
- One vote rule V: P = Ui ENPi.

4.3. Rules from the Central Class

Let us introduce the operation of lexicographic composition of relations, which for

two binary relations P 1 and P2 defines a third one as follows:

xPy X {xPIy or xPly,yPlx, but xP 2 y}.

In symbolic form, this operation will be denoted by *, i.e., P = P1 * P 2- If I is

an indifference relation for a given P (i.e., Ip = A2 \ (P U P-l)), the lexicographic
composition may be written as

P = P * P2 = Pi U (Ip, n P2 ).

The operation * reveals the inequality of agents: if P 1 is the relation representing

the opinion of the first agent, and P2 of the second one, then the social decision P
contains all pairs from P1 as well as those pairs (x,y) from P2 to which agent 1 does
not object (the pair (y,x) being absent in PI).
The operation * can obviously be extended to any finite number of components.
Using this operation let us introduce various rules from the central class Ac.
Definition 4. A local rule F is called a hierarchicalfederation rule if P = F({Pi})
can be defined through the individual relations as

P =Un *Pljq, Pjq {Pi}.

l=lj=l

The hierarchical federation rule defines a social decision as follows: the totality of
individuals is divided into y "blocks", which may intersect; the social decision is
obtained as the union of the independent decisions of all blocks. In turn, each block
decision is obtained as follows: the agents of each block are gathered in a number of
subgroups (numbered as 1, ... , fi), which may intersect as well; the agents in each
106 F Aleskelou

subgroup are ordered by their "ranks", the decision being made hierarchically, and the
decision of a block reflects the unanimous opinion of all its subgroups.
Definition 5. A local rule F is called a hierarchicalrepresentationrule if P = F({Pi})
can be defined through the agents' relations as

p i

= j= I

Denote the classes of hierarchical federation and hierarchical representation rules

as AU n * and AnU *, respectively.

Theorem 3. The Central Class and the classes of hierarchical federation and
hierarchicalrepresentation rules in III coincide, i.e.,

Ac = A u *n = A n u *

A number of particular cases of the hierarchical federation and hierarchical repre-

sentation rules are considered below, and their mutual relationships are established.
Definition 6. A hierarchical federation rule is called a hierarchicalueto rule, denoted
by HV, if it can be represented as

P=, [(Pi,* Pik,) (n*IPi,,)j.

Thus, the hierarchical veto rule is a hierarchical federation in which at least one
hierarchy is included in all intersections, i.e., it has a right of veto. The right of veto
in this case is given to the agents in that group by "seniority".
Definition 7. A hierarchical federation rule is called a hierarchicaloligarchy (denoted
by n*) if in the definition of the hierarchical federation y = 1, i.e., if

P=n*qaojlPJ
j=-

It is interesting to note that the first member in each hierarchical component of the
rule plays a role of vetoer. Indeed, she can "block" any decision (x,y) by choosing
(y,x). On the other hand, to enforce a decision (x,y) all first members in each
hierarchical component should accept the decision, and in this sense these agents create
the oligarchy.
Ch. 2. Categories of Arrovian Voting Schemes 107

Definition 8. A hierarchical representation rule is called a hierarchical syndicate

(denoted by U*) if p = 1.
Thus, for the hierarchical syndicate the rule is written as follows:

P U *q= Pjq.
j=1

Definition 9. A hierarchical oligarchy rule with 3 = 1 (or, equivalently, a hierarchical

syndicate rule for which a = 1) will be referred to as a hierarchicaldictator rule.
In this case

P=*q=IPq, aGn.

The rule classes of hierarchical oligarchy, hierarchical syndicate, and hierarchical

dictator will be denoted by An *, A u *, and A*, respectively.
Consider now the rules from the Symmetrically Central Class.
Definition 10. A rule F E Asc is called
(a) "(kl, k2 )-majority" if the decision (i.e., the pair (x,y) included in P) is made if
n > kl voters vote for it and n2 < k2 vote against it (the rest n - (nl + n2 ) voters
abstain from voting);
(b) "absolute k-majority" if the decision is made if nl > k people vote for it and the
decision does not depend on the opinion of the rest of the voters;
(c) "relative k-majority" if the decision is made if nl > k voters vote for it and the
rest of voters abstain from voting.
The classes of these rules will be denoted by A(kk2) M, kM, and ARkM, respectively.
The (kl,k 2)-majority rule with its subtypes of absolute k-majority and relative
k-majority can be easily interpreted. Thus, using a (kl, k2 )-majority rule, a decision
is made (the pair (x,y) is included in the social binary relation P) if nl > kl agents
vote for this decision and n2 < k2 vote against it (with the remaining n - (nl, n2)
agents abstaining from voting). Of course, when kl < k2 the term "majority" is used
conditionally. In the case of an absolute k-majority rule a decision is made if it is
supported by nl > k agents. The remaining n - n agents may vote against this decision
or abstain. The term "majority" then actually can be used if k > [n/21, i.e. if more
than half of the agents support the decision. Finally, the relative k-majority rule implies
that the decision is made if it is supported by n > k agents while the remaining
n - nl agents abstain (no votes against the decision are permitted). This type of rules
leads to a situation described in Sen (1970). Consider the relative -majority rule.
A pair (x,y) in this case will be included in the social decision P if there is at least
one agent who selects this pair, all others being indifferent to the decision. Then, this
decision privately made by a single member of society becomes the social decision.
108 E Aleskerov

Definition 11. A rule F E ASC is said to be a "r-system of (kl, k 2) majorities" (denote

by (kl, k 2)T -M) if it is a union of r (kl, k2)-majority rules.
For example, if r = 3, then an admissible rule is {(3,2), (4, 3), (7, 6)}-majority, i.e.,
the decision is made if at least 3 voters vote for it, and not more than 2 voters vote
against it; or at least 4 voters vote for it, and not more than 3 people vote against it,
etc.
Theorem 4. Asc = A{(k,k 2 )-M}.
Remark. It is worth mentioning here that strongly local rules from the Central Class do
not have a "hierarchical component" in their definition, i.e., one can obtain federation
and representation rules, veto and collegium rules, oligarchy, syndicate and dictator
rules. For a dictator rule the social decision P coincides with the binary relation of
some agent io, i.e., P = Pi,. An oligarchy rule constructs the social decision P as
unanimous decision of members of some group C N, i.e., P = i E ,,Pi. The class
of dictator rules will be denoted by Ad, and the class of oligarchy rules will be denoted
by A n .

4.4. Rationality constraints and rules from the Central Class

Definition 12. Let Qd, Q, be the sets of binary relations. A rule F will be said to be
a generating rule for a pair of sets (Qd, Qr) if for all Pi E Qd (i N)

F({Pi}) E Qr

holds. The class of all generating rules for (Qd, Q,) is called the complete generating
class (or, briefly, the complete class) for (Qd, Qr) and will be denoted by A(Qa, QJ).
Since only the class W(O of weak orders will be considered below as a domain QJ,
the complete class A(WO, Q,) will be denoted by A(Q).
First we study the rules from the intersection ASC n A(AZ), and then give the
theorem about rules from the Central Class as well.
Theorem 5. With kl < k2, A(A7R)nA(k,'k2)-M = 0 holds. With kl > k2 the intersection
A(AR) n A(kk2)M coincides with the subclass Ak2)M of the class A(kk2)M for
which [kl/k 2 ] > card(A), where [a] is the least integer greater than or equal to a. If
card(A) > card(N), A(AR) n A(k k)-M = ARkM.
Corollary. The intersection of the class A(AZA) with the class of operators "absolute k-
majority" coincides with the class of operators which satisfies the additional condition
k/(card(N) - k)] > card(A). If card(A) > card(N), this intersection consists only of
the operator U.
Remark. For the general case of "r-system of (kl, k2) majorities" the following results
can be obtained:
(a) if there is t such that k < k, then A(A) n Asc = 0;
Ch. 2: Categories of Arrovian Voting Schemes 109

(b) if V t, kf > k2, then A(AR) n AsC = A{(kk2) M}, where m < mint [kl/kj;
(c) if card(A) > card(N) then A(AR) n As c = ARkM.
Now we can formulate the main result concerning local Social Decision Rules.
Theorem 6. The intersection of the Central Class Ac and the Symmetrically Central
Class ASc with complete classes A(Qr) given in the first line of Table 2 coincides
with the classes of rules presented in the cells of Table 2.

Table 2

A(WO) A(SP70) A(AR)

AC A* An * +
l -M
ASC 0 ARkM AMkk 2) }, m < min t [kl/k2l

Remark. The rules from the class Ac n A(AR) marked by the + sign in Table 2, are
described by the following property: the list Q2 in the list representation of the rule
from this class must not contain pairs {w f ', w2J}, j = 1, . . , p, such that
P P
U W1 = Uj
=l j-=1
It can be shown, in particular, that AHC C Ac n A(AR).
For strongly local rules the corresponding results are presented in Table 3.

Table 3

A(WV) A(S'PO) A(AR)

AC Ad An Aun
ASC 0 U AkM

In this Table, A Un is the subclass of federation rules such that the rank6 v(Q2) of
list £2 is greater than the number of alternatives m, i.e., v(Q2) > m; ATM is the subclass
of k-majority rules such that m < [rnkl, where n is the number of agents.

4.5. Remarks and guide to the literature

Remark 1. The statement of Theorem 6 does not mention the class of hierarchical
syndicate rules. The reason is that these rules, as well as hierarchical federation and

6 The rank(or Nakamuranumber) of the list 2 is a natural number v(Q) such that •2 contains v(2) sets,
intersection of which is empty, while the intersection of fewer sets from 2 is not empty. If n,,
w ; 0,
then by definition v(Q) = oo.
110 F Aleskelro

hierarchical representation rules, determine a social binary relation which generally

contains cycles, and this has always been interpreted in social choice theory as an
unacceptable result.
Remark 2. The assumption that the domain of Social Decision Rules consists of
all n-tuples of binary relations of a special type, say, all n-tuples of linear orders,
or all n-tuples of weak orders, etc., plays a crucial role in a theory developed. This
assumption, following Arrow (1951, 1963), is called Unrestricted Domain condition.
Remark 3. Under the condition of strong locality it is not necessary to demand
Neutrality when the range of the rule is the class of strict partial orders or the class
of weak orders. It turns out that

Au' n A(SPO) = A n , Au n A(WO) = Ad.

For the local rules there are some other properties of the same nature - see Aleskerov
and Vladimirov (1986), Aleskerov (1999).
Remark 4. Studies of strongly local Social Decision Rules generally use the
restriction Qd c Q, i.e., any individual relation is admissible as a social one. It turns
out that violation of it usually violates the Non-Imposition condition.
Remark 5. Another constraint on the domain of a local rule single-peakedness -
was introduced by Black (1958). It was proved that a simple majority rule preserves
all axioms needed with that restriction on the domain Qd. The problem of preserving
transitivity (or other necessary rationality constraints) when the majority rule (or, in
general, local rule) is used, was studied in many publications [see, e.g., Sen and
Pattanaik (1969)]. Since this property can not be preserved with Unrestricted Domain,
restrictions on the agents' binary relations were developed.
Remark 6. A very interesting model in which the locality condition has been
reformulated for the case of hyper-relations was considered in Naiton and Watanabe
(1995).
The studies of local rules in H11 were initiated by the classic monograph by K. Arrow
(1951, 1963). He first proved the theorem which in our notation is expressed as follows:
with AI > 2, INI > 2

AC n A(WVO) = Ad.

In the framework of strongly local rules the result

AC n A(SPO) = A'n ,

was obtained independently in Gibbard (1969), Guha (1972), Mas-Colell and Son-
nenschein (1972). This result shows that Arrow's Paradox can be weakened when the
range of the rule is the class of strict partial orders.
Ch. 2: Categories of Arrovian Voting Schemes 111

The Veto Rule and Collegium rule were introduced by Brown (1975) along with the
proof of the theorem

AC n A(A1R) = AVR

Federation and representation rules and the theorem

u n
Ac = = Anu

were first obtained in Mirkin (1979). He studied the problem of the aggregation of
equivalences as well.
Wilson (1972) showed that if in the formulation of Arrow's General Impossibility
Theorem the monotonicity condition (Pareto principle) is omitted then the anti-dictator
rule arises, i.e., social decision P coincides with the inverse relation Pl of some
agent io, Pio'= {(x,y) I (y,x) E Pi}. Strictly speaking, Wilson (1972) showed that

ANI n ANe A(WO) = Ad U A d -

Very detailed studies of rules in Hi were made in Monjardet (1978), who elaborates
the problem of tournament aggregation, and gives the characterization of the class AU n
in the algebraic form.
Acyclicity of social decision constructed by strongly local rules was studied by
Nakamura (1975); see also Moulin (1988). A broad exposition of results concerning
strongly local rules is given by Kelly (1978). New constraints on the range of the
strongly local rules, namely, -acyclic and single-source relations were studied by
Aizerman and Aleskerov (1983b) and Aleskerov (1999).
Local rules were first studied by Fishburn (1974, 1975) and Schwartz (1980).
Schwartz showed that the class Ac n A(W/O) consists of weak dictators, which was
later understood to be the first by seniority agent in the Hierarchy rule. Fishburn proved
the theorem

Ac n A(WO) = A*.

In Danilov (1982, 1985) it was first proved that

AC n A(WO) = An *

and the problem of aggregation of the equivalence relations and relations of tolerance
(reflexive and symmetric relations) using local rules was studied.
In the exposition of the results concerning local rules in H1 I followed mainly
Aleskerov (1985a) and Aleskerov and Vladimirov (1986).
Levchenkov (1987) obtained another representation of local rules using algebraic
polynomials.
112 F Aieskerou

The use of three-valued logic functions to study the problem of social choice is due
to Murakami (1968). In Makarov, Vinogradskaya, Rubchinskiy and Sokolov (1982)
this approach was used widely to describe different choice problems.
Vladimirov (1987) showed that the locality of a rule in HI is equivalent to the
existence of a boolean function Cp(x,y) of 2n variables, which is constructed as follows:
let Z be a vector of 2n variables such that Z = Z(x,y, V), and define components of
Z, V i N, by

z2i- , (Z) = 0 X (y, X) E Pi.

The function cp is defined as follows

p(x,y)(Z) = 1 X (x,y) C F().

Sholomov (2000) studied the aggregation problem using the representation of local
rules via boolean functions of three-valued variables. In particular, he seems to be the
first to obtain the following theorems

AC n A(SO) = A*, and AC n A(ZO) = A*,

for the classes of semi-orders SO, and interval orders ZO. He also characterized
the class of weighted majority rules for monotonic and non-monotonic cases of local
rules.

5. Functional voting rules

5.1. Locality
In the Functional Aggregation Rules category individual opinions as well as social
decisions are presented as choice functions. A mapping F of profile {Ci(.)} into a
social choice function C(.) is called a Functional Voting Rule. A rule F is defined
over the n-tuples of functions from C, and its values are the functions from C, i.e., F
is the mapping

F: C= C x ... x C -- C.

Definition 13. A rule F is said to be local if for every two profiles C, C' and arbitrary
X e A and x E X for which

V i Nx Ci(X) xE C, (X),
it is true that

x E C(X) X x E C'(X).
Ch. 2: Categories of Arrovian Voting Schemes 113

As usual we define the rules U (unanimity), as V X E A

c(x)= n ci(x);
iEN

and V (one vote), as V X · A

C(X)= U Ci(X).
iGN

The representation of local FVRs using the winning coalitions approach is defined
similarly for the case of strongly local Social Decision Rules with only one
difference that winning coalition Wo(x-) is defined for an alternative x and set X. The
representation of local rules through logic functions can be defined analogously.

5.2. Normative conditions

We now introduce a number of normative conditions on FVRs. The conditions
introduced below are counterparts of the conditions introduced above for Social
Decision Rules.
- Non-Imposition (NI): for all X E A and all x from X there exists a profile {Ci(.)}
such that x E C(X), where C(.) = F({Ci(.)}i GN); and for all X c A and all x from
X there exists a profile {C()} such that x X C(X);
- Monotonicity (M): let two profiles {C(.)} and {Ci(.)} be given and x e C(X) be sat-
isfied for some X E A and x G X. Denote V(x,X; {Ci(.)}) = {i E N I x E Ci(X)}.
Let V(x,X; {Ci(.)}) C V(x,X; {C.(.)}). Then, x · C(X);
- Neutrality (Ne): let two profiles {C'(.)} and {Ci'(.)} be given and let the following
assertion hold for all X',X",x',x" such that x' X' and x" X": V i N
(x' G C(X') x" E C(X")). Then, x' C'(X') if and only if x" E C"(X"),
where C'(.) = F({C(.)}) and C"(.) = F({C'"(.)}). Condition Ne consists of
the two conditions: Context (presentation X) independence, and Alternatives (x)
independence.
- Anonymity (An): let be any one-to-one mapping of N to N. Then, F({C,(.)} =
F({Ci(i)(')}).
Definition 14. The class in H 1 where conditions NI, M and Ne are satisfied
simultaneously, will be referred to as the Central Class; the class satisfying the all
four conditions (NI, M, Ne, An) will be referred to as the Symmetrically CentralClass.
These classes will be denoted by Ac and ASC, respectively.

5.3. Rules from the Central Class

Some explicit cases of rules from Ac are introduced in the following definition.
Definition 15. The rule F is said to be a
(a) Federation, if C(X) = Us= l Pi C i(X);
114 E Aleskerou

(b) Representation, if C(X) = il Ui, ci();

(c) Oligarchy, if C(X) = ni i~ Ci(X);
(d) Syndicate, if C(X) = Ui, l Cj(X);
(e) Dictator, if C(X) = Ci(X);
(f) k-majority, if the sets coj in the definition of the operator "federation" contain all
k-element subsets of N.
The corresponding classes of operators will be denoted by Au , A n ,, A", AU , Ad
and AkM, respectively.

Theorem 7. Ac = A u = AnU; ASC = AkM.

5.4. Rationality constraints and rules from the Central Class

Definition 16. The class Q c C is said to be closed with respect to a rule F, if V C

such that V i E N C() Q F(C) C Q holds. The class Q is said to be closed with
respect to the class Y of rules F, if Q is closed with respect to each F. The set of all
rules relative to which Q is closed is called a complete class of closedness (briefly,
complete class) and denoted by A(Q).
Below we consider the complete classes for domains H, C and O and their
intersections. The following general result holds.
Theorem 8. The intersection of the Central Class Ac and the Symmetrically Central
Class Asc with complete classes A(Qr) given in the first line of Table 4 coincides
with the classes of rules presented in the cells of Table 4.

Table 4

A(H) A(C) A(O) A(HnC) A(H n ) A(CnO) A(H n C n ) A(ACA)

AC Aun An AU An AL Ad Ad Ad

ASC AkM U V U V 0 0 0

Thus, no local rule satisfies the set of normative conditions NI -An with respect to
which the domains of classically rational choice functions (H n c n O or ACA) are
closed. It implies that, if all agents use some classically rational functions, there exists
no local rule satisfying those conditions whose application would guarantee classical
rationality of the social choice function. This result is a direct counterpart of Arrow's
General Impossibility Theorem.
At the same time, the domain H is remarkable in the following sense: if the agents
are restricted in their choice of functions C() only by the condition H, no difficulty
arises with construction of the desired rules: social choice functions belonging to H
Ch. 2: Categories of Arrovian Voting Schemes 115

are constructed for any rule satisfying the characteristic conditions NI-An and any
profile from H. Moreover, such rules are k-majority rules.
As follows from the aforesaid, as long as consideration is confined only to local
voting rules, and the conditions NI-An are regarded as necessary, the classical
rationality requirements should be abandoned, and one should accept as reasonable
all the choice functions from the domain H in C. Generally speaking, this violation
of the conditions C and O is necessary only in reference to the choice function C(.)
generated by the aggregation rule, whereas the individual choice functions Ci(.) can
be arbitrarily limited by any subdomain of H (e.g., even ACA C H).

5.5. Remarks and guide to the literature

Remark 1. If the monotonicity condition is omitted then broad classes of operators
can be obtained including those which are defined with "anti-component", i.e., anti-
dictator, anti-oligarchy, etc. The complete study of this case is given by Aleskerov and
Duggan (1993).
To mention one non-monotonic result let us introduce the condition C which is
dual to the condition C of Concordance:

C-: YC XI X 2,C(X)n Y = C(X2)n Y= XC(XUx,,2 )n Y =

We denote by AN lnNe the class of rules which satisfy non-imposition and neutrality.
Theorem 9. ANI n Ne n A(C U C-) = Adu d , i.e., the intersection of the class ANInNe
with the complete class of closedness for the domain C U C- coincides with the class
Ad u d which consists of the operators "dictator" and "anti-dictator".

Theorem 9 is a complete analogue of Wilson's theorem [Wilson (1972)] for functional

voting rules.
Remark 2. Throughout Section 5, it was assumed that Qd C Q,, i.e., any kind of
choice functions used by agents is admissible as social choice function. However, the
question arises what kind of rules can be obtained if this restriction is omitted, i.e., the
domains Qd and Qr are not enclosed. The answer to that question given in Aleskerov
(1985c) is that if Qd t Qr, then in practically all cases such mappings violate Non-
Imposition.
In contrast to the Social Decision Rules there are only several publications in which
local functional voting rules have been studied.
An early article on local FVRs was Parks (1976), but in that work the problem of
axiomatic synthesis wasn't even stated. The problem of axiomatic synthesis of local
FVRs was stated in Aleskerov (1983) and Aizerman and Aleskerov (1983a). In the
latter article practically all results concerning the Central Class of local FVRs were
obtained. Detailed study of those rules was done in Aizerman and Aleskerov (1986),
Aleskerov (1999). Classes of rules under different kind of rationality constraints with
and without normative restrictions were studied in Aleskerov (1985c). The conditions
116 F Aleskel-o

which guarantee the non-emptiness of local FVRs were obtained in Popov and El'kin
(1989). It is worth mentioning that in that article a more restricted version of FVRs
was considered - the rules were defined only on A, not on the subsets of A. Non-neutral
but monotonic FVRs were the subject of study in Aleskerov (1985c) and Aizerman
and Aleskerov (1986, 1995). Non-monotonic neutral rules were studied in Aleskerov
and Duggan (1993).
In Stefanescu (1997), the above results for the local Functional Voting Rules are
generalized to the infinite case.

6. Social choice correspondences

6.1. Locality
In this Section we consider the space H111, i.e., rules which transform individual binary
relations to social choice functions. For simplicity, the binary relations Pi, i = 1, ... n,
are assumed to be linear orders. Thus, rule F is defined as

F: £" - C.

These rules are called Social Choice Correspondences.

Definition 17. The upper contour (or, dominant) set D(x) for the alternative x in the
binary relation P is defined as

D(x) = y e A I (y,x) e P}.

For the moment, we will not restrict the set C, i.e., all choice functions are admissible
for collective decision.
Definition 18. A SCC F: CO" -- C will be said to satisfy the locality condition if for
any two profiles V,P and any x,X, x E X E A, such that

Vi E N X n ;Di(x)
=X D'(x)

it is true that

x E C(X) if and only if x e C'(X),

where C() = F(P) and C'(.) = F(P').

It can be proved that the weak Paretian rule Fpar:

C(X) = {x E X I y X such that V i e N yPix}

satisfies the locality condition.

Ch. 2: Categories of Arrovian Voting Schemes 117

On the other hand, the Borda rule is not local.

Now let us construct the list representation for local Social Choice Correspondences.
The totality of sets {2}, Z = (Z, ... , Z,,) such that V i N Zi C X\{x} is called
a list for the pair (x,X), x E X E A. The list for pair (x,X) will be denoted by
Q(x,X): 2(,X)= ({2}).
The Q- (or, list) representation of a rule F E HIr is defined by the set of lists
OF = {•2(x,X)}(xx) and the rule for any profile If determines a social choice on any
pair (x,X) as follows:

x E C(X) X (X nDi(x)..., X D,(x)) Q(x,X) 2F.

Let us emphasize that in contrast with previous cases of the local rules in H 1
and HI, for which lists in Q2-representation contain winning coalitions or pairs of
coalitions, i.e., subsets of the set of agents N, in the case of local rules in Hi, lists in
Q-representation are subsets of the set of alternatives A.
Remark. Let us assume that agents' binary relations are represented by more
"reach" type of binary relations, say, by weak orders. Then the weakly local rules can
be defined generalizing local rules - to make a decision about inclusion (or exclusion)
of an alternative in the social choice from X, one should check not only upper but
also lower contour sets in agents' binary relations.

6.2. Normative conditions

The normative conditions for SCCs include the following.
- Non-Imposition (NI): for all x,X (x X A) there exists a profile V such that
x C C(X), where C(.) = F(If), and there exists a profile P such that x C'(X).
The following condition of monotonicity was introduced using the lower contour sets
in Maskin (1977).
- Maskin's Monotonicity (MM): if V and P are two profiles, such that V i N
DI(x) n X C Di(x) n X, then, x e C(X) X x E C'(X).
It is obvious that the condition of monotonicity is a strengthening of the condition of
locality, i.e., a monotonic rule is local.
- Neutrality (Ne): this condition is divided into two conditions:
(a) Independence of alternative (of x) (Nex): suppose for two profiles P> and A,
and x and y in X it is true that V i,X n Di(x) = X n D'T(y). Then y E C(X)
iffx C C'(X);
(b) Independence of context (of the subset X) (Nex): suppose for two pro-
files V and P, and two sets X and X' in A(= 2 A\{ 0}), that V i,
Di(x) n X = D;I(x) n X'. Then x C(X) iffx C C'(X').
Conditions (Nex) and (Nex) are stronger than the locality condition: put
x = y in Nex or X' = X in Nex. The Neutrality condition, as usual,
guarantees that the alternatives are treated equally by the rule (even in different
contexts X).
118 E Aleskerov

- Anonymity (An): let : N >N be a one-to-one mapping from the set N to N. Then
C(.) = C'(.) where C(.) = F(PI, . .. , P,) and C'(.) = F(P(l),... P(,)).
- Non-Dominance (ND): this condition is divided into two following conditions:
(a) Positive Non-Dominance (ND-): for all x,X (x X A), if there exists
io N such that Dio(x) nx = 0, then x C(X).
(b) Negative Non-Dominance (ND-): for all x,X (x X E A), if V i C N
Di(x) nX 0, then x C(X).
- Positive Unanimity (U+): if for some x,X (x E X E A), and for all
i N Di(x) nX = 0, then x E C(X).
- Negative Unanimity (U-): if for some x,X (x X e ) i N such that
Di(x) n X 0, then x C(X).
- No Veto Power (NVP): Let

card({i I X n Di(x) = 0}) = n - 1.

Then x C C(X).
The NVP condition states that if n - I agents evaluate some alternative x as their best,
the last agent cannot "veto" this decision, i.e., x will be socially chosen. This condition
was first introduced by Maskin (1977).
It can be shown that the weak Paretian rule belongs to the class NI n MM n Ne n
An n ND+ n U+ .
Two special rules are:
-Unanimity rule b: C(X) = {x E X V i E N X n Di(x) = 0};
- One vote rule V: C(X) = {x C X I io N Dio(x) X = 0}.
The class in H111 satisfying simultaneously the conditions NI, MM and Ne will be
referred to as the Central Class; the class satisfying NI, MM, Ne and An will be
referred to as the Symmetrically Central Class. These classes will be denoted by Ac
and ASC, respectively.

6.3. Rules from the Central Class

Representation of the rules from the Central Class is established below. Because the
variety of the rules from the class AC is very large, we begin with particular cases,
and then give generalizations.
Consider the following rule:

.T(N, 0): V x,X x C C(X) n (x n Di(X))a =0.

iEN

In words, x belongs to the social choice C(X) if the number of alternatives which are
more preferable than x in each Pi is null, or, in other words, there is no alternative y
which is more preferable than x for every agent. One can see that this rule is exactly
the weak Paretian rule Fpar introduced above, i.e., Fpar = oT(N, 0).
Ch. 2: Categories of Arrovian Voting Schemes 119

To generalize, consider

r(I,q'): x C(X) = n (xnDi(x)) < q ,

i.e., the alternative x is chosen even if there are at most q alternatives which are
preferred to x for every member of coalition I. This rule will be called a partial
q-Pareto rule.
Let us construct a rule, called coalitionalq-federation rule, by the same "scenario"
which we used before: for a given totality of sets of coalitions we consider q-Pareto
optimal elements unanimously chosen in each set of coalitions, and take the union of
these elements over the sets from the given totality of sets. It is important to note that
the parameter q' for each coalition in each set can be different. Stated formally, we
obtain

F =U
t=l I
nI,
7(Iq'

The class of these rules will be denoted by Aq un

Consider now the rules introduced above with extremal values of parameter q. The
following equalities are obvious: V i and V I, if q > m - 1 then

({i}, q)= Jr(, q)= 1;

if q < 0 then

7({i}, q) = 7(I, q) = 0,

where 1 is the constant rule which defines for all profiles C(X) = X for all X; 0 is the
constant rule which, for all profiles, defines C(X) = 0 for all X.
All the rules just introduced are local and satisfy the conditions of Non-Imposition
(if 0 < q < m - 1), Maskin's Monotonicity and Neutrality. So, these rules belong to the
Central Class Ac. Varying coalition I and number q in the definition of these rules,
one can obtain the class of rules, that depends on I and q.
Theorem 10. Rule F belongs to the Central Class Ac if f there exists an integer
s > O, a set of coalitions ,, t = 1, ... , s, the integers {q'}, such that for all X, V I,
0 < q < m - 1, andfor all i

0 n r(I,q).
F=U
t I I E It

That is, any rule from Ac can be represented as a coalitional q-federation rule. In
turn, any coalitionalq-federation rule is a rule from the Central Class.
120 E Aleskerou

Stated differently, the above Theorem establishes that Ac = Aq n.

At the end of this Section we study the following problem: what is the form of a
coalitional q-federation rule which satisfies No Veto Power (NVP)? NVP states that the
unanimous decision of any n - 1 agents concerning their best alternative (top element
in their linear orders) should be in the social decision. Consider the rule

o= U nm (i, o),
I: card(l) = n- iE

which constructs social decisions exactly as unanimous decisions of any of n - 1 agents

about their top elements. Then, obviously, the coalitional q-federation rule which
satisfies NVP can be written as

F = OU O',

where (9' is any rule from the Central Class.

For instance, it can be the rule "q-dictator" ("dictator" if q = 0)

O' = T({i0}, qiO).

In this case, the social decision consists of q" top elements of agent i0, and the common
top element of any of n - 1 agents if such an element exists. Thus, this result establishes
the form of the Nash implementable SCCs [Maskin (1977)].

6.4. Rules from the Symmetrically Central Class

Let us study coalitional q-federation rules which satisfy condition An as well. A rule
satisfies condition An if the families , t = 1, ... , s, are formed in a special way.
In contrast to Social Decision Rules and Functional Voting Rules, the present case
has two "degrees of freedom" - the size of coalition I (the only one to be varied in
previous cases), and the structure of the families Z,. Consider a system of r families
of subsets of N

2=(l ... ,5-)J

such that Jj (J .. J ), where , =ki, = , ., rj, and rj = (k), i.e., each

family Gj contains all coalitions of size k.
Let HO= {1, ... , ri, and construct the family of sets

3(0) = (3:,, , 3c,(09)

such that each 3j, = 1, .. , p(), contains t-majority families of indices from Hi,
and p(0/) = (j)-
Ch. 2: Categories of Arrovian Voting Schemes 121

Consider
r P
F=UU n (j,q').
j=l =1 E3

We call this rule a -system of (t, k)-majority families of q-Pareto rules. The
corresponding class of rules will be denoted by A q, -(t,k)M
Let us illustrate the rule via an example.
Example. Let N = {1,2, 3}, = (), and J1 contains all single-valued coalitions
1 = ({1}, {2}, {3}). Let 3(01) = (1, ... , Cp) with t(0 1) = 2. Then p = (3) = 3, and

J(01) = (al, a2}, {al, a3}, {a2, a3}),

where al = {1}, a2 = {2}, a3 = {3}.
The rule is written as follows:

F = (({ 1}, q) nft({2}, q)) U (({ 1}, q) n r({3 }, q)) U (Jr({2}, q) n ({3}, q)).

In words, this rule chooses common top q + 1 alternatives of the first and second
agents, or common top q + 1 alternatives of the first and third agents, or common top
q + 1 alternatives of the second and third agents. ·

Theorem 11. A s c = Aq, T-(t,k)M, i.e., the Symmetrically Central Class coincides with
the class of r-system of (t, k)-majority families of q-Pareto rules.
Four special cases of these rules are of interest. Consider first the weak -majority
q-Pareto rule, i.e., the rule

F= U :r({i},q).
iEN

It is important to note that the parameter q is the same for each i. This rule will be
called a q-one-vote rule. Note also that 0-one-vote rule is just a one-vote rule which
chooses undominated alternatives in all ith preferences.
The class of q-one-vote rules contains m - 1 rules which correspond to different
values of q, q = 0, 1, ... , m - 2. This class of rules will be denoted by qV, while
one-vote rule is denoted by V.
Let us study strong n-majority q-Pareto rule, i.e., the rule

F= n T({i}, q).
iEN

This rule will be called a q-unanimity rule (but not qi-unanimity), and in case q = 0,
it is unanimity rule. It chooses maximal elements common to all agents. Obviously,
such alternatives may not exist.
122 F Aleskerou

The class of q-unanimity rules contains m 1 rules for q = 0, 1,..., m - 2. This

class will be denoted by qU/, while unanimity rule is denoted by U.

6.5. Rationality constraints and rules from the Central Class

Until now, we did not restrict the range of Social Choice Correspondences under
study, although the domain of the rule was fixed, i.e., the rule F was defined over
the n-tuples of linear orders from LC, and its values were the functions from C, i.e.,
F was considered as the mapping F: LO" - C.
Definition 19. The set of choice functions Q c C will be said to contain the image
of a rule F c HII1 if for any profile {Pi} such that for all i N Pi E LO,
F({Pi}) = C() E Q holds. The set Q c C will be said to contain the image of
the class of rules F if Q contains the image of each rule F E F. If F consists of all
rules such that Q contains their images, then F will be called the complete class for
the set Q and denoted by A(Q).
Next, we study an important rationality constraint on social choice correspon-
dences - single-valuedness of the social choice function. The fundamental result
for single-valued monotonic SCCs is Muller-Satterthwaite's Theorem [Muller and
Satterthwaite (1977)]. Let us recall that the class of single-valued choice functions
is denoted by C; hence the complete class for C' is denoted by A(C).
Theorem 12. Let card(A) > 3, and F: L0 --* C be a single-valued SCC which
satisfies Non-Imposition and Maskin s Monotonicity. Then F is dictatorial, i.e.,
ANI n AMM n A(C) C Ad.
Remark. The single-valuedness ofF permits to consider a set W of agents' coalitions,
i.e., to use even in this third model "an apparatus" of winning coalitions. This shows
deep connections between all three models
- strongly local Social Decision Rules. In this case, transitivity of social binary relation
guarantees that the rule under study is neutral;
- Functional Voting Rules when social decision is searched for in the classes of choice
functions satisfying the Outcast (0) condition. In this case, the fact that F belongs
to A(O) guarantees neutrality of the rule; and
- single-valued Social Choice Correspondences. In this case, neutrality of the rules
are obtained as well.
As range constraints we consider next the classes of choice functions determined
by Expansion-Contraction Axioms - those of Heredity (H), Concordance (C),
Outcast (0), and Arrow's Choice Axiom (ACA).
It turns out that rules from the Central Class generate choice functions which belong
to the domain H.
Theorem 13. AC C A(H), i.e., any rule from the Central Class generates choice
functions which satisfy condition H.
Ch. 2: Categories of Arrovian Voting Schemes 123

As a corollary to previous Theorems 10, 11, and 13 we obtain

Theorem 14. The intersection of the Central Class with the complete class A(H)
coincide with the class of coalitional q-federation rules, i.e., AC n A(H) = AqUn;
the intersection of the Symmetrically Central Class with the complete class A(H)
coincide with the class of r-system of (t, k)-majority families of q-Pareto rules, i.e.,
Ac n A(H) = Aq,r- ( t,k)M
To formulate the next Theorem we need to define some special subclasses of
coalitional q-federation rules. The rule F = Nl c rZ(I, 0) will be called the coalitional
oligarchy rule, the rule r(I, 0) will be called the partial Pareto rule, and the
rule 7r({i}, 0) will be called the dictator rule. The classes of these rules will be denoted
by A n , A P, and Ad, respectively.
Theorem 15. (1) Ac n A(H n C) = An , i.e., the intersection of the Central Class
with the complete class for the domain H n C coincides with the class of coalitional
oligarchy rules. (2) The intersection of the Central Class with the complete class for
the domain H n C n 0 coincides with the class of partialPareto rules, i.e.,

AC n A(H n C n O) = AP;

(3) The intersection of the Central Class with the complete class for the domain ACA
coincides with the class of dictator rules, i.e.,

AC n A(ACA) = Ad.

Since any choice function from H n C n O is rationalizable via strict partial order,
and any choice function from ACA is rationalizable via weak order (Theorem 1), the
following corollaries to Theorem 15 can be obtained.
Corollary 1. A social choice function constructed from an arbitrary profile by a
rule from the Central Class in H17i is rationalizable via strict partial order iff the
corresponding rule is the partial Pareto one.
Corollary 2. A social choice function constructed from an arbitrary profile by a rule
from the Central Class in H17l is rationalizable via weak order iff the corresponding
rule is the dictator one.
This last Corollary is a direct counterpart of Arrow's General Impossibility Theorem
which was obtained under a different condition of locality - that of Independence of
Irrelevant Alternatives.

6.6. Remarks and guide to the literature

Remark. There are not many publications concerning the axiomatic synthesis of the
rules from the space H
1 1, although indirectly these rules were intensively studied in
the framework of the implementation theory. The concept of Nash implementability
124 R Aleskerou

of SCCs was first introduced by Maskin (1977). In this framework the rules themselves
are not described - the problem is to find the conditions which the rules must satisfy to
be Nash implementable. Maskin (1977) showed that the Monotonicity condition (MM)
is necessary, and both MM and No Veto Power (NVP) are sufficient for a SCC to be
Nash implementable.
The next important step in the implementation theory was made by Moulin and Peleg
(1982) with the introduction of the notion of implementation in strong equilibria. In
this case, not only does any single agent have no incentive to deviate from her sincere
preference, but also no coalition can be better off as a result of unanimous deviation
of its members. Again, Maskin's Monotonicity turns out to be a sufficient condition
for the implementability of social choice correspondence in strong equilibria.
In publications in this field, no attention has been drawn to the problem of
representation of procedures implementable via Nash or strong equilibria. Rather, the
aim has been to establish conditions to which the implementable procedures must
satisfy.
It looks like from that period, i.e., from 1977 on when E. Maskin formulated
the problem of implementation in Nash equilibria, studies on these two important
problems - an axiomatic synthesis of SCCs and implementability problem - have been
developed in parallel and no intersection between them has been established. The first
and seemingly unique attempt to connect implementation theory with the problem of
axiomatic synthesis of Arrovian voting schemes was made by Plott (1984). In that
work, the Independence of Irrelevant Alternatives condition was discussed in general
form, and it was shown that any implementable SCC must satisfy the IIA condition.
The model of local social choice correspondences considered above was first
formulated in Aleskerov (1985b) in the framework of multicriterion choice, or, strictly
speaking, in the case when instead of binary relations Pi ordinal utility functions
are considered. In that work a condition of weak locality (more general than that
of locality) was formulated but attention was given only to the rules from the
Symmetrically Central Class. In Aleskerov (1991) weakly local SCCs were studied
further. Practically all main results obtained above can be found in Aleskerov (1992). In
Aleskerov (1992) the main representation Theorem 10 was formulated, which gives a
complete description of the form of Nash implementable voting schemes. In Aleskerov
(1993, 1994, 1997, 1999) further generalizations of the results dealing with rules from
/Ill were obtained.
In publications dealing with the rules from I7i1, there are several studies in which
locality of rules means the satisfaction of Independence of Irrelevant Alternatives
in general form. At the same time, the social choice function is restricted by some
additional condition, e.g., by Arrow's Choice Axiom (ACA); however, the social choice
correspondence is defined not on the set A but on its subsets of fixed cardinality.
Nevertheless, the existence of a dictator is obtained [see, e.g., Grether and Plott
(1982)]. In Blair, Bordes, Kelly and Suzumura (1976) the condition of locality for
Ch. 2: Categories of Arrovian Voting Schemes 125

SCCs was formalized in a pair-wise form, and the existence of a dictator in some
form was established.
In Denicol6 (1987) [see also Peris and Carmen Sanchez (2001)] "a bridge" between
local social decision rules and independent (local in a special sense) social choice
correspondences was constructed. In that model to check the fact that y is not included
in the social choice on A one should check pair-wise relations between x and y in both
profiles Pl and P. In other words, the locality concept is based on the idea similar to
that for Social Decision Rules. In this model, in particular, the existence of a dictator
has been established.

7. Conclusion

The Independence of Irrelevant Alternatives axiom is the key concept in the problem
of axiomatic synthesis of voting rules introduced by K. Arrow. It separates local rules
from all other rules, including such non-local rules as the Borda rule, the Approval
Voting rule, etc.
The main advantage of locality is that it allows to decrease the informational
complexity of decision making, which was pointed out in Arrow (1963) for the first
time. Indeed, for the Social Decision Rules to make a decision about the inclusion of
a pair (x,y) to the social decision it is necessary to check 7 in general n _m2 pairs, i.e.,
in total to construct the social decision it is necessary to make n - m3 comparisons. If
we consider the case of strongly local Social Decision Rules, only n m2 comparisons
should be made to construct the social decision, and in the neutral case this number
decreases to just n.
For the Functional Aggregation Rules in general cases one should check n · m2 ·22m
pairs of the type (x,X). In local cases, only n m 2m comparisons should be made.
For the case of the Social Choice Correspondences to make a decision about the
choice of an alternative x from a set X n - m2 · 23m are to be made. For the local case,
however, only n m2 2m comparisons are enough to obtain the social decision.
The use of locality condition allows to construct a complete theory of local
(Arrovian) voting rules. It has been done above for three different aggregation models:
for Social Decision Rules, for Functional Aggregation Rules and for Social Choice
Correspondences. In this construction we also separated the normative conditions and
rationality constraints which the rules under study should or may satisfy.
The study of the rules from the Central Class (satisfying usual normative conditions
of Non-Imposition, Monotonicity and Neutrality) shows that they possess the similar
structural properties and can be described in a similar way: the social decision is

7 The comparison of local and non-local rules is made below under the assumption of the same
coalitional structure of the rules. The evaluations given are achievable upper bounds. More accurate
evaluations can be made; however, that is outside the scope of this chapter.
126 F Aleskerou

constructed as the union of the unanimous decisions of agents in designated groups. For
the first two models these unanimous decisions are straightforward - they are defined
as the intersection of individual opinions. In the third model such decision is made in
a more complicated way - it coincides with the set of alternatives chosen by q-Pareto
rule for some coalition of agents.
The rationality constraints are formulated either in terms of the properties of binary
relations (for Social Decision Rules, for instance, acyclicity of the social binary
relation) or in terms of the properties of choice functions (for Functional Aggregation
Rules and Social Choice Correspondences, for instance, the social choice function
should satisfy the Heredity condition). It turns out that the rules from the Central Class
satisfying most restrictive rationality constraints (the social decision is a linear order
for Social Decision Rules, and the social decision satisfies Arrow's Choice Axiom
for Functional Aggregation Rules and Social Choice Correspondences) are dictatorial
rules. These results are direct counterparts of Arrow's Impossibility Theorem for the
three models under study. At the same time, less restrictive rationality constraints lead
to very rich classes of rules in these models.
In contrast with local (Arrovian) rules, the complete theory of axiomatic synthesis
for non-local rules is far from completeness. It is worth mentioning that there
are several important publications concerning non-local rules [see, e.g., Brams and
Fishburn (1982), Chebotarev (1994), Ilyunin, Popov and El'kin (1988), May (1952),
Moulin (1988)]. The analogues of the normative conditions used above can be applied
to non-local rules, as well as the corresponding rationality constraints. The problem is
to define a "narrow" notion of a non-local rule to construct a complete theory of such
rules. However, it is still an open problem.

References

Aizerman, M., and F Aleskerov (1983a), "Local operators in models of social choice", Systems and
Control Letters 3:1-6.
Aizerman, M., and F Aleskerov (1983b), "Arrow's problem in the theory of social choice", Automation
and Remote Control 9:127-151.
Aizerman, M., and F.Aleskerov (1986), "Voting operators in the space of choice functions", Mathematical
Social Sciences 11(3):201-242.
Aizerman, M., and F Aleskerov (1987), "Structural properties of voting systems", in: M. Aizerman and
E. Caianiello, eds., Topics of General Theory of Structures (Reidel Publishing Company, Amsterdam)
pp. 137-149.
Aizerman, M., and E Aleskerov (1995), Theory of Choice (Elsevier, Amsterdam).
Aizerman, M., and A. Malishevski (1981), "Some aspects of general theory of the choice of best
alternatives", Automation and Remote Control 2:65-83.
Aleskerov, E (1983), "Principles of mutually exclusive neutralities in Arrow's social choice problem",
Preprints of IX All-Union Congress on Control Sciences (VINITI, Moscow) pp. 171-172. In Russian.
Aleskerov, F. (1984), "Formal methods of constructing collective decisions", in: Yu. Tyurin, ed.,
Proceedings of 2nd All-Union Conference on Statistical and Discrete Analysis of Ordinal Data
(VINITI, Moscow) pp. 18-28. In Russian.
Ch. 2: Categories of Arrovian Voting Schemes 127

Aleskerov, E (1985a), "Decision making procedures making use of voting procedures", in: B. Stanulov,
ed., Proceedings of the Conference of Complex Control Systems (Institute of System Analysis, Varna,
Bulgaria) pp. 357-366.
Aleskerov, E (1985b), "Procedures of multicriterial choice", in: Jang Jiachi, ed., Preprints of the IFAC/
IFORS Conference on Control Science and Technology for Development, Beijing, China (International
Federation on Automatic Control) pp. 858-869.
Aleskerov, E (1985c), "Local collective decision making procedures in different classes of the space
of choice functions", in: A. Dorofeyuk, ed., Proceedings of the Institute of Control Sciences 'Data
Analysis in Control Systems' (Institute of Control Sciences, Moscow) pp. 62-69. In Russian.
Aleskerov, E (1991), "Qualitative models of multicriterial choice", in: A. Dorofeyuk, ed., Methods of
Collecting and Analyzing of Complex-Organized Data (IPU, Moscow) pp. 61-69. In Russian.
Aleskerov, E (1992), "Relational-functional voting operators", Social Science Working Paper 818
(California Institute of Technology).
Aleskerov, E (1993), Local Models of Voting (Institute of Control Sciences, Moscow). In Russian.
Aleskerov, E (1994), "Multicriterial interval choice models", Information Sciences 80:25-41.
Aleskerov, F (1997), "Voting models in Arrovian framework", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Social Choice Theory: Re-examined, Vol. 1 (St. Martin Press, New York) pp. 47-67.
Aleskerov, E (1999), Arrovian Aggregation Models (Kluwer Academic Publishers, Dordrecht).
Aleskerov, E, and J. Duggan (1993), "Functional voting operators: the non-monotonic case", Mathematical
Social Sciences 26(2):249-272.
Aleskerov, F, and A. Vladimirov (1986), "Hierarchical voting", Information Science 39(1):41-86.
Arrow, K.J. (1951), Social Choice and Individual Values, 1st Edition (Wiley, New York).
Arrow, K.J. (1959), "Rational choice functions and orderings", Econometrica 26:121-127.
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Yale University Press).
Aumann, R.J. (1986), "Rationality and bounded rationality", Nancy L. Schwartz Memorial Lecture
(J.L. Kellogg School of Management, Northwestern University).
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, Cambridge).
Blair, D.H., G.A. Bordes, J.S. Kelly and K. Suzumura (1976), "Impossibility theorems without collective
rationality", Journal of Economic Theory 13:361-379.
Borda (J.-C. de Borda) (1781), "M6moire sur les elections par scrutin", M6moires de l'Academie Royale
des Sciences ann6e 1781, pp. 657-665. Translated in English in 1953 by A. de Grazia: "Mathematical
derivation of an election system", Isis 44:42-51.
Brams, S.J., and P.C. Fishburn (1982), Approval Voting (Birkhauser, Boston).
Brown, D.J. (1975), "Aggregation of preferences", Quarterly Journal of Economics 89:456-469.
Chebotarev, P (1994), "Aggregation of preferences by the generalized row-sum method", Mathematical
Social Sciences 27:293-320.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Danilov, V (1982), "Aggregation of tolerant preferences", in: Yu. Ovseenko, ed., Categories of Social
Utility: Problems of Methodology and Structurization (CEMI, Moscow) pp. 91-108. In Russian.
Danilov, V (1985), "The structure of binary rules for aggregating preferences", Matecon 21(4):44-65.
Denico16, V (1987), "Some further results on nonbinary social choice", Social Choice and Welfare
4:227-285.
Dodgson, C.L. (Lewis Carroll) (1873), A Discussion of the Various Methods of Procedure in Conducting
Elections (Imprint by E.B. Gardner, E. Pickard Hall and J.H. Stacy, Printers to the University, Oxford).
Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections (Cambridge University
Press, Cambridge) pp. 214-222.
Ferejohn, J.A., and P.C. Fishburn (1979), "Representation of binary decision rules by generalized
decisiveness structures", Economic Theory 1:28-45.
128 E Aleskerou

Fishburn, P.C. (1974), "On collective rationality and a generalized impossibility theorem", Review of
Economic Studies 41(4):445-457.
Fishburn, PC. (1975), "Axioms for lexicographic preferences", Review of Economic Studies 42(1):
19-35.
Gibbard, A.F (1969), "Social choice and Arrow condition", Unpublished manuscript.
Green, D., and I. Shapiro (1994), Pathologies of Rational Choice Theory: A Critique of Applications in
Political Science (Yale University Press, New Haven, CT).
Grether, D.M., and C.R. Plott (1982), "Nonbinary social choice: an impossibility theorem", Review of
Economic Studies 49:2.
Guha, A.S. (1972), "Neutrality, monotonicity and the right of veto", Econometrica 40:821-826.
Ilyunin, O., B. Popov and L. El'kin (1988), "Functional majority operators in voting theory", Automation
and Remote Control 7:255-267.
Kelly, J. (1978), Arrow Impossibility Theorems (Academic Press, New York).
Levchenkov, V (1987), "Algebraic representation of group choice", Doklady Akademii Nauk SSSR
293(6):1324-1329.
Makarov, I., T. Vinogradskaya, A. Rubchinskiy and V Sokolov (1982), Theory of Choice and Decision
Making (Nauka, Moscow). In Russian.
Malishevski, A. (1998), Qualitative Models In the Theory of Complex Systems (Nauka, Moscow).
Mas-Colell, A., and H. Sonnenschein (1972), "General possibility theorems for group decisions", Review
of Economic Studies 39:185-192.
Maskin, E. (1977), "Nash equilibrium and welfare optimality", Mimeo (MIT).
May, K.O. (1952), "A set of independent, necessary and sufficient conditions for simple majority
decision", Econometrica 20(4):680-684.
Mirkin, B. (1979), Group Choice (Whinston, New York).
Monjardet, B. (1978), "An axiomatic theory of tournament aggregation", Mathematics and Operation
Research 3(4):334-351.
Moulin, H. (1985), "Choice functions over a finite set: a summary", Social Choice and Welfare 2(2):
147-160.
Moulin, H. (1988), Axioms of Cooperative Decision Making (Cambridge University Press, Cambridge).
Moulin, H., and B. Peleg (1982), "Cores of effectivity functions and implementation theory", Journal
of Mathematical Economics 10(1): 115-145.
Muller, E., and M.A. Satterthwaite (1977), "The equivalence of strong positive association and strategy-
proofness", Journal of Economic Theory 14(4):412-418.
Murakami, Y. (1968), Logic and Social Choice (Dover, New York).
Naiton, W, and T. Watanabe (1995), "An extended formulation of Arrows theorem for k-social welfare
function", Social Choice and Welfare 12(3):215-224.
Nakamura, K. (1975), "The core of a simple game without ordinal preferences", International Journal
of Game Theory 4(1):95-104.
Nanson, E.J. (1882), "Methods of election", Transactions and Proceedings of the Royal Society of
Victoria 18.
Nurmi, H. (1987), Comparing Voting Systems (Reidel, Dordrecht).
Nurmi, H. (1999), Voting Paradoxes and How To Deal With Them (Springer, Berlin).
Parks, R. (1976), "Further results on path independence, quasi-transitivity and social choice", Public
Choice 26:75-87.
Peris, J.E., and M. Carmen Sanchez (2001), "Characterization of social choice sets in terms of individual's
maximal sets: the fixed agenda framework", Social Choice and Welfare 18(1):113-127.
Plott, C.R. (1973), "Path independence, rationality and social choice", Econometrica 41:1075-1091.
Plott, C.R. (1984), "A relationship between independence of irrelevant alternatives, game theory, and
implementability", Mimeo (California Institute of Technology, Pasadena).
Popov, B., and L. El'kin (1989), "Functional voting operators of non-complete and non-empty choice of
options", Automation and Remote Control 7:987-992.
Ch. 2: Categories of Arrovian Voting Schemes 129

Samuelson, P.A. (1938), "A note of the pure theory of consumer's behavior", Economica 5:353-354.
Schwartz, T. (1980), The Logic of Collective Choice (School of Social Sciences, University of
California).
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San-Francisco).
Sen, A.K. (1993), "Internal consistency of choice", Econometrica 61(3):495-521.
Sen, A.K., and PK. Pattanaik (1969), "Necessary and sufficient conditions for rational choice under
majority decision", Journal of Economic Theory 1:1-24.
Sholomov, L. (2000), "Explicit form of neutral social decision rules for basic rationality conditions",
Mathematical Social Sciences 39:81-107.
Stefanescu, A. (1997), "Impossibility results for choice correspondences", Mathematical Social Sciences
33(2):129-148.
Suzumura, K. (1974), "Rational choice and revealed preferences", Review of Economic Studies 44(1):
38-47.
Suzumura, K. (1983), Rational Choice, Collective Decisions and Social Welfare (Cambridge University
Press, Cambridge).
Vladimirov, A. (1987), "Functions of algebra of logic in the theory of social choice", in: A. Volkov, ed.,
Control of Complex Technical Systems, Preprints (Institut Problem Upravleniya, Moscow) pp. 17-20.
In Russian.
Vol'skiy, V, and Z. Lezina (1991), Voting in Small Groups (Nauka, Moscow). In Russian.
Wilson, R.B. (1972), "Social choice theory without the Pareto principle", Journal of Economic Theory
5:476-486.
Chapter 3

DOMAIN RESTRICTIONS*

WULF GAERTNER
Universitit Osnabriick, Germany

Contents

Abstract 132
Keywords 132
1. Introduction 133
2. Notation and definitions 135
3. The existence of collective choice rules under exclusion conditions for
finite sets of discrete alternatives 136
3.1. Basic concepts 136
3.2. The case where the individual preferences are orderings - The method of majority
decision is a social welfare function or a social decision function 137
3.3. The case where the individual preferences are quasi-transitive - The method of majority
decision is a social welfare function or a social decision function of type QT for
quasi-transitive individual preferences 139
3.4. Alternative single-stage social decision rules and multi-stage majority decision rules 141
3.5. The probability of no Condorcet winner 143
3.6. Coalition formation and simple games 144
4. The existence of Arrovian social welfare functions and the domain of the
simple majority rule 146
4.1. Domains for Arrow-type social welfare functions 146
4.2. The domain for the simple majority rule 152
5. Distributional restrictions over the set of individual preferences under simple
majority rule 153
5.1. Conditions on the distribution of preferences in the original profile 153
5.2. Conditions on the distribution of preferences in the reduced profile 156
6. Social choice in continuous space 160
6.1. Impossibilities 160
6.2. Contractibility and existence results 162
6.3. Discrete vs. continuous choices 165

* Helpful comments and suggestions by Nick Baigent, Eric Maskin, and Kotaro Suzumura are gratefully
acknowledged. I also wish to thank Volkswagen-Stiftung for generous financial support.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K. Sen and K. Suzumura
© 2002 Elseuier Science B. V All rights reserved
132 W Gaertner

7. Concluding remarks 166

References 167

Abstract

This chapter discusses different types of domain restrictions. We begin by analyzing

various qualitative conditions on preference profiles. Value-restricted preferences (with
single-peaked preferences as one of its subcases), limited agreement as well as
antagonistic and dichotomous preferences are relatively easy to interpret. In our view,
the property of single-peakedness stands out in particular. It proves to be a central
restriction under majority voting. However, it also plays an essential role in the
context of strategy-proof voting rules (which is the topic of another chapter in this
Handbook). Furthermore, we consider quantitative or number-specific requirements
on the distribution of voters over different preference orderings, and we shall see that
some of those requirements are logically related to the qualitative conditions such
as extremal restriction and value-restricted preferences. While the latter restrictions
are requirements on combinations of individual orderings, the domains of individual
orderings that admit n-person nondictatorial social welfare functions la Arrow
result from restrictions on permissible preferences for individuals. While the first five
sections study the aggregation problem within the framework of arbitrary finite sets of
discrete alternatives, the final section discusses continuous choice rules; the alternatives
are assumed to be n-dimensional vectors in Euclidean space. Contractibility as a
condition on the topological space of preferences proves to be necessary and sufficient
for the existence of continuous aggregation rules.

Keywords

domain restrictions, Arrow social welfare functions, topological social choice,

quantitative restrictions

JEL classification:D7, D6
Ch. 3: Domain Restrictions 133

1. Introduction

Unrestricted domain is a widely used assumption in social choice theory, most notably
postulated in three of the best known impossibility results in this research area, viz.,
Arrow's General Possibility Theorem (1951, 1963), Sen's Impossibility of the Paretian
Liberal (1970) and the Gibbard-Satterthwaite result on the nonexistence of strategy-
proof voting procedures (1973, 1975). Is the axiom of unrestricted domain a natural
requirement? There seem to be several answers to this question. One possible response
has to be viewed in close relationship to Arrow's concept of a social welfare function.
We borrow from Arrow's 1951 text (p. 24): "If we do not wish to require any prior
knowledge of the tastes of individuals before specifying our social welfare function,
that function will have to be defined for every logically possible set of individual
orderings. Such a social welfare function would be universal in the sense that it would
be applicable to any community" (the italics are ours). A related response touches
upon the aspect of limited information. We quote from the same page of the 1951 text.
"We will ... suppose that our a priori knowledge about the occurrence of individual
orderings is incomplete, to the extent that there are at least three among all the alterna-
tives under consideration for which the ordering by any given individual is completely
unknown in advance". On the other hand, Sen (1970, p. 165) argued that "individual
preferences are determined not by turning a roulette wheel over all possible alterna-
tives, but by certain specific social, economic, political, and cultural forces. This may
easily produce some patterns in the set of individual preferences". This view seems to
be shared by Pattanaik (1971) and, in very different ways, by Marglin (1995), among
others. Marglin speaks of a homogenization of preferences accomplished by culture.
The problem that we run into when we admit an unrestricted domain is that a social
welfare function or a social decision function may not exist. Is it possible that certain
properties of the individual preferences themselves introduce a natural restriction
that circumvents impossibility results? Again, the answer is not so clear. Consider
a division-of-a-cake-problem and let us postulate that there are three individuals
who possess monotonic preferences with respect to their own share only (a typical
assumption from microeconomics). Suppose the three alternative divisions for this
three-person society are a = (, , ), b = (,0, ), and c = (5, , 4). According
to the monotonicity assumption, the three preference orderings can be taken to be
bPlcPla, cP2aP2 b, and aP3bP3c. We have here a preference structure that we shall
later on call a "latin square". If this three-person society agrees to use the simple
majority decision rule as their vehicle in order to reach a social outcome, the depicted
case ends up in what is known as a "Condorcet paradox". A majority prefers a to b,
b to c, and c to a. The simple majority rule generates a preference cycle for society 1.

1 Fixing a particular sequence of pairwise decisions would, of course, be a way out of this impasse,
but among self-interested individuals none of the three possible proposals would win a simple majority
over the other two [see Dixit and Nalebuff (1991, Chapter 10) for several examples illustrating this
phenomenon].
134 W Gaertner

There is nothing pathological about the individuals' orderings so that the majority
cycle occurs "quite naturally". However, for a slightly different set of alternatives, the
monotonicity requirement can "save the day". Change alternative b into b = ( ',°),
and assume again that the persons are monotonic in their own share of the cake.
This time person 3 will have the preference b'P 3aP3c, the profile now satisfies
a particular structural property called single-cavedness to be defined later, so that
the majority rule generates a transitive social preference relation. To elaborate a
little further on the occurrence of preference cycles, one can easily see in the case
of the Condorcet paradox that one single "wrong" ordering added to an "orderly"
preference profile can bring about "disorder". Consider one thousand voters who have
the preference aPbPcPd,another one thousand voters stating bPcPdPa, yielding no
preference cycle, and add one person with the ordering dPaPbPc. A "disorderly"
preference cycle is the immediate consequence.
Modern social choice theory started with Arrow's book from 1951 and work by
Black (1948). According to McLean and London (1990), the roots of the theory
of collective choice can be traced back to the end of the 13th century (possibly
earlier than that). A bit more than two hundred years ago, two French scientists,
Borda (1781) and the Marquis de Condorcet (1785) made important contributions to
the mathematical theory of voting. Condorcet extensively discussed the election of
candidates under the majority rule. Condorcet demonstrated the existence of cyclical
majorities for particular preference profiles and proposed a resolution scheme for such
cases. Roughly another one hundred years later, Dodgson (1876) explicitly dealt with
the case of cyclical majorities under various voting schemes.
To the best of our knowledge, none of the earlier authors thought of a solution to
the problem of cyclical majorities via restricting "the variety" of individual preference
orderings. Domain conditions of various forms for different collective choice rules are
the topic of this investigation. The literature on domain conditions for social choice
rules can be split up into two large classes. The contributions to the first class study the
aggregation problem for arbitrary finite sets of discrete alternatives such as political
parties or candidates representing these parties. The contributions to the second class
assume that the set of options has a topological structure. Most authors in this category
suppose that the choice space is the n-dimensional Euclidean space. Possible situations
to consider in this class are decisions on the composition of the federal budget or
decisions on the production of various public goods. Other examples would be choices
over a set of social states that are characterized by numerical values only such as the
unemployment rate, the inflation rate, the federal deficit, the trade surplus or deficit,
and other indicators. Within the first class, a further distinction has been made. The
domain conditions with respect to the individual preference relations either have the
characteristic of being exclusion conditions (particular individual preference relations
are not permitted to be held by any member of society), or the domain conditions admit
all logically possible individual preference orderings, for example, but make certain
requirements as to the distribution of individuals over these orderings.
The structure of this survey follows the different aspects just outlined. Section 2
Ch. 3: Domain Restrictions 135

will give a minimum of notation and definitions indispensable to our presentation.

Section 3 discusses the existence of collective choice rules under exclusion conditions
for finite sets of discrete alternatives. Section 4 deals with the existence of Arrow social
welfare functions and the preference domain of the simple majority rule. Section 5
describes restrictions on the distribution of individuals' preferences. Section 6 deals
with topological social choice. It gathers some results on the existence of social choice
rules in n-dimensional continuous space. We end with a few concluding remarks.

2. Notation and definitions

The notation and definitions that we are introducing in this section will remain valid
throughout Sections 2-5. Section 6 will use a somewhat different terminology.
Let X = {x,y,z, ... } denote the set of all conceivable social states and let
N = {1, .... n} denote a finite set of individuals or voters (n > 2). Let R stand
for a binary relation on X; R is a subset of ordered pairs in the product X x X.
We interpret R as a preference relation on X. Without any index, R refers to the
social preference relation. When we speak of individual i's preference relation we
simply write Ri. The fact that a pair (x,y) is an element of R will be denoted xRy;
the negation of this fact will be denoted by -xRy. R is reflexive if for all x E X: xRx.
R is complete if for all x,y E X,x y: xRy or yRx. R is said to be transitive if for
all x,y,z E X: (xRy A yRz) -- xRz. The strict preference relation (the asymmetric
part of R) will be denoted by P: xPy - [xRy A yRx]. The indifference relation
(the symmetric part of R) will be denoted by I: xly -* [xRy A yRx]. We shall
call R a preference ordering (or an ordering or a complete preordering) on X if
R is reflexive, complete and transitive. In this case, one obviously obtains for all
x,y X: xPy -3 -yRx (reflexivity and completeness of R are sufficient for this
result to hold), P is transitive and I is an equivalence relation; furthermore for all
x,y,z C X: (xPy AyRz) - xPz. R is said to be quasi-transitive if P is transitive. R is
said to be acyclical if for all finite sequences {xl,... , xk } from X it is not the case that
X1PX2 A x 2Px 3 A ... A xk 1Pxk and xkPX . The following implications clearly hold: R
transitive - R quasi-transitive R acyclical.
An important result due to Sen (1970) states that if R is reflexive and complete, then
a necessary and sufficient condition for the existence of a nonempty choice set C(S),
i.e. the existence of a best element for all finite subsets S of X is that R be acyclical
(the existence of a best element for all S means that for each S, there exists x c S such
that for all y: y E S xRy). The interest in the property of acyclicity is due to the
close relationship between this result and the existence of social decision functions.
In the context of social choice theory, the following interpretations can be attached
to the relations R, P, and I. xRy means that "x is socially at least as good as y";
xPy means that "x is socially better than y", and xly means that x and y are socially
indifferent.
136 W Gaertner

Let A denote the set of preference relations on X, /3 denote the set of preference
relations which are reflexive and complete on X, C denote the set of preference
relations which are reflexive, complete and acyclical on X, D the set of preference
relations which are reflexive, complete and quasi-transitive on X, and £ the set of
preference orderings on X. Obviously, C D C C C B C A. A', 1, ... will stand
for subsets of A, 3, ... fulfilling particular restrictions. A"' will denote the cartesian
product A' x ... x A', n-times. An element of A is an n-tuple of preference relations
(R 1, ... , R,) or the profile of an n-member society consisting of preference relations.
A collective choice rule is a mapping from A"n to A. A social welfare function
is a mapping from l"' to . A social welfare function for quasi-transitive individual
preferences is a mapping from D'n to . A social decision function of type QT is a
mapping from £"' to D. A social decision function is a mapping from £E" to C. A social
decision function is a collective choice rule such that a choice function is generated
over the set of alternatives. We shall also consider social decision functions which
are defined on D" [see Arrow (1963), Fishburn (1970), Inada (1970), Sen (1970),
Pattanaik (1971), Mas-Colell and Sonnenschein (1972)].

3. The existence of collective choice rules under exclusion conditions for

finite sets of discrete alternatives

In this section we study various weakenings of the condition of unrestricted domain.

In other words, we are looking for restrictions on individual preferences such that
social preference relations with varying degrees of rationality exist (transitive, quasi-
transitive or acyclical). A common feature of these restrictions is that either certain
individual orderings are not allowed to occur at all or particular individual orderings
are excluded in the presence of other orderings.

3.1. Basic concepts

The first collective choice rule that was examined in the context of domain conditions
was the method of simple majority decision 2 . We assume that the individual preference
relations are reflexive and complete. We let N(xRiy) denote the number of individuals
for whom xRiy.
Definition. The method of majority decision is a collective choice rule defined on /3"
such that

(R 1 . . . Rn), Vx,y E X: xRy - [N(xRiy) > N(yRix)].

Clearly, the range of this collective choice rule is the set B. May (1952) provided a
complete characterization of the simple majority rule [see also Sen (1970, Chapter 5)],

2 See Black (1948).

Ch. 3: Domain Restrictions 137

using the conditions of anonymity (all voters are treated symmetrically or equally - the
outcome does not depend on which individual expresses which preference), neutrality
(if alternatives x and y, and z and w, respectively, are preference-wise arranged in an
identical way in two profiles, the social relation between x and y in the first profile
should be exactly the same as the relation between z and w in the second profile), and
positive responsiveness (a change in favour of x in only one individual's preference
with respect to alternatives x and y, let's say, is sufficient for a change from social
indifference to social preference between x and y).
Since we are about to examine particular conditions on the pattern of individual
preferences, we have to make clear what we understand by sufficiency and necessity
of a condition. We shall say that an exclusion condition restricting to ' is
sufficient for the existence of a social welfare function, for example, if the generated
social preference relation is an ordering for any set N and any n-tuple or profile
n.
(R 1, ... ,Rn) C a£ Furthermore, an exclusion condition is said to be necessary if
for all N and all n-tuples (R1, ... R,n) the social relation is an ordering only if E' c E
satisfies this condition. In other words, whenever E' c violates this condition,
there is some set N and some n-tuple (R 1, ... R,) such that the social relation does
not have the ordering property3 . These definitions of sufficiency and necessity are
particularly useful in the present context where the aim is to find structural conditions
on preferences without considering numbers of voters and the distribution of voters
over the set of admissible preferences.

3.2. The case where the individual preferences are orderings - The method of
majority decision is a social welfare function or a social decision function
Firstly, we shall characterize the sets £' for which the method of majority decision
generates a transitive social relation or a social relation that lies in C. Since the
transitivity property is a condition which is defined over triples, we can state the
relevant conditions in terms of triples of distinct alternatives (x,y,z); a,b, and c
are three distinct variables each of which can assume one of the values x,y, and z.
Before we introduce various restrictions, we wish to separate out those individuals
who are indifferent among all the alternatives (the so-called unconcerned individuals).
A concerned individual for a set of options then is one who is not indifferent between
every pair of elements.
Definition. Extremal restriction (ER) [Sen and Pattanaik(1969)]: Va, Vb, Vc:
[3i: aPib A bPic - Vj:
l (cPja -- cPjb A bPja)].
If for some Ri, a is preferred to b, and b is preferred to c, then for all Rj such that c is
preferred to a, c is also preferred to b and b is preferred to a. In other words, if there

3 These are not the only possible definitions of necessary and sufficient conditions. For a justification as
well as alternative definitions see Inada (1969), Sen and Pattanaik (1969), Pattanaik (1971), Fine (1973),
Kelly (1974a), and Kaneko (1975).
138 W Gaertner

is some individual with a strong ordering over the triple (a, b, c) and if there are other
individuals who regard as uniquely best the alternative that the first person considers as
uniquely worst, then the other persons regard as uniquely worst the alternative which
the first person regards best.
It is not difficult to show that this condition is equivalent to the union of three
restrictions proposed by Inada (1964, 1969), viz., "antagonistic" preferences, "echoic"
preferences, and "dichotomous" preferences 4 . Dichotomous preferences, for example,
are given if within a triple there is at least one pair of alternatives between which
each individual is indifferent (the pair need not be the same for different individuals).
Condition ER satisfies this property trivially.
Theorem 1 [Sen and Pattanaik (1969), Inada (1969)]. If and only if the set of
individual orderings 8' satisfies condition ER for every triple of alternatives, the
method of majority decision is a social welfare function.
Condition ER is unique in the sense that it is the only condition in this category
that proves to be necessary and sufficient for the existence of a social welfare function.
Since Black's (1948) and Arrow's (1951) investigations we know that there are other
restriction conditions which permit the existence of a social welfare function. For these
conditions, however, an additional requirement has to be made, viz., the number of
concerned voters has to be odd.
Definition. Value restriction (VR) [Sen (1966)]: 3a, Eb, 3c such that for all Ri
concerned:

[Vi: aPib V aPic] V [Vi: bPia V cPia]

V[Vi: (aPib A aPic)V (bPia A cPia)].

In the triple (x,y, z) there is an option a such that all concerned individuals agree that
it is not worst, or agree that it is not best, or agree that it is not medium.
The property of value restriction generalizes various conditions proposed by Black
(1948, 1958), Arrow (1963, Chapter 7), Inada (1964), Vickrey (1960), and Ward
(1965), among others. The first case in the above definition, for example, describes
the situation of single-peaked preferences, the second case defines single-caved
preferences.
The next condition is a weaker version of Inada's (1969) restriction of "taboo
preferences".
Definition. Limited agreement (LA) [Sen and Pattanaik (1969)]: a, ]b: [Vi: aRib].
In the triple (x,y, z) there is an ordered pair (a, b) such that every individual regards
a to be at least as good as b.

4 See Section 3.3 below.

Ch. 3. Domain Restrictions 139

Theorem 2 [Inada (1969)]. If and only if the set of individual orderings£' satisfies
for every triple of alternatives at least one of the conditions ER, VR and LA, and
provided the number of concerned voters is odd, the method of majority decision is a
social welfare function.
If the oddness requirement appears unacceptable and the condition ER somewhat
difficult to interpret, one can decide to be less demanding and resort to the concept
of a social decision function which requires the existence of best elements, not
complete orderings. The existence of a best element guarantees that there is a so-called
Condorcet winner.
Theorem 3 [Sen and Pattanaik (1969)]. If and only if the set of individual orderings
E' satisfies at least one of the conditionsER, VR andLA for every triple of alternatives,
the method of majority decision is a social decision function.
Note that the antecedent in this Theorem is identical to the one in Theorem 2 except
for the suppression of the oddness requirement.

3.3. The case where the individual preferences are quasi-transitive - The method of
majority decision is a social welfare function or a social decision function of
type QTfor quasi-transitive individualpreferences
For the results that follow some of the conditions which were presented in Section 3.2
have to be modified.
Definition. Generalized value restriction VR* [Fishburn (1970)]: In the triple (x,y,z)
there exists an element, say a, such that for b # a, c # a(b # c) in (x,y,z)

(aPib V aPic) for all Ri concerned,

V (bPia V cPia) for all Ri concerned,
V [(aPibA aPic) V (bPia A cPia)
V (alib A aIic)] for all Ri.

It is clear that if every Ri is an ordering, then VR* is equivalent to VR (the individuals

with aib A aIic in the definition above would then automatically become unconcerned).
If, however, every Ri is reflexive, complete, and quasi-transitive, then VR* is weaker
then VR. For example [Pattanaik (1970a)], if (xPiy A yliz A ziix) for some individuals
and (yPix Ayliz A zlix) for the others, then VR* is satisfied though VR is not.

Definition. Generalized limited agreement LA* [Inada (1970)]: In the triple (x,y,z)
there exists some ordered pair, say (a, b), such that for all i {1, ... , n},

if Ri is transitive, then aRib;

if Ri is not transitive, then aPib.

Again, if every Ri is an ordering, LA* is equivalent to LA.

 140 1 Guertrer

Definition. Dichotomous preferences (DP) [Inada (1969)]: For all i a {1,..., n},
Ri is transitive and for some distinct a, b in (x,y, z): aI b.

Definition. Antagonistic preferences (AP) [Inada (1969)]: For all i e {1,..., n},
Ri is transitive and there exists an ordered triple (a,b,c) such that for some
j,k E 1, ... , n}, aPjbPjc and cPkbPka and for all i distinct fromj and k, aic.

Definition. Strict agreement (SA) [Pattanaikand Sengupta (1974)]: There are three
distinct alternatives a, b, c in (x, y, z) such that one of the following three cases holds:
(i) for all i E {1, ... , n}, if Ri is concerned and transitive on (x,y,z), then
aPic A bPic, and if Ri is not transitive on (x,y,z), then aRic A bRic;
(ii) for all i E {1, ... , n}, if Ri is concerned and transitive on (x,y,z), then
aPib A aPic, and if Ri is not transitive on (x,y,z), then aRib A aRic;
(iii) for all i E {1, ... , n}, if Ri is concerned and transitive on (x,y,z), then
aPic A (aPib V bPic), and if Ri is not transitive on (x,y,z), then aRib A bRicA
aRic.
The following Theorem which was formulated by Fishburn (1972) and, independently,
by Pattanaik and Sengupta (1974) characterizes the sets of preference relations D'
which guarantee that the social relation is an ordering generated by the method of
majority decision.
Theorem 4 [Fishburn (1972), Pattanaik and Sengupta (1974)]. If and only if the
set of individual relations D' satisfies one of the conditions VR*, LA*, DP AP and
SA for every triple of alternatives, the method of majority decision is a social welfare
function for quasi-transitiveindividualpreferences, provided the number of concerned
individuals with a transitive preference relation Ri is odd.
This result is a generalization of Theorem 2 above. It is of particular importance
if there are reasons to believe that individual weak preferences are likely to be
quasi-transitive rather than transitive. Note that transitivity of the weak preference
relation implies transitivity of the indifference relation. The latter property can become
problematic in cases where the decision maker is unable to perfectly discriminate
among alternatives. For such cases, the assumption of quasi-transitivity may be found
to be on safer ground. We are finally interested in a characterization of the sets D'
such that a quasi-transitive social preference relation exists.
Theorem 5 [Fishburn (1970), Inada (1970)51]. If and only if the set of individual
preference relations D' satisfies at least one of the conditions VR*, LA*, DP and AP
for every triple of alternatives, the method of majority decision is a social decision
function of type QTfor quasi-transitiveindividualpreferences.

5 For a correction of Inada's proof see Salles (1974).

Ch. 3: Domain Restrictions 141

3.4. Alternative single-stage social decision rules and multi-stage

majority decision rules

Sen and Pattanaik (1969) have shown that if a binary collective choice rule is neutral
and non-negatively responsive 6 , then a quasi-transitive social preference relation
will be generated if the individual orderings are value restricted for every triple of
alternatives. An example of a collective choice rule that is neutral, anonymous, and
non-negatively responsive is the strict majority rule.
Definition. The strict majority rule: V(R,., R,),Vx,y X: xPy "- N(xPiy) >
2 NI, where NI is the total number of individuals. Further, xRy t- -yPx.

Starting from this definition, one can now define a continuum of collective choice
rules that will lie between the strict majority rule and the method of simple majority
decision. Let N* be the number of non-indifferent individuals in the relation between
two distinct alternatives x and y.
Definition. The semi-strict majority rule: (R 1, ... , R,),Vx,y E X: xPy <,
N(xPiy) > [p · INI + (1 -p)N*], for some given p chosen from the open interval
]0, 1[. Further, xRy t- yPx.
If p = 0 we obviously obtain the simple majority rule, for p = 1 we have the
strict majority rule. We know that if every individual has dichotomous preferences,
the simple majority rule generates a quasi-transitive (even fully transitive) social
preference relation. The strict majority rule also yields a quasi-transitive social
relation under condition DP. However, for no p ]O, 1[ does the semi-strict majority
rule generate a quasi-transitive social relation for dichotomous preferences of the
individuals [Jain (1986)]. As for condition ER, we have seen that it is a (necessary and)
sufficient condition for the existence of a transitive social relation under the simple
majority rule. Extremal restriction is no longer sufficient even for quasi-transitivity
if we apply the strict majority rule or the semi-strict majority rule. Considering the
continuous variation of p in the open interval ]O, 1[ we, therefore, observe some sort
of discontinuity, for with p appropriately chosen we can come arbitrarily close to the
points p = 0 and p = 1.
Jain (1986) has shown that a necessary and sufficient condition for quasi-
transitivity of the social preference relation under every semi-strict majority rule is
that condition VR or the condition of "absence of a unique extremal value" holds for
every triple of alternatives. In the same paper, Jain has given a necessary and sufficient
condition for transitivity of the social relation under every semi-strict majority rule.
A multi-stage majority decision rule (some authors prefer to call it a representative
majority group decision function) is a collective choice rule which is composed of a
hierarchy of one or more majority voting operations in each of which the entries consist

6 For a definition of these properties, see Sen (1970, Chapter 5).

142 W Gaertner

of individual weak preference relations or the results of majority voting operations

which have already been performed at a lower level. For this class of group decision
rules it is possible that the preference relation of individual k, let's say, enters into
more than one majority voting operation. It can also enter more than once into one
majority voting operation.
The method of majority decision is a special element in this class of social decision
rules in so far as it involves one and only one majority voting operation in which the
preference relation of each individual enters exactly once.
A majority voting operation at the lowest level is an operation of order zero where
each entry is an individual weak preference relation. A majority voting operation of
order one is characterized by the fact that at least one entry is the value of a majority
voting operation of order zero. More generally, a majority voting operation of order n
is one in which each element is the value of a majority voting operation of order n - 1
or n - 2 or, ... , or zero, and at least one of these entries is the value of a majority
voting operation of order n - 1.
Given a finite sequence of individual preference relations, there are many different
multi-stage majority decision rules. For the individual relations R1 , R2, R 3 one can, for
example, have ((R1,R 3),R 2) or ((Ri,R 1,R 2 ,R 3),R 2,(R,R 3 ,(R 2,R 3))). ((R1,R 3),R 2),
for example, means that there is a majority voting operation of order zero, applied
to the profile (R1, R 3), and a majority voting operation of order one that is based on
the outcome of order zero, together with the individual ordering R2. For any particular
rule, however, the structure of majority voting operations is the same for all pairs of
alternatives.
Value restriction, limited agreement and the oddness requirement are the essential
characteristics in the following possibility result for the case of transitive individual
preferences.
Theorem 6 [Batra and Pattanaik (1971)]. If a finite sequence of individualorderings
satisfies VR or LA over every triple of alternatives, any multi-stage majority decision
rule generates a socialpreference ordering, given that the number of entries concerned
with respect to any triple of alternatives is odd for every majority voting operation
involved.
The requirement that the number of entries concerned with respect to any triple
be odd for every majority voting operation is indispensable. Otherwise, the above
Theorem may not hold even if the number of concerned individuals is odd for every
triple.
In the case where the individual preferences are quasi-transitive, the oddness
requirement is less stringent. It is only demanded that the number of transitive and
concerned entries in the final majority voting operation be odd.
Theorem 7 Batra and Pattanaik (1972)]. Any multi-stage majority decision rule
generates a socialpreference ordering over the triple (x,y, z) if the finite sequence of
individualpreference relationsfrom set D' satisfies condition VR* or condition LA*
Ch. 3: Domain Restrictions 143

over (x,y,z) and the number of entries in the final majority voting operation which
are concerned with respect to (x,y,z) and transitive over it, is odd.

3.5. The probability of no Condorcet winner

In the introduction we already referred to Condorcet's paradox of voting which occurs

when each alternative in a voting process can be beaten by some other alternative
on the basis of the simple majority decision rule. Imagine that there are three
alternatives, three voters, and only strict individual orderings are admitted. We then
have six logically possible strict orderings and arrive at 6 x 6 x 6 = 216 logically possible
preference profiles for the three-member society. For this simple case one can easily see
that twelve profiles out of the 216 situations give rise to a majority cycle. If we make
the assumption that every individual ordering is as likely as every other (several authors
in this area of research call this assumption the "impartial culture condition"), we see
that cyclical majorities occur with a probability of 0.0555. How does this probability
change when we increase the number of options or (and) increase the number of voters?
Guilbaud (1952) was the first to calculate the probability that there is no majority
winner for the case of three alternatives when the number of voters increases. He
could show that the probability of a majority cycle goes up slightly with the number
of voters and reaches a limiting value just under 9% (0.0877).
Several authors have attempted to give analytical representations for the probability
of no majority winner [Gehrlein and Fishburn (1976, 1979), Gillett (1978), and others].
Let n for the moment be the number of voters and m be the number of alternatives.
Then we define Q(m, n) to be the proportion of the (m!)' profiles for which there are
no Condorcet winners. For the case of at least three alternatives, Kelly (1974b) has
proved that

Q(m, n + 1)<Q(m,n) for odd n > 3,

Q(m, n + 1) > Q(m, n) for even n > 2.

Fishburn, Gehrlein and Maskin (1979) have proved that Q(m, 3) < Q(m + 1,3) for
all m > 2 and Q(3, n) < Q(4, n) for odd n > 3. The situation becomes quite intractable
when m increases. Therefore, approximations for Q(m, n) have been developed, when
m and n are not relatively small [see e.g. Gehrlein (1983, 1998)]. Generally speaking,
Q(m, n) increases with n increasing. Q(m, n) increases even faster when the number of
alternatives grows. For 9 alternatives, for example, Q(9, n) reaches a limiting value of
45.45%. However, the picture brightens when we allow individuals to be indifferent
between distinct alternatives [Fishburn and Gehrlein (1980)]. Furthermore, once the
impartial culture condition is relaxed and it is asssumed, for example, that one of the
logically possible preference orderings occurs with a slightly higher frequency than
any of the others, the probability for preference cycles also decreases [Williamson
and Sargent (1967)].
144 W Gaertner

3.6. Coalition formation and simple games

The introduction of the concept of a simple game into the area of preference
aggregation goes back to Guilbaud (1952) and Blau (1957). We define a coalition
as a nonempty subset of N, the set of individuals. Let K be a coalition; xRKy means
that for all i K: xRiy, and XPKy means that for all i K: xPiy. A coalition K is
winning if and only if V(Rl, ... , Rn),Vx,y E X: xPKy - xPy.
Definition. A simple game is an ordered pair (N, W), where W, the set of winning
coalitions, satisfies:
K ,1 W and K1 C K2 - K2 C W.

This condition is called the monotonicity property for simple games. A simple game
can be interpreted as a committee where N represents the set of members of the
committee and )W gives the set of coalitions that fully control the decisions of the
committee.
Definition. Let (N, W) be a simple game. It is said to be proper if
KEW - N\K )/V.
It is said to be strong if
K ¢ W- N\K E W.

We now wish to relate the notion of a simple game to the concept of a collective
choice rule.
Definition. A simple voting game is a triple (N, W, f) where (N, W/) is a simple
game and f, the collective choice rule, is a mapping from A'" to A such that
V(R1... , R),Vx,y e X: xPy -+ [3K G W: xPKy], where P is generated byf.
A well-known example for a simple game is the voting system of the United Nations
Security Council [see Riker and Ordeshook (1973)]. The Council consists of five
permanent members and ten non-permanent members. In order to pass substantive
motions nine votes are required, including the five permanent members. This game is
proper since the complement of a winning coalition clearly is not winning. This game
is not strong since according to the rules neither the five permanent members, for
example, nor the ten non-permanent members of the Council form a winning coalition.
The strict majority rule which we have defined in Section 3.4 above is a proper simple
game. It is strong only if the total number of voters is odd. We now examine a
slight modifications of the strict majority rule that was proposed by Dummett and
Farquharson (1961).

7 This modification introduces a tie-breaking rule in the case of an even number of voters.
Ch. 3: Domain Restrictions 145

Definition. A majority voting game is a simple voting game (N, W, f) where W = {K:
[IK > INI, when INI is odd] V [(IKI > INI) V (IKI = NI A1 K) when
INI is even]} and where f is a mapping from B'"n to B. IKi denotes the cardinality of
K, IN[ is the number of voters, and voter 1 can be interpreted as the chairman having
a tie-breaking vote as well as an ordinary vote.
One can easily see that this rule defines a strong and proper simple game. In the
following, the sets B' in the definition shall be replaced by sets £' so that the domain
off will be '".
Definition. (a) Not-strictly-worst value restriction(NSW) [Dummett and Farquharson
(1961)]: There exists an alternative in a given triple (x,y,z) such that it is not strictly
worst in any preference ordering, i.e.,

3a, 3b, 3c such that for all Ri: aRib V aRic.

Definition. (b) Not-strictly-best value restriction (NSB) [Pattanaik (1970b)]: There

exists an alternative in a given triple (x,y,z) such that it is not strictly best in any
ordering, i.e.,

3a, 3b, 3c such that for all Ri: bRia V cRia.

We define the choice set C(S) as the set {x S: xRy for all y E S}, where R
is generated by rule f of the majority voting game. We can now state the following
result.
Theorem 8 [Dummett and Farquharson (1961), Pattanaik (1970b)]. If the set of
individual orderings ' satisfies condition NSW for every triple of alternatives or
condition NSB for every triple of alternatives belonging to a finite set S, the majority
voting game (N, W), f) generates a nonempty choice set C(S).
Note that it is not permissible in the present case that the restrictions vary from triple
to triple. In other words, a nonempty choice set need not exist when some triples satisfy
condition NSW and the other triples fulfil condition NSB.
Wilson (1972) analyzed Arrow's general impossibility theorem within a game-
theoretical framework. He could show that the winning coalitions in Arrow's theory
meet the requirements for the winning coalitions in a strong and proper simple game.
Definition. A strong and proper simple voting game is a simple voting game (N, W, f)
where (N, W) is a strong and proper simple game and wheref is a mapping from B'3
to B.
For our purposes we let the domain off again be '". Salles (1975) used Sen's
value restriction condition and a new exclusion condition that he called "cyclical
indifferences" in order to arrive at the following result.
 146 W Gaertner

Theorem 9 [Salles (1975)]. If and only if the set of individual orderings £' satisfies
for every triple of alternatives one of the exclusion conditions of value restriction and
cyclical indifferences, every strong andpropersimple voting game generates transitive
social preferences (the collective choice rulef is a social welfare function).
Note that conditions ER and LA which are sufficient for the existence of a social
ordering under the method of majority decision, are no longer sufficient under the
present collective choice rules which clearly are more general than the majority rule.
Salles (1976) obtained a characterization of individual preference orderings such
that every proper simple voting game is a social decision function of type QT.

4. The existence of Arrovian social welfare functions and the domain of the
simple majority rule

Arrow's impossibility theorem stated that there exist no nondictatorial social welfare
functions when the set of individual orderings is unrestricted. In the following
investigations, we characterize domains of individual orderings that admit n-person
nondictatorial social welfare functions la Arrow. It is also shown that in the case of
strict individual preferences, simple majority rule is an aggregation mechanism that is
transitive on the widest class of domains of individual preferences.

4.1. Domains for Arrow-type social welfare functions

We begin by considering the following two examples which can be found in Kalai,
Muller and Satterthwaite (1979).
Example 1. Let X = {xl,X2,x3,yl,y2,y3} and assume that there are at least three
individuals. Consider the domain of preferences consisting of all n-tuples of orderings
with the following characteristics: the sets {xl,x 2,X3} and {Yi,Y2,Y3} are free triples
so that all logically possible orderings of the elements in these sets are permissible.
Furthermore, all elements from {Xi,X2,X 3 } are always arranged above all elements
from {Yl,Y2,Y3} in the admissible orderings, i.e. every individual ranks each xi above
each yj for i,j {1,2, 3 }. It is well known that any Arrow-type social welfare function
on this domain must have a dictator on each of the free triples. If we now construct
a social aggregation procedure in the way that individual 1 is made a dictator over
{x1 , 2,x3}, individual 2 is made dictatorial over {Yl,Y2,Y3 }, and unanimity determines
the social rankings between xi andyj, i,j {1, 2, 3}, we have chosen a rule that satisfies
Arrow's conditions I and P, and there is no dictator in Arrow's sense. ·

Example 2. Let X be as defined in Example 1. This time the preference domain

consists of all n-tuples of orderings such that the set {xl,x 2,X3} is a free triple, every
individual ranks each x i in {x, x2,x3} above each yj in {yl, Y2,Y3} and the elements in
the latter set are ranked unanimously in the descending order ylPy 2Py 3. If we consider
Ch. 3: Domain Restrictions 147

a social welfare function satisfying conditions I and P, there is a dictator on the free
triple and via the Pareto condition, this person becomes dictatorial over the whole
set X. So no Arrow social welfare function exists in this case. ·
In order to explain these two examples and interpret them in a wider context, we have
to introduce some notation and definitions. We want to call a pair of distinct alternatives
x,y E X trivial relative to the set of admissible preference orderings, if there is some
individual i who has only one admissible preference over the pair {x,y}. Pairs which
are not trivial are called nontrivial. Two nontrivial pairs B = {x,y} and C = {c,z}
are said to be strongly connected if the cardinality ] B U C = 3 and B U C is a free
triple. Thus B and C are strongly connected if they share an element in common and
together form a free triple. Two pairs B and C are called connected if a finite sequence
of pairs B1,B 2, ... , B 1,Bn with B1 = B and B, = C exists such that Bi and Bi+ 1 are
strongly connected for each i = 1,2, ... , n - 1. We wish to say that there is a common
preference domain if each person has the same set of admissible preferences. Finally,
a set of admissible preference orderings is called saturating if (a) the set of alternatives
contains at least two nontrivial pairs, and (b) any two nontrivial pairs are connected.
The following Theorem by Kalai, Muller, and Satterthwaite establishes an illumi-
nating result.
Theorem 10 [Kalai, Muller, Satterthwaite (1979)]. If a social welfare function f is
defined on a common saturatingpreference domain and satisfiesArrow s independence
condition and weak Pareto, then f is dictatorial.
The mechanism behind this result can be described as a local Arrovian approach
[Bordes and Le Breton (1989)]. First in this procedure, free triples are identified; then a
local version of Arrow's theorem is used to prove the existence of a dictator on a triple.
One then shows a contagion property, viz., that an individual's sphere of dictatorship
expands by a connecting process to other free triples, and finally to the whole set
of alternatives. Returning to the two examples above, in the first case, the domain
of preferences is not saturating. The two triples are free, but they are not connected.
Therefore, the Theorem above is not applicable and, as pointed out, a nondictatorial
Arrow social welfare function can be established. In the second case, the domain is
saturating. The set {xl,X2,x 3} is a free triple, all pairs involving {yl,Y2,Y3} are trivial
and all preference relations between the two sets are trivial as well. Therefore, in the
second example, the Theorem above applies.
Kalai, Muller and Satterthwaite show furthermore that if the space of alternatives
is R, n > 1, where each dimension represents a different public good, and if each
individual's preference ordering is restricted to be convex, continuous, and strictly
monotonic (the standard assumptions in microeconomic theory), Arrow's impossibility
result for social welfare functions holds. Le Breton and Weymark (1996) provide
various other examples for Arrow-type impossibilities. They consider, for example,
linear orderings on the (m - )-dimensional unit simplex and spatial preferences, where
alternatives are ranked according to their Euclidean distance from an ideal point.
148 t Gaertner

The assumption in the Theorem above that the preference domain is common and
saturating is merely sufficient, and not necessary in order to generate a dictatorial
welfare function. Le Breton and Weymark present an example where the preference
domain is common but not saturating; yet there is dictatorship. They also give an
example which shows that the common preference domain assumption is essential for
the result in Theorem 10, viz., there exist saturating preference domains such that a
nondictatorial social welfare function exists. Consider the following situation.
Example 3. Let there be two individuals and let the set of alternatives be
X = {x1, x 2,X 3,x 4 } . Furthermore, for both persons preferences are unrestricted
on {xI,X 2,X 3} and any pair of alternatives in this subset is nontrivial. However,
X4 PIXi for i E {1,2,3} and xiP2 x4 for i E 1,2,3}. So the two individuals have
opposite strict preferences on the trivial pairs. Le Breton and Weymark define a social
welfare function that sets xRy iff xRy for x,y C {xl,x 2 ,x 3 } and xPx4 whenever
x E {xl,x 2,x 3}. So person 1 dictates the social preference on {xl,X 2,x 3} and person 2
is a dictator on all the trivial pairs. This social welfare function satisfies independence,
weak Pareto, and nondictatorship on a preference domain that is saturating but not
common. U
The next step is much more ambitious. Are there necessary and sufficient conditions
on the domain of preferences such that an Arrovian social welfare function exists?
Maskin (1975, 1976) and, independently, Kalai and Muller (1977) proved a remarkable
result which greatly simplifies the analysis: There exists an n-person social welfare
function (voting procedure) for a given domain if and only if there exists a 2-person
social welfare function (voting procedure) for the same domain. Thus, the question of
domain restriction can be dealt with independently of the number of individuals in the
society.
In the subsequent model, only strict orderings will be considered. Let X denote a set
of mutually exclusive social alternatives with at least two elements and let 2 denote the
set of all strict preference orderings on X. Furthermore, let Q C 12 denote a nonempty
subset of Z. The elements of 2 represent the admissible preference orderings in
society N. For any n > 2, Qn stands for the set of all n-tuples of preference orderings
from 2, with (PI, ... , P,,) being a typical element of •2".
Definition. An n-person social welfare function on •2 is a mappingf: Q" -* 2 which
satisfies the following two conditions. (1) The Weak Pareto Condition (Unanimity).
For all (P1 ... , P,) CE "2 , for all x,y C X and all i E N: xPiy - xf(P 1, ... , Pn)y.
(2) Independence of Irrelevant Alternatives. For all x,y E X, all (PI, ... , P,,),
(Pl, ... , P;,) CE 2": [xPiy +- xP'y for all i E N] - [xf(P 1, .. ,P,)y +
xf (Pi, .... e',,Y]
The functionf is called dictatorial if there exists an i C N such that for every pair
x,y E X and for all (P 1 ... , P,) E 2n, f(P 1, ... , P,) = Pi. This notion of dictatorship
is known from Arrow's analysis.
Let T = {(x,y) e X 2 : x # y}, TR= {(x,y) C T: there exist no P 1,P2 CEQ such that
Ch. 3: Domain Restrictions 149

xP 1y and yP2 x} and NTR = T - TR. Thus, T consists of all distinct ordered pairs, TR
represents the set of trivial pairs (either xPy for all P E 2 or yPx for all P E Q2)and
NTR consists of the nontrivial ordered pairs (there exist Pi,P2 E 2 such that xPI y
and yP2x) 8. With these definitions Kalai and Muller's decomposability condition can
be defined.
Definition. Nondictatorial decomposability [Kalai and Muller (1977)]: The set
of preferences 2 c is nondictatorially decomposable (ND) with respect to a
domain 2n if there exist two sets SI and S2 with TR Si C T, i {1, 2}, having the
following properties:
(a) For all (x,y) G NTR, (x,y) E S1 if and only if (y,x) X S2.
(b) For all nontrivial pairs the following is true for i = 1,2:
(bl) If there are P 1,P 2 E 2 with xP1yPiz and yP2zP2x, then (x,y) E Si implies
(x,z) Si;
(b2) If there are P1, P2 E Q2 with xPI yPlz and YP 2zP 2x, then (z, x) E Si implies
(y,x) E Si;
(b3) If there is a P Q2 with xPyPz, then (x,y) Si and (y,z) Si imply
(x,z) Si.
Si is the set of pairs for which individual i is decisive. The condition states that there
are two persons with some power of decisiveness.
Theorem 11 [Kalai and Muller (1977)]. There exists a nondictatorialn-person social
welfare function on Q2 if and only if 2 satisfies condition ND.
It may be of interest to note that Kalai and Muller's domain condition is a restriction
on permissible preferences for individuals. The conditions discussed in Section 3
were restrictions on combinations of individual orderings (profile restrictions). It can,
however, be shown that a set of single-peaked preferences (which has to be defined
relative to some particular linear ordering) satisfies property ND and therefore admits
nondictatorial n-person social welfare functions [Kalai and Muller (1977, p. 466)].
There are analogues both to the Kalai-Muller-Satterthwaite theorem and to the
Kalai-Muller result for the case that the social alternatives are allocations of private
goods only and not, as discussed above, allocations of purely public goods. We wish
to discuss explicitly an analogue to the former Theorem. For situations that involve
private dimensions alone, we shall assume that the set of alternatives X is Cartesian,
i.e. X = i= 1Xi, where Xi describes that part of the social alternative relevant for
i (i's consumption set, let's say). We shall further assume that individuals are selfish
which means that they only care about their own component in a social alternative.
We shall write Di for individual i's selfish preference domain on the Cartesian set X,
where Di C S for all i E N, and D = Fi= Di is the Cartesian preference domain. For

8 Note that Kalai and Muller's use of the attributes "trivial" and "nontrivial" is different from the one
introduced at the beginning of this section.
150 tW Gaertner

the following result by Bordes and Le Breton (1989), it is not necessary to assume a
common preference domain, i.e. it is not necessary to have Di = Dj for all i,j e N.
Above, we already defined what it means for a set of admissible preference orderings
to be saturating. However, for the Bordes-Le Breton theorem, we need two more
definitions.
Definition. Individual i's selfish preference domain Di is supersaturating iff: (1) Di is
saturating; (2) for all nontrivial pairs x,y with respect to Di in X, there exist u, v E X
such that u,v {x,y}, and {x,y,u}, {x,y,v}, {x,u,v}, and y,u,v} are free triples
with respect to Di. A preference domain D is supersaturating if Di is supersaturating
for all i E N.
Bordes and Le Breton show that if D is supersaturating, any two nontrivial pairs are
connected.
Definition. Individual i's preference domain is hypersaturating iff: (1) Di is supersat-
urating, and (2) for all pairs {x,y} in X that are trivial for i and such that xPi y for all
orderings Ri Di, there exist u C X and R$ E Di such that xP,'uP'y, where the pairs
{x, u} and {y, u} are nontrivial for i. A preference domain D is hypersaturating if Di
is hypersaturating for all i C N.
Clause (2) in the last definition formulates a separation condition for trivial pairs.
A trivial pair {x,y} is separable for person i if i is not indifferent between x and y
and there exists an admissible ordering for i and an alternative u such that u is on an
indifference curve lying between the indifference curves through x and y and such that
{x, u} and y, u} are nontrivial pairs for i.
Theorem 12 [Bordes and Le Breton (1989)]. Suppose X is a Cartesian set of
alternatives. If a social welfare function f is defined on a preference domain D that
is both selfish and hypersaturating and iff satisfies Arrow independence condition
and weak Pareto, then f is dictatorial.
If for each i N, Di is the set of all selfish, continuous, strictly monotonic
(in own consumption), and convex preference orderings and if each consumption
set Xi is some strictly positive orthant, then this preference domain is hypersaturating.
However, the classical domain of microeconomic preferences for private goods admits
the existence of an Arrow social welfare function if the origin is included in the
individual consumption sets. As seen from Example 3, the latter result has to do with
the fact that, with the origin being included in the consumption set, it may not be
possible to separate some nonindifferent trivial pairs. Note, however, that according to
the Theorem above, dictatorship still holds on the subset of alternatives obtained by
deleting the origin from each individual's Xi. The same is true if all goods have to be
consumed in positive quantities.
Kalai and Ritz (1980) have proved an analogue to the Kalai-Muller theorem
above for the case that every social alternative is composed of a bundle of private
alternatives only and each individual i is concerned with the ith component of each
Ch. 3: Domain Restrictions 151

social alternative alone. Kalai and Ritz assumed that all the individual sets of private
alternatives and all the domains of admissible orderings are the same. The authors
proved the independence between the group size and the existence of an Arrow-type
social welfare function and they characterized all the restricted domains which admit
such functions.
Kalai and Ritz were criticized both for assuming a common preference domain
and for requiring the set of private alternatives to be the same for all members of
society. The main criticism, however, pointed to the fact that only strong orderings are
admissible in their model. If the possibility of indifference is allowed, their results do
not hold anymore.
Ritz (1983) was able to remedy some of these defects: individuals may indeed be
indifferent among their private alternatives, and each person may choose her preference
ordering out of a different set of permissible preferences. Ritz proved first that a set
of n restricted domains of preferences admits an Arrow-type social welfare function
independently of the number of individuals and then characterized all the sets of
restricted domains admitting Arrow-type social welfare functions.
Ritz (1985) generalized this last result further when he considered situations where
each social state is a combination of private and public alternatives. Some parts of
each social alternative are assigned separately to individuals while other parts affect
all the individuals together. Again, individuals may be indifferent among alternatives
and individuals need not be symmetric in their private alternatives sets and in the sets
of admissible orderings. Ritz characterized the sets of restricted domains of individual
preferences that admit two-person Arrow social welfare functions. In the earlier results
by Kalai and Muller (1977), Kalai and Ritz (1980), and Ritz (1983) it was always
the case that if a restricted domain of preferences admitted a k-person Arrow social
welfare function for a finite k greater than one, then it admitted an n-person Arrow
social welfare function for any given finite n. A similar result for the case of private
and public alternatives has not been achieved.
We should mention in passing that Kalai and Muller's domain condition which
guarantees the existence of a nondictatorial n-person social welfare function (Theo-
rem 11) secures the existence of a nonmanipulable voting procedure [Kalai and Muller
(1977)]. Ritz (1983, 1985) proved theorems similar to this result for the case of private
alternatives and the case of mixed alternatives.
Kalai and Muller's aggregation procedures are, strictly speaking, nondictatorial.
However, for societies with more than two individuals, these decision rules are hardly
acceptable. Two individuals are socially decisive over all pairs of alternatives, the rest
of the society has no decisive power at all. Therefore, Blair and Muller (1983) proposed
a class of decision schemes for which the distribution of power is somewhat more
egalitarian: it is required that no voter be "inessential" in the sense that he (or she)
is completely deprived of power. Put differently, individual i is essential under some
aggregation mechanism if and only if there exists an admissible profile of voters'
preferences such that an admissible change in person i's preferences would alter the
social outcome. Strengthening the property of nondictatorship to the requirement that
152 LV Gaertner

all members of society are essential has a nontrivial effect on the set of domains for
which Arrow-type social welfare functions exist. Allowing different persons to have
different sets of admissible orderings, Blair and Muller (1983) proved that under the
condition of essential decomposability of the set of ordered pairs of alternatives with
respect to the individual domains, an n-person essential Arrow social welfare function
exists.

4.2. The domain for the simple majority rule

In Section 3.1, we have already briefly referred to May's characterization of the
majority rule in terms of anonymity, neutrality and positive responsiveness. Note
that neutrality, as used here, is stronger than Arrow's independence condition, and
neutrality and positive responsiveness together imply the Pareto rule. Therefore, the
simple majority rule fulfills three of Arrow's four axioms. Unrestricted domain as the
fourth requirement may, unfortunately, yield an intransitive social preference. All this
is, of course, well known. Quite recently, however, Maskin (1995) and Dasgupta and
Maskin (1998) have come up with a characterization of simple majority voting that is
different in perspective from May's seminal work. The two authors are focussing on
the widest domains of preferences on which the majority rule is defined and they offer
a defense of this voting rule in terms of a criterion they call "robustness".
For which restricted domains of individual preferences does a given voting rule
become an Arrow social welfare function. For the case of individual weak orderings,
we have seen in Section 3 that there is not a unique answer. The situation becomes
simpler when we require that all individual orderings be strict and that, in addition,
there be an odd number of voters. Then we know that Sen's (1966) value restriction
condition is necessary and sufficient for the method of majority decision to be an
Arrow social welfare function. Maskin (1995) has now shown the following.
Let f be any collective choice rule that is anonymous, neutral and positively
responsive (and therefore fulfills the Pareto condition). Given the two requirements of
oddness and strict individual preferences, iff is transitive on a particular domain, then
simple majority rule is transitive on this domain as well. Moreover, unlessf itself is the
simple majority rule, there exists a domain of individual orderings on which majority
rule is transitive butf is not. Among all aggregation rules that fulfil the properties
discussed above, simple majority rule is the only one that is transitive on the widest
class of domains. In this sense the majority rule is robust. More precisely, let there be
an odd number of voters. Again, let Y denote the set of all strict orderings on X, with
Q2 C Z denoting a nonempty subset of 2.
Theorem 13 [Maskin (1995)]. Letf be an n-person collective choice rule that satisfies
anonymity, neutrality and the Pareto condition. Suppose thatf is transitive on Q2 (i.e.
f is an Arrow social welfare function on 2). Then so is the simple majority rule
on 2. Moreover, unless f is identical with simple majority voting, there exists some
domain 2Q'on which the majority rule is an Arrow social welfare function but f is
not.
Ch. 3: Domain Restrictions 153

Very recently, Dasgupta and Maskin (1998) derived an analogous result for the
situation that there is a continuum of voters (a large population). Individual indifference
is ruled out as before and it is shown that the simple majority rule is "maximally"
robust 9.
Maskin's result can be illustrated by comparing the majority rule with the Borda
rule or the Pareto extension rule. The latter says that for two alternatives x and y,
x is socially at least as good as y if and only if y does not strictly Pareto-dominate
x. We have already mentioned that for strict individual orderings only and for three
alternatives a, b, and c, let's say, value restriction (plus oddness) is necessary and
sufficient for the simple majority rule to be an Arrow social welfare function. For three
alternatives a, b, and c, there are six strict orderings, viz. abc, bca, cab (latin square 1),
and cba, acb, and bac (latin square 2). Sen's value restriction requires that one strict
ordering be deleted from each latin square. The Pareto extension rule (as well as the
Borda-count) is transitive if and only if the property of - what Maskin calls - quasi-
agreement is satisfied. The latter condition is stronger than value restriction. It is a
domain restriction that demands that for all triples (x,y, z), there exists one alternative,
say x, such that for all profiles (i) x is ranked higher than both y and z, or (ii) x is
ranked lower than both y and z, or (iii) x is ranked between the other two alternatives.
Thus, quasi-agreement requires that more than one strict ordering be deleted from each
of the two latin squares 10

5. Distributional restrictions over the set of individual preferences under simple

majority rule

The domain restrictions that we discussed in Section 3 were such that certain individual
preference orderings were not allowed to occur at all (as in the case of dichotomous
preferences) or were excluded in the presence of other preference relations (as, for
example, in the case of value restriction or extremal restriction. Once these exclusions
became effective, any distribution of individuals over the remaining preferences was
admissible. In this section, all logically possible individual orderings are allowed to
occur and we shall identify conditions on the distribution of voters' preferences such
that a consistent (i.e. transitive or quasi-transitive) majority decision will come about.

5.1. Conditions on the distribution of preferences in the originalprofile

A very simple, almost trivial condition for transitivity of the majority relation is that in
the case of strict orderings only, let's say, one strict preference strictly outnumbers all

9 Campbell and Kelly (2000) present yet another characterization of simple majority voting for an
arbitrary, but finite number of voters. Their proofs look at the family of almost-decisive coalitions.
10 Dasgupta and Maskin (1998) have shown furthermore that if the anonymity condition is somewhat
weakened, simple majority voting is no longer uniquely maximally robust. The class of weighted majority
rules is maximally robust as well.
154 W Gaertner

the other orderings taken together, or the situation where the number of voters having
each one of the possible strict orderings is exactly as large as the number of people
having precisely its inverse ordering. In Section 4.2, we mentioned the Pareto extension
rule and a necessary and sufficient condition such that this rule yields a transitive social
relation. Whenever for one of the strict orderings over x, y, and z, let's say, that occur
in society, it can be guaranteed that at least one person strictly opposes, the issue of
transitivity vanishes into thin air under this Pareto-based rule.
In his examination of Arrow's impossibility result Tullock (1967) discussed an
example where the space of alternatives is an arbitrary subset of RI2 and where each
individual i is characterized by a point ai in the space R2 such that for every x and y in
X, x is at least as good as y if and only if the distance between x and ai is not greater
than the distance between y and ai. Indifference curves of each person are circles of
center ai. The distribution of individuals among preferences can then be described by a
probability distribution on the plane. In Tullock's example the individuals, i.e. the best
points ai are uniformly distributed over a rectangle with center a*. It is not difficult
to show that the simple majority rule yields a relation that is transitive.
Grandmont (1978), who began his analysis with Tullock's example, established
a theorem which - roughly speaking - says that if the preference relations of the
members of society are "nicely" distributed around some relation R*, then the method
of majority decision selects a relation which coincides with R*. Grandmont starts out
by considering a family of relations (Ra)a E A, defined on a set of alternatives X, which
is indexed by a point a running over a convex subset A of an arbitrary Euclidean
space R'. The family (Ra,)a A is required to have a continuity property (C.1) and a
"betweenness" property (C.2) .
(C.1) For every x and y in X, the set {a E A Ray} is closed in A.
(C.2) For every a' and a" in A, Ra e (R,,,R,,) whenever a = Aa' + (I - A)a", where
A e (0, 1).
Ra (Ra,, R,,) is the "betweenness" property saying that for any two points a' and a",
relation Ra must lie between the relations R,, and Ra,,. Grandmont shows that a family
of relations (Ra)aE A, where A is an open convex subset of R' satisfies (C. 1) and (C.2)
if and only if for every x and y in X, one of the following conditions holds: (1) either
xPay for all a, or xa y for all a, or yPax for all a; (2) there exists q in R', q o, and
a real number c such that qa > c when XPa y, qa = c when xIa y, and qa < c when
YPax.
Consider the class of societies whose individuals have preferences which belong to
(Ra)aCA, with A as defined, satisfying properties (C.1) and (C.2). Any society in this
class can be characterized by a probability distribution v on A. For given alternatives x
and y in X, v(xRay) represents the proportion of voters having xR,y. The majority
relation RM is then defined as xRMy if and only if v(xR,y) > v(yRx).

11Note that in Grandmont's approach it is the set of individuals which has a topological structure and
not (necessarily) the set of social states X.
Ch. 3: Domain Restrictions 155

For any hyperplane H in iR, let A' and A" denote the intersections of A with the
two closed half spaces determined by H. Consider x and y, and let H be the unique
hyperplane of equation qa = c associated to x and y due to condition (2) above. Let 7-
be the family of such hyperplanes which is generated when x and y vary in X. Now
the majority relation RM coincides with Ra* for some a* in A if and only if:
(M) For every H in 7-t, v(A') > v(A") if and only if a* A'.
A stronger condition is obtained by requiring that the property stated in (M) be valid
for every hyperplane H of I ", not only for those of 7I:
(M.1) There exists a* in A such that for every hyperplane H of RI, v(A') = v(A") if
and only if a* E H.
Grandmont established the following result:
Theorem 14 [Grandmont (1978)]. Let (Ra) CA be a family of relations on a set of
alternatives X, where A is open and convex in RIP, which satisfies (C.1) and (C.2).
Then for any probabilitydistribution v on A which satisfies (M.1), the majority relation
RM coincides with Ra.
Note that the result in the Theorem also holds when the relations of the family
(Ra), CA are not transitive. If one considers individual orderings only, as in Tullock's
example, the majority relation RM is transitive "automatically". The majority relation
can be identified with a representative member of the family, a*. The weaker
condition (M) can be used when one-dimensional families of relations are studied.
Grandmont shows (1) that when (R,), eA is a family of transitive preference relations
on a set X satisfying (C. 1) and (C.2), where A is an open interval of the real line, then
preferences are single-peaked or single-caved; (2) when (Ra,), A is a one-dimensional
family of relations and society comprises an odd number of voters, then for every triple
of alternatives, there is an a* in A such that the majority relation coincides with R,.
on that triple. This shows that the well-known possibility theorem for single-peaked
(or single-caved) preferences is just a special case within a more general framework.
Saposnik (1975) presented a sufficient condition for transitivity where a distribu-
tional requirement is only made with respect to the set of strict individual orderings.
In the author's terminology, for any ordered triple (x,y, z) of alternatives the "clockwise
cycle" of individual orderings is defined to be
(1) xPi Piz,
(4) yPizPix,
(5) zPixPiy,
and the "counterclockwise cycle" of individual rankings is similarly defined as
(6) zPiyPix,
(2) xPizPiy,
(3) yPixPiz.
According to our earlier terminology, the clockwise and the counterclockwise cycle
represent two latin squares. "Cyclical balance" of the voters' preference profile is then
156 W Gaertner

given if and only if there is the same number of individual orderings constituting the
clockwise cycle and the counterclockwise cycle, i.e. ni + n4 + n5 = n2 + n3 + n6.
Saposnik's result is:
Theorem 15 [Saposnik (1975)]. Under the majority decision rule the socialpreference
relation is transitive if the individualpreference relations are cyclically balanced.
The domain of cyclically balanced preferences may appear rather small but Saposnik
observed that in his approach the distribution of voters over the subset of orderings with
at least one indifference has no bearing on the transitivity issue. This sounds surprising.
Gaertner and Heinecke (1977) proved that cyclical balance is a rather unique property
in the sense that it is not only sufficient but also necessary for this result to hold.
A clue to this is the fact that a preference profile that fulfills Saposnik's condition
can be transformed into a so-called reduced profile that either comprises dichotomous
preferences only or is the empty set (the latter is, for example, the case where all strict
and weak orderings have inverse orderings of exactly the same type and frequency).

5.2. Conditions on the distribution of preferences in the reduced profile

Under the method of simple majority decision all those voters who are indifferent
between every pair of alternatives out of a given set have no influence on the formation
of the collective preference relation. Therefore, there are good reasons to disregard
these unconcerned individuals. This (trivial) observation can immediately be extended
to other cases. Consider a preference profile where five individuals, let's say, have
the ordering yPizPix and five other individuals have the opposite or inverse ordering
xPizPiy. Obviously, these ten voters can be excluded from further analysis under
simple majority decisions as well.
Independent from each other, Gaertner (1977, 1988), Gaertner and Heinecke (1978),
and Slutsky (1975, 1977) have taken this observation as their point of departure to con-
struct new (imaginary) societies which, under majority voting, are equivalent to origi-
nally given societies but are more "tractable" due to a simplified preference profile.
Let N(xPiy) denote the number of voters for whom xPiy as before. The vector
(N(xPiy) - N(yPix),N(xPiz) - N(zPix),N(yPiz) - N(zPiy)) will be called majority
vector of the underlying preference profile II. Under the method of majority decision
the majority vector determines the collective preference relation R. Clearly, two prefer-
ence profiles characterized by the same majority vector yield the same social preference
relation. For our analysis, these two profiles will be considered as equivalent.
In the following it proves very useful to come back to Saposnik's notion of clockwise
and counterclockwise cycles and define the following cycles (or groups) of individual
orderings for a given triple (x,y,z):
C+: C-
type 1: xRiyRiz type 1: zRiyRix
type 2: yRizRix type 2: xRizRiy
type 3: zRixRy type 3: yRixRiz
Ch. 3: Domain Restrictions 157

We shall call C+ and C- the clockwise and the counterclockwise R-cycle

respectively. The corresponding P-cycles of strict preference are generated from C+
and C- by substituting P-P orderings for R-R orderings everywhere. Similarly, the
P-I cycle (with one strict preference and one indifference in each individual ordering)
and the I-P cycle (indifference first, then strict preference) are obtained from C+ and
C- by replacing the R-R orderings everywhere by P-I orderings and I-P orderings,
respectively.
Consider a preference profile consisting of only two individual orderings. We wish
to say that these two orderings are inverse to each other if the majority vector of this
profile is (0, 0, 0). The orderings xPiYPiz and zPjyPjx with i,j E N, for example, yield
the majority vector (0, 0, 0); so do the orderings xPiyIiz and zIjyPjx. In both cases,
therefore, we say that the individual orderings are inverse.
More interestingly, the majority vector (0, 0, O) is also obtained in the situation of
a complete P-I cycle; by this we mean a P-I cycle where each of the three possible
P-I orderings is held by exactly one individual. The same result is, of course, obtained
for a complete I-P cycle.
In our analysis, we are primarily interested in qualitative or structural properties,
and this should also refer to reduced preference profiles. Consequently, the property of
dominance is of no great appeal to us. We shall therefore use the concept of necessity
in the sense in which it was used by Inada, Pattanaik, Sen, and others who were
looking for structural properties of originalpreference profiles, focusing only on types
of permissible preference orderings and not on numbers of individuals holding the
various orderings.
We now propose to transform the preference profile H of a given (original) society
into another more tractable one, denoted H*. This process of simplification (reduction
procedure) is performed by eliminating and/or combining certain orderings (or sets of
orderings) of H in such a way that the majority vector is left unchanged, the number
of types of orderings in H*, however, being considerably diminished (if a reduction
is at all possible). The following types of simplifications will be distinguished:
(pl) elimination of all pairs of individual P-I and I-P orderings which are inverse
to each other;
(P2) elimination of all complete P-I cycles and I-P cycles;
(p3) transformation of all remaining individual P-I orderings and I-P orderings into
strict preference orderings P-P. As an example, the ordering xPizliy, which
occurs ni-times, let's say, is transformed into the two strict orderings xPzPy
and xPyPz, each occurring ni/2-times. Analogously, the ordering yIjxPjz, which
occurs, let's say, nj-times, will be transformed into yPxPz and xPyPz, each
occurring nj/2-times.
(p4) elimination of all pairs of individual strict orderings which are inverse to each
other.
We wish to postulate that all four types of reductions are always performed in the
given order. The preference profile H of an original society is said to have obtained its
158 W Gaertner

reduced equivalent strict preference profile H1*if and only if all four reductions have
been executed, i.e. if

pi(7*) = I*, i e 1,2,3,4}.

Let Nj(Kj) be the number of individuals in H* who hold the type j ordering of
C+(C- ) in its strict form P-P.Then profile H* is characterized by the following simple
property:

Vj E {1,2,3}: Nj >0 -Kj= O.

We now wish to work with a notion of nearness or proximity of individual

preferences which uses the idea of binary inversions of adjacent alternatives. Given
any set of individual orderings making up society's preference profile, we construct the
reduced equivalent strict preference profile and ask whether it is possible to find a path
which links all remaining orderings in the following sense: moving from any one type
of ordering to the next requires one and only one inversion of adjacent alternatives.
Repeating this operation step by step leads from the first picked to the last type of
individual orderings, thus completing the reduced preference profile. Let us depict this
procedure with the help of the following profile:

x y z
y z
x z y

*, )
z x y

Such a path cannot be found for a latin square preference profile, for example. We
are now in a position to define what we call the binary inversion property.
Definition. The reduced equivalent strict preference profile of a given society is said
to satisfy the binary inversion property (BIP), if (trivially) this profile does not contain
more than one type of individual orderings, or if (non-trivially):

(NI + N2 + N3 ) (Ki + K2 + K3 ) > 0.

The latter requirement essentially captures the idea of a path composed of binary
inversions.
Theorem 16 [Gaertner (1988)]. The method of simple majority decision generates a
transitive social relation for a given preference profile HI if and only if its reduced
equivalent strictpreference profile II* satisfies the binary inversion propertyfor every
triple of alternatives.
Ch. 3: Domain Restrictions 159

For original unreduced profiles the condition of extremal restriction has been found
to be both necessary and sufficient for collective transitivity under the simple majority
rule. What is the logical relationship between condition ER and property BIP? It can be
proved [Gaertner (1988)] that if the original profile H fulfills extremal restriction, then
the reduced profile H* satisfies BIP. As for Saposnik's condition of cyclical balance, the
following result holds: an original preference profile satisfies the property of cyclical
balance if and only if the reduced set of preferences either is the empty set or the
reduced profile satisfies the condition (N1 + N 2 + N3 ) · (K1 + K 2 + K 3) > 0. Thus,
cyclical balance implies that there is a path of binary inversions of adjacent alternatives
within the reduced profile.
A sufficient condition for quasi-transitivity of the social preference relation can
easily be stated by modifying the non-trivial requirement in the definition of BIP. If
(N1 +N2 + N3) (Ks +K2 + K3) = 0 with any two of the N, I C {1,2, 3}, being strictly
positive and the third being zero (any two of the K1 being strictly positive and the third
being zero), then simple majority decisions generate a quasi-transitive social relation.
The reduced profiles H* which fulfill property BIP always satisfy both value restriction
and limited agreement, once these restrictions are reformulated for H*-profiles.
Slutsky (1977) presented sufficient conditions for social transitivity which shed a
somewhat different light on Sen's value restriction. Assume that there are only strict
individual preferences. Let m = min(ni), i = 1, ... , 6, be the smallest number of
individuals and g = max(n/), i = 1, ... , 6, be the largest number holding one of the
six strict orderings. Furthermore, let h(f) be the maximum number (minimum number)
of individuals who agree that one alternative in a given triple is strictly best or that
one is strictly worst, i.e.

h = max{N(aPib A aPic), N(bPia A cPia), for all (a, b, c) in (x,y, z)},

and

f = min{N(aPib A aPic),N(bPia A cPia), for all (a, b, c) in (x,y, z)}.

Then the following result can be stated.

Theorem 17 [Slutsky (1977)]. For a society consisting of n members with only strict
individual orderings, majority decision will be transitive if either h > ()n- m or
f < ()n - g; that is, if there is an alternative which is felt to be strictly best or one
which is felt to be strictly worst in the triple by either more than ()n - m or less
than (2)n - g individuals.
The conditions in this Theorem are, of course, also sufficient for quasi-transitivity
of the social preference relation. Weaker conditions, sufficient only for acyclicity over
the triple, require that the strict inequalities in the Theorem be replaced by weak
inequalities [for more details, see Slutsky (1975, 1977)].
160 W Gaertner

6. Social choice in continuous space

In the foregoing sections we studied the existence of social choice rules within the
framework of arbitrary finite sets of discrete alternatives. In this section, we shall
examine the existence of continuous aggregation rules in n-dimensional Euclidean
space. We imagine that social alternatives are described as vectors where each of the
n components is specified by a real number.

6.1. Impossibilities
Kramer (1973) has shown that in economic environments where the social alternatives
can be represented as points in an appropriately defined multi-dimensional choice
space, with the usual quasi-concavity and differentiability conditions satisfied by the
individual utility functions, the standard restriction conditions from Section 3 such
as value restriction, extremal restriction and limited agreement do not permit even a
modest degree of heterogeneity of tastes. Therefore, the transition from finite sets of
discrete alternatives to multi-dimensional choice spaces has serious consequences. We
shall now examine this problem in greater detail.
We consider a choice space X which is contained in the positive orthant of
the n-dimensional Euclidean space ItR. Let X be a cube in RI+. Each individual
preference pi is a C' vector field over the space of alternatives, which is locally
integrable. This means that to each alternative x in X, one attaches a vectorpi(x) in a
continuously differentiable fashion which indicates the direction of the largest increase
of utility. This direction is that of the normal to the tangent plane of the indifference
surface through x. Since ordinal preferences are considered, it is the direction of the
vectors rather than the length which is important. Therefore, all vectors are assumed
to be of length 1. The space of preferences P is defined as the set of all C' integrable
unit vector fields defined on the choice space X. There are k individuals in society.
A social aggregation rule for individual preferences is a function that assigns to each
k-tuple of individual preferences (each preference profile) a social preference in P. If
ri stands for a social aggregation rule, we have

:Px ... x P P
k times

Chichilnisky (1980, 1982) requires the rule cg to satisfy the following three
properties. First of all, the social preference has to define at each choice a most
desirable direction in a continuous manner. The continuity of (P is defined in terms of
convergence in the space of preferences which implies that proximity of preferences
in P is equivalent to the proximity of their indifference surfaces. Chichilnisky provides
several reasons for the continuity assumption, among them the argument that continuity
introduces a form of stability of society's choices with respect to small changes in the
preferences of the members of society.
Ch. 3: Donmain Restrictions 161

I(P1,P2)

I2

Fig. 1. A clash between continuity and unanimity.

Secondly, the choice rule (P is assumed to satisfy the condition of anonymity, i.e.
'i(PI,P2, ... , Pk) = qbP(r(1),Pr(2), ... , Pr(k)), where r(1), r(2), ... , r(k) denotes any
permutation of the set of integers {1,2, ... , k}. Thirdly, the choice rule is required
to respect unanimity, i.e. (P(p,p, ... , p) = p for all p e P. Respect for unanimity is
a condition which is weaker than the Pareto condition since it makes a requirement
only in the case where all preferences within a profile are the same. The condition of
anonymity is stronger than Arrow's nondictatorship condition [Arrow (1963)], because
it requires equal treatment of the individuals' preferences, while Arrow's condition only
forbids an extreme unevenness of treatment.
Given these three conditions, Chichilnisky arrives at an impossibility theorem that
will be stated below. Before doing this, we present a geometrical argument underlying
the negative result which is also due to Chichilnisky. Let the choice space X be two-
dimensional, a unit cube denoted j2. Furthermore, let the space of preferences consist
of linear preferences only. This space is denoted by Q. Now let o be the center of
X. Then each preference Q can be uniquely identified by a point on the circle Sl
(see Figure 1). Let us further assume that there are only two individuals. In what
follows, an aggregation rule is used that is some kind of an averaging rule. Let Pl
and P2 be two vectors in S1 and let I(pl,P2) be the unit vector in a direction that is
determined by half the angular distance between Pl and P2 in the clockwise direction.
Now letpl rotate clockwise around the unit circle S 1. Then aspl - P2, 1(pl,p2) must
converge to P3. On the other hand, due to unanimity, (Pi,P2) must also converge to
P2, as pi converges to P2. Thus, this anonymous choice rule suggests a clash between
continuity and unanimity. The impossibility can now be stated more generally.
Theorem 18 [Chichilnisky (1980, 1982, 1983)]. Any continuous social aggregation
rulefor smooth individualpreferences cannot simultaneously satisfy the conditions of
anonymity and respect of unanimity 12.

12The impossibility changes into a possibility when the requirement no longer is that a complete social
ordering exist on the choice space X but that there always be an optimal social choice x in X, i.e. a
continuous choice function exists with a unique outcome x [Chichilnisky (1993)].
162 W/ Gaertner

If the unanimity condition is replaced by a Pareto condition, this negative result can
be extended to preferences that allow for satiation.
Using the present topological framework, Baigent (2002) derived a comparable
impossibility result, both for social aggregation rules and choice functions, using the
conditions of anonymity and respect for unanimity on the one hand and a property
that he calls proximity preservation on the other. The author originally formulated the
latter condition within the finite framework and obtained impossibility results for a set
of discrete alternatives, both for social welfare functions and social choice functions
[Baigent (1987)].

6.2. Contractibility and existence results

We now wish to ask what kind of meaningful restrictions on individual preferences

and profiles of individual preferences can be identified such that for any finite number
of voters a social choice rule >P exists.
Let X again be a topological space and let I be the unit interval of the real line. Then
X is said to be contractible if there exists a point xo C X and a continuous function
F: X x I --+X such that F(x, 0) = x and F(x, 1) = xo for all x G X. The introduction
of the condition of contractibility into social choice theory is due to Chichilnisky and
Heal (1983). Roughly speaking, a contractible space is one that has no holes in it and
can, therefore, be contracted continuously through itself into one of its points. The unit
disk is, for example, contractible since it has no holes. A simple example of a space
which is not contractible is the circle in IR2, or any geometrical figure homeomorphic 13
to it (e.g. the boundary of a square or a triangle). The spaces that are being considered
in this subsection are parafinite CW complexes. They can be expressed as a countable
union of finite dimensional spaces.
Heal (1983) pointed out the relationship between contractibility and convexity. Any
convex space is contractible. This can easily be seen. Take a convex space X and a
point x, C X. By convexity, {(1 - t) x + t xo} E X for any x C X and any t E [0, 1].
If we now set F(x, t) = (1 - t). x +t xo, which is a map ofX x [0, 1] -- X, we obtain
F(x, 0) = x and F(x, 1) = xo for all x E X, as required in the definition of contractibility.
Any space X homeomorphic to a convex space C can be shown to be contractible. The
converse, however, is not true. As an example for the latter assertion, Heal mentions the
unit disk in R2 with a line segment attached to its boundary. Consider two topological
spaces A and B, A C B. We say that A is a retract of B if there exists a continuous
function f: B - A such that f(a) = a for all a A. The function f is said to be
a retraction. For spaces which are manifolds, polyhedra or parafinite CW complexes,
any contractible space is a retract of a convex space.

13 Two topological spaces A and B are said to be homeomorphic if there exists a function h: A - B
which is continuous, one to one, onto B, and has a continuous inverse. Intuitively speaking, two spaces
are homeomorphic if one can be continuously deformed into the other.
Ch. 3: Domain Restrictions 163

Heal focusses his discussion of the relationship between contractibility and

convexity on an aspect that is particularly important for the issue of aggregation. In a
space X, take a k-tuple of points (xl, ... , xk), each of which is inX and assign to them
another point y in X which is their aggregate. In mathematical terms, a map f from
the k-fold product of X with itself into X has to be constructed,f: X x ... x X X.
Let f satisfy continuity.
When X is convex, it certainly admits the following continuous aggregation rule:
k

f(x) = - Exi,
i=

which is convex addition. For convex spaces, such a type of aggregation is always
possible. For the class of topological spaces considered here, the fact that a contractible
space is a retract of a convex space makes it possible to assert the same for contractible
spaces. Consider Y C X, with X being convex and Y contractible, and let r: X - Y
be a retraction. We can then define the rule g: Y x ... x Y Y by

g(Y, ... , Yk) = r(f(yl,.. , Yk))-

Here the functionf maps the k-tuple (y1, .. , Yk) into a point in X, and r then pulls
this back to Y. Hence one can also aggregate continuously in contractible spaces. As
the reader will easily recognize, the continuous aggregation rule f(x) above satisfies
the properties of unanimity and anonymity. The mappingf(x) is a particular element
from the class of means. This mean obviously fulfills the three conditions Chichilnisky
requires to be satisfied by an aggregation rule in continuous space.
Chichilnisky and Heal (1983) distinguish between restrictions on the types of
preferences that can be held by individuals and properties of spaces of profiles
(condition on profiles). In both cases, the essential point is that the domain of
individual preferences P should be contractible. Consider preferences at a choice y
in Y. A preference is then fully determined by the normalized gradient vector Pi(Y)
giving individual i's preferred direction at y. If there are no restrictions at all on
individual preferences, then this normalized gradient vector may take any direction,
so the set of possible preferences at point y is isomorphic to the set of points on the
unit sphere centered at y. This is denoted by S"- , the (n - )-dimensional sphere in
]R" (see S' again in Figure 1). A restriction on individual preferences now takes the
form of the specification of a subset of S - 1 within which the normalized gradient
vector must lie. With no restrictions, the space of individual preferences at a point is
Sn - 1 , and this (n - )-dimensional unit sphere is not contractible, as we already know.
If, however, there is a convex cone C of directions, no matter how small, which no
individual may have as most preferred directions, the space of preferences becomes
pC = Sn - \(Sn- I n C), and this space is contractible. Contractibility then means that
there exists a continuous way of deforming the preferences so that at the end of this
deformation a single preference remains. Within the space of linear preferences, the
164 W Gaertner

contractible space pc is extremely general. The cone C can be arbitrarily small as

long as its interior is nonempty. For the case of general smooth preferences a similar
statement can be made. If P is the overall space of smooth preferences, define a
subspace P of the space P in the following way:

D= p E P: p(y) C for all y Y}.

As with pc above, P is a space of preferences such that all individuals agree that
there is one direction of preference which is totally undesirable. In particular, the space
of all smooth non-decreasing preferences is contained in P, and is contractible. The
same statement holds for the space of all smooth, convex, and monotone preferences.
Coming back to the distinction between restrictions on the type of individuals'
preferences and conditions on profiles, in the first case the requirement is that the
preferences in the space of preferences are sufficiently similar so that they can be
simultaneously continuously deformed to a single preference. In the second case, the
requirement is that within any admissible profile there must be sufficient agreement
that the preferences in that profile can be continuously deformed into one single
preference or a homogeneous profile. What is then required is a sufficiently high level
of agreement in any preference profile. Contractibility limits the types of disagreement
among individual preferences, thus determining the variation in the individuals'
preferences which can be accepted.
Let us now consider the theorems.
Theorem 19 [Chichilnisky and Heal (1983)]. Let the space of preferences P be a
parafinite CW complex and let k be the number of individuals. Then for all k > 2
a necessary and sufficient condition for the existence of a social choice rule from
the k-fold Cartesianproduct of P to P, which is continuous, anonymous and respects
unanimity, is that the space ofpreferences P be contractible.
The next result gives a characterization of the social aggregation rules that can be
defined on contractible spaces.
Theorem 20 [Chichilnisky and Heal (1983), Heal (1983)]. Let the preference space P
be contractible. Then any continuous anonymous social choice rule that respects
unanimity is equivalent to a rule f constructed as follows. Take a convex space C of
which P is a retract, with r the retraction, r: C P. Then define a convex averaging
rule on C, and letf be the composition of this with r.
Equivalence in the above formulation means the following. Two maps are equivalent
if they can be linked by a parameterized family of maps F(x, t) which are jointly
continuous in x and parameter t, where x E X, some topological space, and t e [0, 11.
For the case of P being convex, we have already indicated that
k
, Pk)
.(Pl = k-EPi,
Ch. 3. Domain Restrictions 165

would be an appropriate choice rule according to Theorem 20. If P is not convex but
contractible, then the composite map

P(Pl, ... k)= r o Pi

defines a continuous anonymous rule that respects unanimity 14.

6.3. Discrete us. continuous choices

The literature on continuous social choice has witnessed a longer debate on which
topology should be chosen for the analysis of collective choice problems. This issue
truly is an important and interesting one but instead of entering a longer discussion,
we wish to refer to Baigent's Chapter 17 in Vol. 2 of this Handbook (forthcoming)
[see also Gaertner (2001, Chapter 6)].
Equally important, perhaps more important for social choice theory is the question
whether the finite framework or the topological framework is the more appropriate
one to work in. An answer to this question involves at least two issues, that, however,
appear to be interrelated.
In our introduction, we referred to decisions on the composition of the federal
budget or decisions on the production of various public goods that can, more or
less, be conceived as continuous selections. Other examples would be choices over
a set of social states that are characterized by numerical indicators only, such as the
unemployment rate, the inflation rate and the change of a currency's exchange rate
versus other currencies.
On the other hand, very often, so it seems, the members of society are confronted
with a finite number of clearly specified options which represent different economic
and (or) social programs, and the individuals are asked to work out their own ranking
of these discrete alternatives. This task is being performed within a finite framework
with no trace whatsoever of a smooth substitution procedure. There definitely are lots
of discrete choice alternatives and the social outcome over these alternatives does not
(cannot by its very nature) change in infinitesimally small amounts. As typical social
choice examples, we have already referred to party programs, political candidates,
investment projects. But there are many more, for example, the creation of new
institutions such as the European Central Bank, the introduction of a new European
currency, or, during the cold war era, the construction of how many missile sites and/or

14 In a historical review and analysis, Horvath (1996) draws attention to the fact that, among others,
Aumann (1943) and Eckmann (1954) studied the existence of means. The latter, for example, stated that
contractibility is a necessary and sufficient condition for a connected polyhedron to have an n-mean for
each positive integer n. See also Horvath (2001).
166 W Gaertner

the deployment of how many nuclear warheads, where each side took great pains in
detecting and registering the numbers (integers!) on the other side.
Our argument here is that there are many discrete alternatives which are the object
of social choice. But there is a second aspect which cannot be separated from the
foregoing. And we now wish to refer again to Chichilnisky's argument that continuity
is a form of stability of society's choices with respect to small changes in individuals'
preferences. In the arena of politics, Weingast (1995) observed that "... policy
outcomes are not a smooth function of public opinion" (the italics are Weingast's).
And he continued that "explaining a major policy change ... requires not only an
examination of public opinion but how public opinion is aggregated via the relevant
institutions and, in particular, how it affects the identity and preferences of those who
control the agenda" (1995, p. 17) i5. Admittedly, Weingast's argument involves the
aspect of implementation and this is different from the aspect of evaluation (in terms
of a social welfare function or a social choice function), but both aspects are not
totally disjoint. Chichilnisky herself (1997, p. 122) emphasizes, when she introduces
her topological set-up, the institutional requirements that are needed to "organize
effectively a diverse society".
A related aspect is that political elections can produce narrow majorities which then
shape politics over a longer period. The point is not so much that an election is won just
by one vote (this is highly unlikely in a larger election) but that the distribution of seats
in the representative body is such that the winning party may have a majority of just
one seat 16 Such narrow majorities can also be brought about by coalition formation, a
procedure which is quite common within the political landscape of continental Europe.
Adenauer, the first chancellor of the Federal Republic of Germany, was elected by the
narrow margin of one vote (his own vote, people say). During his chancellorship quite
a few far-reaching decisions were taken for West Germany. In other words, narrow
margins among votes or elected representatives can produce vast differences for social
welfare judgments. In modern politics, there apparently is a multitude of cases in which
continuity of the choice procedure just cannot be found.
All in all, there seems to be room for more than one framework and for more
than one technique, but in each case arguments have to be given, and these should
be provided with considerable detail and care.

7. Concluding remarks
This chapter has discussed different types of domain restrictions. We began by analyz-
ing various qualitative conditions on preference profiles. Value-restricted preferences

15 On the aspect of agenda formation, see also Shepsle and Weingast (1982).
16 The most recent presidential election in the US has taught us that majorities in the popular vote can be
extremely narrow. Also elections in the electoral college can be very close. In 1876, the closest possible
occurred, a margin of one. In the current US Senate, Democrats have a 50-49 majority, giving them
control over committee chairmanships and the legislative agenda, again an aspect of implementation.
Ch. 3. Domain Restrictions 167

(with single-peaked preferences as one of its subcases) as well as antagonistic and

dichotomous preferences are relatively easy to interpret. In our view, the property
of single-peakedness stands out in particular. It proved to be a central restriction
under majority voting. However, it also plays an essential role in the context of
strategy-proof voting rules (which is the topic of another chapter in this Handbook).
Furthermore, we have considered quantitative or number-specific requirements on the
distribution of voters over different preference orderings, and we saw that some of
those requirements are logically related to the qualitative conditions. While the latter
restrictions are requirements on combinations of individual orderings, the domains
of individual orderings that admit n-person nondictatorial social welfare functions
a la Arrow result from restrictions on permissible preferences for individuals. While
the first five sections studied the aggregation problem within the framework of
arbitrary finite sets of discrete alternatives, the final section discussed continuous
choice rules; the alternatives were assumed to be n-dimensional vectors in Euclidean
space. Contractibility as a condition on the topological space of preferences proved to
be necessary and sufficient for the existence of continuous aggregation rules.

References

Arrow, K.J. (1951), Social Choice and Individual Values, 1st Edition (Wiley, New York).
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York).
Aumann, G. (1943), "ber Riume mit Mittelbildungen", Mathematische Annalen 119:210-215.
Baigent, N. (1987), "Preference proximity and anonymous social choice", Quarterly Journal of Economics
102:161-169.
Baigent, N. (2002), "Continuity and preference proximity", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) Chapter 17, forthcoming.
Batra, R.N., and P.K. Pattanaik (1971), "Transitivity of social decisions under some more general group
decision rules than the method of majority decision", The Review of Economic Studies 38:295-304.
Batra, R.N., and EK. Pattanaik (1972), "Transitive multi-stage majority decisions with quasi-transitive
individual preferences", Econometrica 40:1121-1135.
Black, D. (1948), "On the rationale of group decision making", The Journal of Political Economy
56:23-34.
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, Cambridge).
Blair, D.H., and E. Muller (1983), "Essential aggregation procedures on restricted domains of
preferences", Journal of Economic Theory 30:34-53.
Blau, J.H. (1957), "Social choice functions and simple games", Bulletin of the American Mathematical
Society 57:243-244.
Borda (J.-C. de Borda) (1781), "M6moire sur les elections par scrutin", Memoires de l'Acad6mie Royale
des Sciences ann6e 1781, pp. 657 665. Translated in English in 1953 by A. de Grazia: "Mathematical
derivation of an election system", Isis 44:42-51.
Bordes, G.A., and M. Le Breton (1989), "Arrovian theorems with private alternatives domains and selfish
individuals", Journal of Economic Theory 47:257-281.
Campbell, D.E., and J.S. Kelly (2000), "A simple characterization of majority rule", Economic Theory
15:689-700.
Chichilnisky, G. (1980), "Social choice and the topology of spaces of preferences", Advances in
Mathematics 37:165-176.
168 18W
Gaertner

Chichilnisky, G. (1982), "Social aggregation rules and continuity", Quarterly Journal of Economics
97:337-352.
Chichilnisky, G. (1983), "Social choice and game theory: recent results with a topological approach", in:
P.K. Pattanaik and M. Salles, eds., Social Choice and Welfare (North Holland, Amsterdam) Chapter 6.
Chichilnisky, G. (1993), "On strategic control", Quarterly Journal of Economics 108:285-290.
Chichilnisky, G. (1997), "A unified perspective on resource allocation: limited arbitrage is necessary and
sufficient for the existence of a competitive equilibrium, the core and social choice", in: K.J. Arrow,
A. Sen and K. Suzumura, eds., Social Choice Re-examined, Vol. 1 (Macmillan, London) Chapter 6.
Chichilnisky, G., and G. Heal (1983), "Necessary and sufficient conditions for a resolution of the social
choice paradox", Journal of Economic Theory 31:68-87.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Dasgupta, P., and E. Maskin (1998), "On the robustness of majority rule", Discussion paper (Department
of Economics, Harvard University, Cambridge, MA).
Dixit, A.K., and B.J. Nalebuff (1991), Thinking Strategically (Norton, New York).
Dodgson, C.L. (Lewis Carroll) (1876), A Method of Taking Votes on More than Two Issues (Clarendon
Press, Oxford). Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections (Cambridge
University Press, Cambridge) pp. 224-234.
Dummett, M., and R. Farquharson (1961), "Stability in voting", Econometrica 29:33-43.
Eckmann, B. (1954), "R/iume mit Mittelbildungen", Commentarii Mathematici Helvetici 28:329-340.
Fine, K. (1973), "Conditions for the existence of cycles under majority and non-minority rules",
Econometrica 41:889-899.
Fishburn, P.C. (1970), "Conditions for simple majority decision functions with intransitive individual
indifference", Journal of Economic Theory 2:354-367.
Fishburn, P.C. (1972), "Conditions on preferences that guarantee a simple majority winner", The Journal
of Mathematical Sociology 2:105-112.
Fishburn, P.C., and W.V. Gehrlein (1980), "The paradox of voting. Effects of individual indifference and
intransitivity", Journal of Public Economics 14:83-94.
Fishburn, PC., W.V Gehrlein and E. Maskin (1979), "Condorcet proportions and Kelly's conjectures",
Discrete Applied Mathematics 1:229-252.
Gaertner, W. (1977), "Zum Problem der Existenz von Sozialen Wohlfahrtsfunktionen im Sinne von
Arrow", Zeitschrift fr die gesamte Staatswissenschaft 133:61-74.
Gaertner, W. (1988), "Binary inversions and transitive majorities", in: W. Eichhorn, ed., Measurement
in Economics (Physica, Heidelberg) pp. 253-267.
Gaertner, W. (2001), Domain Conditions in Social Choice Theory (Cambridge University Press,
Cambridge).
Gaertner, W., and A. Heinecke (1977), "On two sufficient conditions for transitivity of the social
preference relation", Zeitschrift fuir Nationalrkonomie 37:61-66.
Gaertner, W, and A. Heinecke (1978), "Cyclically mixed preferences. A necessary and sufficient condition
for transitivity of the social preference relation", in: H.W Gottinger and W. Leinfellner, eds., Decision
Theory and Social Ethics (Reidel, Dordrecht) pp. 169-185.
Gehrlein, W.V (1983), "Condorcet's paradox", Theory and Decision 15:161-197.
Gehrlein, W.V. (1998), "Approximating the probability that a Condorcet winner exists", Discussion Paper
(University of Delaware, Department of Business Administration, Newark, DE).
Gehrlein, WV, and P.C. Fishburn (1976), "The probability of the paradox of voting: a computable
solution", Journal of Economic Theory 13:14-25.
Gehrlein, W.V, and P.C. Fishburn (1979), "Proportions of profiles with a majority candidate", Computers
and Mathematics with Applications 5:117-124.
Gibbard, A.F. (1973), "Manipulation of voting schemes: a general result", Econometrica 41:587-602.
Ch. 3: Domain Restrictions 169

Gillett, R. (1978), "A recursion relation for the probability of the paradox of voting", Journal of Economic
Theory 18:318-327.
Grandmont, J.M. (1978), "Intermediate preferences and the majority rule", Econometrica 46:317-330.
Guilbaud, G.Th. (1952), "Les theories de l'intert gn6ral et le problme logique de l'agr6gation",
Economie Applique 15:502-584.
Heal, G.M. (1983), "Contractibility and public decision-making", in: P.K. Pattanaik and M. Salles, eds.,
Social Choice and Welfare (North Holland, Amsterdam) Chapter 7.
Horvath, Ch.D. (1996), "Some aspects of the mathematical background of the topological social
choice problem". Manuscript, preliminary version (D6partement de Math6matiques, Universit de
Perpignan).
Horvath, Ch.D. (2001), "On the topological social choice problem", Social Choice and Welfare 18:
227-250.
Inada, K. (1964), "A note on the simple majority decision rule", Econometrica 32:525 531.
Inada, K. (1969), "The simple majority decision rule", Econometrica 37:490-506.
Inada, K. (1970), "Majority rule and rationality", Journal of Economic Theory 2:27-40.
Jain, S.K. (1986), "Semi-strict majority rules: necessary and sufficient conditions for quasi-transitivity
and transitivity", Unpublished manuscript (Jawaharlal Nehru University, Centre for Economic Studies
and Planning, New Delhi).
Kalai, E., and E. Muller (1977), "Characterization of domains admitting nondictatorial social welfare
functions and nonmanipulable voting procedures", Journal of Economic Theory 16:457-469.
Kalai, E., and Z. Ritz (1980), "Characterization of private alternatives domains admitting Arrow social
welfare functions", Journal of Economic Theory 22:23-36.
Kalai, E., E. Muller and M.A. Satterthwaite (1979), "Social welfare functions when preferences are
convex, strictly monotonic, and continuous", Public Choice 34:87-97.
Kaneko, M. (1975), "Necessary and sufficient conditions for transitivity in voting theory", Journal of
Economic Theory 11:385-393.
Kelly, J.S. (1974a), "Necessity conditions in voting theory", Journal of Economic Theory 8:149-160.
Kelly, J.S. (1974b), "Voting anomalies, the number of voters, and the number of alternatives",
Econometrica 42:239-251.
Kramer, G.H. (1973), "On a class of equilibrium conditions for majority rule", Econometrica 41:
285-297.
Le Breton, M., and J.A. Weymark (1996), "An introduction to Arrovian social welfare functions on
economic and political domains", in: N. Schofield, ed., Collective Decision Making: Social Choice
and Political Economy (Kluwer, Dordrecht).
Marglin, S.A. (1995), "Deconstructing individualism: a step towards an economics of community",
Mimeo (Harvard University, Department of Economics, Cambridge, MA).
Mas-Colell, A., and H. Sonnenschein (1972), "General possibility theorems for group decisions", The
Review of Economic Studies 39:185-192.
Maskin, E. (1975), "Arrow social welfare functions and cheat-proof game forms on restricted domains:
the two-person case", Discussion paper (Harvard University, Cambridge, MA, and the University of
Cambridge, UK).
Maskin, E. (1976), "Social welfare functions on restricted domains", Discussion paper (Harvard
University, Cambridge, MA, and Darwin College, Cambridge, UK).
Maskin, E. (1995), "Majority rule, social welfare functions, and game forms", in: K. Basu, PK. Pattanaik
and K. Suzumura, eds., Choice, Welfare and Development, Festschrift for Amartya Sen (Clarendon
Press, Oxford).
May, K.O. (1952), "A set of independent necessary and sufficient conditions for simple majority decision",
Econometrica 20:680-684.
McLean, I., and J. London (1990), "The Borda and Condorcet principles: three medieval applications",
Social Choice and Welfare 7:99-108.
170 W Gaertner

Pattanaik, PK. (1970a), "On social choice with quasi-transitive individual preferences", Journal of
Economic Theory 2:267-275.
Pattanaik, PK. (1970b), "Sufficient conditions for the existence of a choice set under majority voting".
Econometrica 38:165-170.
Pattanaik, PK. (1971), Voting and Collective Choice (Cambridge University Press, Cambridge).
Pattanaik, PK., and M. Sengupta (1974), "Conditions for transitive and quasi-transitive majority
decisions", Economica 41:414-423.
Riker, W.H., and P.C. Ordeshook (1973), An Introduction to Positive Political Theory (Prentice-Hall,
Englewood Cliffs).
Ritz, Z. (1983), "Restricted domains, arrow social welfare functions and noncorruptible and
nonmanipulable social choice correspondences: the case of private alternatives", Mathematical Social
Sciences 4:155-179.
Ritz, Z. (1985), "Restricted domains, Arrow social welfare functions and noncorruptible and
nonmanipulable social choice correspondences: the case of private and public alternatives", Journal
of Economic Theory 35:1-18.
Salles, M. (1974), "A note on Ken-Ichi Inada's majority rule and rationality", Journal of Economic
Theory 8:539-540.
Salles, M. (1975), "A general possibility theorem for group decision rules with Pareto-transitivity",
Journal of Economic Theory 11:110-118.
Salles, M. (1976), "Characterization of transitive individual preferences for quasi-transitive collective
preference under simple games", International Economic Review 17:308-318.
Saposnik, R. (1975), "On transitivity of the social preference relation under simple majority rule",
Journal of Economic Theory 10:1-7.
Satterthwaite, M.A. (1975), "Strategy-proofness and Arrow's conditions: existence and correspondence
theorems for voting procedures and social welfare functions", Journal of Economic Theory 10:
187-217.
Sen, A.K. (1966), "A possibility theorem on majority decisions", Econometrica 34:491-499.
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San Francisco).
Sen, A.K., and PK. Pattanaik (1969), "Necessary and sufficient conditions for rational choice under
majority decision", Journal of Economic Theory 1:178-202.
Shepsle, K.A., and B.R. Weingast (1982), "Institutionalizing majority rule: a social choice theory with
policy implications", American Economic Review (Papers and Proceedings) 72:367-371.
Slutsky, S. (1975), "Majority voting and the allocation of public goods", Ph.D. dissertation (Yale
University, New Haven, CT).
Slutsky, S. (1977), "A characterization of societies with consistent majority decision", The Review of
Economic Studies 44:211-225.
Tullock, G. (1967), "The general irrelevance of the general impossibility theorem", Quarterly Journal of
Economics 81:256-270.
Vickrey, W.S. (1960), "Utility, strategy and social decision rules", Quarterly Journal of Economics
74:507-535.
Ward, B. (1965), "Majority voting and alternative forms of public enterprises", in: J. Margolis, ed., The
Public Economy of Urban Expenditures (Johns Hopkins Press, Baltimore) pp. 112-126.
Weingast, B.R. (1995), "Self-enforcing democracy: its maintenance and collapse in antebellum America",
Manuscript (Hoover Institution and Stanford University, Department of Political Science, Stanford,
CA).
Williamson, O.E., and T.J. Sargent (1967), "Social choice: a probabilistic approach", Economic Journal
77:797-813.
Wilson, R.B. (1972), "The game-theoretic structure of Arrow's general possibility theorem", Journal of
Economic Theory 5:14-20.
Chapter 4

VOTING PROCEDURES
STEVEN J. BRAMS
New York University

PETER C. FISHBURN
AT&T Labs-Research

Contents

Abstract 175
Keywords 175
1. Introduction 176
2. Voter preferences and social choice functions 180
3. Voting procedures for two candidates 183
4. Introduction to voting procedures for three or more candidates 186
5. Nonranked voting and dominated strategies 189
5.1. Examples 189
5.2. Voter preferences 190
5.3. Dominance between strategies 191
6. Strategic analysis of nonranked voting 193
6.1. Admissible strategies 193
6.2. Sincere voting and strategyproofness 195
6.3. Efficacy 197
6.4. Plurality ballots and the median outcome 198
7. Nonranked multistage voting: successive elimination 198
7.1. Examples 198
7.2. Binary multistage voting 199
7.3. Paradoxes 201
8. Condorcet choices and ranked voting 203
8.1. Condorcet combinatorics 203
8.2. Nonranked voting and Condorcet candidates 206
8.3. Condorcet voting procedures 210
9. Positional scoring procedures and Borda choices 212
9.1. Positional scoring procedures 212
9.2. Axioms 213
9.3. Paradoxes 214
9.4. In praise of Borda, mostly 216

Handbook of Social Choice and Welfare, Volume 1, Edited by KJ Arrow, A.K. Sen and K. Suzumura
© 2002 Elsevier Science B. V All rights reserved
174 S.J. Brains and PC. Fishburn

10. Point distribution procedures 218

11. Proportional representation 219
11.1. The Hare system of single transferable vote 220
11.2. Additional-member systems 223
11.3. Minimizing representational imbalance 225
12. Conclusions 225
References 227
Ch. 4: Voting Procedures 175

Abstract

Voting procedures focus on the aggregation of individuals' preferences to produce

collective decisions. In practice, a voting procedure is characterized by ballot responses
and the way ballots are tallied to determine winners. Voters are assumed to have
clear preferences over candidates and attempt to maximize satisfaction with the
election outcome by their ballot responses. Such responses can include strategic
misrepresentation of preferences.
Voting procedures are formalized by social choice functions, which map ballot
response profiles into election outcomes. We discuss broad classes of social choice
functions as well as special cases such as plurality rule, approval voting, and
Borda's point-count method. The simplest class is voting procedures for two-candidate
elections. Conditions for social choice functions are presented for simple majority
rule, the class of weighted majority rules, and for what are referred to as hierarchical
representative systems.
The second main class, which predominates in the literature, embraces all proce-
dures for electing one candidate from three or more contenders. The multicandidate
elect-one social choice functions in this broad class are divided into nonranked
one-stage procedures, nonranked multistage procedures, ranked voting methods, and
positional scoring rules. Nonranked methods include plurality check-one voting and
approval voting, where each voter casts either no vote or a full vote for each
candidate. On ballots for positional scoring methods, voters rank candidates from
most preferred to least preferred. Topics for multicandidate methods include axiomatic
characterizations, susceptibility to strategic manipulation, and voting paradoxes that
expose questionable aspects of particular procedures.
Other social choice functions are designed to elect two or more candidates for
committee memberships from a slate of contenders. Proportional representation
methods, including systems that elect members sequentially from a single ranked ballot
with vote transfers in successive counting stages, are primary examples of this class.

Keywords

voting systems, voting paradoxes, social choice, Condorcet candidate, proportional

representation

JEL classification: D7
176 S.J Brains and PC. Fishhurn

1. Introduction

Voting procedures describe the manner in which the preferences of individuals are
aggregated to produce a collective decision. The individuals, whom we refer to as
voters, might be registered voters, legislators, trustees, committee members, jurors,
or members of some other body whose decisions not only are binding on their
members but often a larger community that the body represents. The alternatives
among which the voters choose will be referred to as candidates. Depending on the
context, candidates might be people running for office, passage or defeat of a bill,
alternative budgets, applicants for a faculty position, or jury verdicts that a judge
permits. Although we refer throughout the chapter to voters and candidates, it should
be clear that voting procedures cover a multitude of voting situations that are often
described in other ways.
As the term is used in this chapter, a voting procedure is defined by two
characteristics. The first is the type of vote, or ballot, that is recognized as admissible
by the procedure. This could range from an open show of hands in an assembly to an
anonymous best-to-worst ranking of all the candidates by the voters (a secret ballot).
We denote by (A) the set of admissible ballots for a given procedure in which A
is the set of feasible candidates. Given A, it is assumed that every voter selects a
member of B(A) as his or her vote, or ballot. When there are n > 2 voters indexed
by i = 1, 2, ... , n, and voter i selects di E (A), the n-tuple d = (dl,d 2, ... , d,) is
the ballot response profile. If there are no restrictions on voting patterns, any n-tuple
in B(A)" might occur as the ballot response profile. If each voter can vote for only
one candidate, for example, a ballot response profile would indicate the candidate for
whom each of the n voters voted.
The second defining characteristic of a voting procedure is how votes are counted
to determine a winner or winners. For this purpose, we need a concrete rule that
aggregates the individual responses in a ballot response profile into a collective
choice or measure (possibly numerical). The criterion by which an outcome is chosen
depends on the collective measure. The rule that whichever of two candidates obtains
a simple majority is a familiar voting procedure. Less familiar is a procedure that
involves successive elimination of lowest-vote candidates and transfers of their votes
to candidates that remain in contention, but it is one that is used in both public and
private elections.
Several important topics are not discussed in this chapter. One is the determination
of eligible voters, which may involve registration, committee membership, or random
choice, as in some jury-selection procedures. Another is the determination of feasible
candidates or official nominees. We will also not discuss agenda formation, use
of polls, campaign finance, or ballot-stuffing and other forms of election fraud. In
addition, strategizing by candidates in campaigns, which can be influenced by the
voting procedure, will not be treated here; it is discussed in Brams and Davis (1973,
1974, 1982), Lake (1979), Enelow and Hinich (1984, 1990), Cox (1984, 1987a,b,
Ch. 4: Voting Procedures 177

1997), Snyder (1989), Riker (1986, 1996), Coughlin (1992) and Myerson (1995a),
among other places.
On the other hand, we will pay attention to strategies that voters employ in making
ballot choices. Their strategic voting can substantially affect election outcomes and,
on occasion, subvert the intention of a voting procedure to treat voters and candidates
fairly. The topic of voter strategy, which is also discussed in Chapters 6, 10 and 11 in
this Volume, and Chapters 15, 23 and 25 in Volume 2 of this Handbook, is often tied
to voter preferences, and we will make that tie-in here.
In the next section, we present our basic assumptions about voter preferences over
the candidates. We define a voter preference profile as an n-tuple of voter preference
orders indexed in the same way as the ballots in a ballot response profile. Because
the structures of ballots and preference orders can be quite dissimilar, we presume no
obvious connection between them. This may even be true when the ballot set 13(A) is
the same as the set of preference orders on A, because some voters might cast ballots
different from their true preference orders in order to secure an outcome they prefer
to that produced by sincere or honest voting.
The theme of strategic voting is a theme of individual or subgroup choice within
a process of group choice. Roughly speaking, if a voting procedure sometimes
allows a voter to secure a preferred outcome by voting in a way that, in isolation,
clearly contradicts his or her true preferences, the procedure is said to be susceptible
to strategic manipulation. A voting procedure that is not susceptible to strategic
manipulation is said to be strategyproof: Common voting procedures used in selecting
between two candidates are essentially strategyproof, but most procedures involving
three or more candidates are not. The following example, which is motivated by Black
(1958) and Farquharson (1969), illustrates the latter point. The approach to strategic
voting discussed here was pioneered by Farquharson (1969).
Example 1.1. We consider a common legislative voting process in which m candidates,
ordered as ala2 ... am, are voted on in a succession of m - 1 pairwise simple majority
votes. The first vote is between al and a 2. For j > 1, the j + 1st vote is between aj, I
and the winner of the jth vote. The winner of the final vote is the overall winner. It is
often true that a candidate's chances of being the overall winner increase the later it
enters the process, for it then has to defeat fewer other candidates to emerge victorious.
There are exceptions, however. Consider three candidates, a, b, and c, and three voters,
1, 2, and 3, who have the voter preference profile

(abc, cab, bca).

This signifies that 1 prefers a to b to c, 2 prefers c to a to b, and 3 prefers b to c to

a. Assume that all voters know one another's preferences and that successive majority
voting applies with voting order abc (or, equivalently, bac). The first vote is between
a and b:
(i) if a wins, then c wins the second vote because 2 and 3 will vote for c over a;
178 SJ Brains and PC. Fishburn

(ii) if b wins, then b wins the second vote because 1 and 3 will vote for b over c.
Because b or c is the overall winner, and voters 1 and 3 prefer b to c, it is in their
interests to prevent (i) by ensuring that b defeats a on the first vote. Consequently, both
1 and 3 will vote for b on the first vote, even though 1 prefers a to b. In effect, voter
manipulates the procedure by voting strategically - in apparent contradiction to his
or her true preferences - to secure a preferred outcome. The vote by 1 for b on the
first vote, and for his or her preferred candidate on the second vote, is an example of
what Farquharson (1969) calls a sophisticated strategy. Similar reasoning shows that a
wins if the voting order is cab (or acb), and c wins if the voting order is bca (or cba).
Whichever order is used, the last candidate is a sure loser under strategic voting. U
In Example 1.l with voting order abc, a ballot in B({a, b, c}) can be defined as a
triple (a, i3, y) in which a denotes a first vote (a, b, or 0, where 0 denotes abstention),
3 e {a, c, 0} for the second vote when a wins the first vote, and y C b, c, 0} when b
wins the first vote. The aggregation rule tallies the votes in each position and specifies
the winner by simple majority comparisons. We have ignored tied-vote outcomes,
which can be factored into the decision criterion if a tie-breaking rule is adopted.
After we discuss individual preferences in the next section, we formalize the notion
of an aggregation rule and its decision criterion by defining a social choice function
as a mapping from ballot response profiles to subsets of feasible candidates. Particular
types of social choice functions are examined in ensuing sections. Section 3 focuses on
binary or two-candidate voting procedures, Sections 5 through 9 discuss social choice
functions for elections of one candidate from among three or more contenders, and
Sections 10 and 11 consider procedures for electing two or more candidates.
Section 4 provides further introduction to elections among three or more candidates.
We will see that there are deep mathematical results behind the enduring fascination
with multiple-candidate voting procedures. Briefly put, while there are innumerable
voting procedures for such elections, all are flawed. The difficulties arise from two
observations. The first, due to Arrow (1950, 1951), is that no social choice function
for three or more candidates simultaneously satisfies a few conditions that can be
viewed as desirable properties of such functions. The second, due to Gibbard (1973)
and Satterthwaite (1975), says that all reasonable voting procedures for three or more
candidates are susceptible to strategic manipulation. More recently, Saari (2001b),
Chapter 25 in Volume 2 of this Handbook, argues that elections can be "chaotic."
The effects of these results on the theory of voting have parallels to the effects
on physics of Heisenberg's uncertainty principle and the effects on logic and the
foundations of mathematics of G6del's incompleteness and undecidability results. The
challenge is not to design a perfect voting system, which is impossible, but rather to
identify those procedures that reflect the desires of voters in as faithful a manner as
possible. Among other things, we would like a voting procedure to encourage sincere
balloting (based on true preferences), be relatively immune to strategic manipulation,
and avoid egregious anomalies or paradoxes, such as the negative responsiveness
paradox that occurs when increased support for a candidate turns it from a winner into
Ch. 4: Voting Procedures 179

a loser. Our analyses of voting procedures in later sections will pay close attention to
these matters.
In doing this, we follow a tradition that goes back to the development of procedures
for conducting democratic elections in ancient Greece and Rome [Stavely (1972)].
Among others, Aristotle, in his Politics, gave considerable attention to better and worse
forms of government, including representative democracies.
Many centuries later, two Frenchmen, Jean-Charles de Borda and the Marquis de
Condorcet, argued on a more modest level for rather different election rules that still
bear their names. Their differences are recounted by Young (1988, 1995), Moulin
(1988a) and Saari (1995a). Borda (1781) [translated by de Grazia (1953)], for example,
advocated the ranked voting procedure in which ballots are complete rankings of
the m candidates, and m - 1, m - 2, ... , 1,0 points are awarded to the best-to-worst
candidates on each ballot. The candidate with the greatest point total wins. When it was
pointed out to Borda that his procedure is quite susceptible to strategic manipulation,
he is said to have replied that his scheme was intended only for honest men [Black
(1958, p. 238)].
Condorcet (1785) took the position that if, based on ranked ballots, one candidate
would defeat every other candidate in pairwise simple majority comparisons, this
candidate, called the majority or Condorcet candidate, should be elected. He showed
not only that Borda's scheme can violate this rule but also that there are ballot response
profiles with a majority candidate who would not be elected by any point-assignment
method that awards more points to a top-ranked candidate than a second-ranked
candidate, more points to a second-ranked than a third-ranked candidate, and so forth
(Section 9.3). Indeed, Condorcet pointed out that there may be no majority candidate,
as in (abc, cab, bca) of Example 1.1, but he was unclear about how to proceed when this
occurs. The nonexistence of a majority candidate has come to be known as Condorcet's
paradox, the paradox of cyclical majorities, and the paradox of voting.
The writings of Borda and Condorcet initiated a huge literature on voting procedures
[McLean and Urken (1995)]. Prominent nineteenth century examples include Nanson's
(1883) extensive review of voting procedures and Hare's (1861) book on the election of
representative legislatures. Hare's system, which is more widely known as the method
of single transferable vote (STV) or instant runoff (in Australia, it is known as the
alternative vote), was proposed as a way to ensure the representation of significant
minorities. With various modifications, STV has been adopted throughout the world to
elect public officials and representative assemblies, but in some jurisdictions it has been
abandoned, including several cities in the USA. We discuss it further in Section 11.
Two of the most significant developments of the twentieth century are Arrow's (1950,
1951) celebrated "impossibility theorem" and the analysis of elections and voting
procedures by methods of game theory, as represented in Farquharson (1969), Peleg
(1984), Brams (1985), Moulin (1988a, 1994), Coughlin (1992), and chapters in this
Handbook that deal with strategic voting.
The discussion of voting procedures that follows is indebted to a host of
predecessors, including many cited above. We assume throughout that voters act to
180 S.Jd rans and PC. Fishburn

maximize the satisfaction of their preferences, subject to the rules of voting and
what voters believe about other voters' preferences and likely behaviors. In analyzing
voting procedures, we will pay special attention to both their successes and failures in
producing social choices that are responsive to the preferences of voters.

2. Voter preferences and social choice functions

We presume that there are n > 2 voters, indexed by i = 1,2, .. , n, and a set X
of two or more candidates. Unless we note otherwise, X is taken to be finite with
cardinality IXJ, and voter i is assumed to have a preference weak order ti on X,
i = 1,2, ... , n, so that i is transitive (x i y and y i z imply x ti z) and complete
(x ti y or y ti x for all x,y c X). The interpretation of x i y is that voter i prefers
x to y or is indifferent between x and y. We denote the asymmetric (strict preference)
part of ,i by >-i, and the symmetric (indifferent) part of i by i:

x >-i y if x i y and not (y ti x),

x i y if x i y and y i x.

It follows from weak order that x -i x, (x i , >-i z) = x >-i z,

(x >-i y,y i z) = x i z, and that exactly one of x i y, y >-i x, and x i y holds
when x y. Moreover, the indifference relation -i on X is an equivalence relation
(reflexive: x i x; symmetric: x i y = y -i x; transitive: x i y and y i z imply
x -i z) that partitions X into r indifference classes Xi, X2, ... , X,. such that -i holds
within each Xj and x >-i y or y >-i x whenever x and y are in different classes. The
classes can be ordered by preference as Xl -i X2 >-i ... -i XF, where A -i B means
that a >-i b for all (a, b) C A x B.
When x -i y 4> x = y, for all x,y C X, we refer to i or its asymmetric part >- as
a linear order or strict ranking and abbreviate xl -i x2 >-i . >-i x,, as xlx2 m
(with i implicit or explicit). The three-candidate set X = a, b, c} with IX I = 3 admits
13 weak orders, including a - b - c - a, of which six are linear orders, namely abc,
acb, bac, bca, cab and cba.
An n-tuple = ( , t2, .. , n,) of weak orders on X, one for each voter, is a voter
preference profile. We let V denote the nonempty set of voter preference profiles that
are considered as possible voter preference profiles in a particular situation. If IXI = 3
and there are no restrictions on V apart from weak order, then VI = 13 if all voter
f;

preference relations are assumed to be strict rankings, then IVI = 6". If X = {a, b, c},
if voter preference relations are strict rankings, and if preferences are single-peaked
in the order abc (so that b is never least preferred), then VI = 4".
Three inputs determine the domain of a social choice function. The first is the
number k > 1 of candidates to be chosen by a voting procedure. The second is a
nonempty set X of subsets of X, each of which might arise as the feasible set of
candidates or the official set of nominees. We require AI > k for every A C X. The
Ch. 4: Voting Procedures 181

third input is the set DA of ballot response profiles that can occur for each A C X.
Each member of DA is a ballot response profile d = (dl, d2 , ... , d,), where di e 3(A)
for each voter and B(A) is the set of admissible ballots. The domain of a social choice
function is the set

D= U {(A,d): d DA}
AEX

of all ordered pairs (A, d) of a set A of feasible candidates and a ballot response profile
for that set.
A social choice function is a mapping F from a domain D into the family 2 of
subsets of X such that, for all (A, d) C D,

F(A, d) A and IF(A, d)l > k.

We interpret F(A, d) as the subset of feasible candidates chosen by the voting procedure
for situation (A, d). It is assumed to contain at least k candidates. When IF(A, d)l = k
for all (A, d) c D, we say that F is decisive. By not imposing decisiveness, we allow
for the possibility that a choice set contains more candidates than the precise number
to be elected, i.e., we admit the possibility of unresolved ties. This may be inimical
to practical necessity, but it has the technical advantage of sidestepping issues of tie-
breaking procedures.
We now define several conditions for social choice functions that will be involved
in discussions of specific voting procedures. We begin with domain aspects and then
consider anonymity, neutrality, and Pareto-dominance properties.
Given k as defined above, we refer to F as a choose-k social choice function. We
focus on choose-1 procedures until the final two sections.
The cardinality of feasible candidate sets might also be fixed, and when A I = m for
all A X, we say that F is an m-ary social choice function. When F is m-ary, we
assume that m > k. A voting procedure designed to choose two of five nominees is a
5-ary choose-2 procedure.
Most procedures have the same ballot-set structure for all A X of the same
cardinality. When this is true, we say that the ballot sets are similar. More precisely,
given A,B E X for which IAI = IBI, a ballot do 3(A), and a one-to-one mapping a
from A onto B, we denote by o(do) the action of a on do caused, in effect, by replacing
every instance of a in do by (a), for all a A. For example, if A = {a, b, c},
B = {a, b,x}, (a) = b, (b) = x and (c) = a, and if do is the strict ranking bac
of A, we have (do) = xba. Then the ballot sets are similar if

B(B) = {o(do): do E B3(A)}

for all A,B C X for which AI = BI and all bijections a from A onto B. Note that this
definition is nonvacuous even when X contains only one set, say X = {A}. It says, for
182 S.JI Bramns and PC. Fishburn

example, that if A = {a, b, c} and if abc is a ballot, then all permutations of abc are
ballots. A situation in which similarity fails occurs when all ballots are weak orders
or linear orders that are single-peaked with respect to a fixed left-to-right ordering of
the candidates.
Along with similar ballot sets, most voting procedures of practical interest allow
each voter independently to select any member of B(A) as his or her ballot, in which
case DA = B(A) " . We will assume this in all voting procedures discussed later, and in
the rest of the present section.
One implication of unrestricted ballot response profiles is that if d = (dl, .. ,d) is
in B(A)" =DA, and if p is any one-to-one mapping from {1, ... , n} onto {1, ... , n},
then pd, defined as (dp(l), ... , dp(n)), is also in DA. We say that F is anonymous if

F(A, pd) = F(A, d)

for all (A, d) C D and all p from {1, ... , n} onto { 1, ... , n}. Anonymity says that the
social choice set is invariant to permutations of ballots among voters, and it therefore
embodies the notion that all voters are treated equally. It fails if there is a tie-breaking
chairperson or if some voters' ballots are weighted more heavily than others.
The term neutrality is used to convey the notion that all feasible candidates are
treated equally. Assume that ballot sets are similar. Given A,B X with A = lBI
and a mapping a from A onto B, let (d) = ((dl), ... , (d,,)) for d E DA, and let
r(A') = I{o(a): a A'} for every A' C A. Then (d) E DB and or(A') C B. We say
that F is locally neutral if

F(A, (d)) = o(F(A, d))

for all (A, d) ) and all permutations a on A, and that it is globally neutral, or simple
neutral, if

F(B, a(d)) = (F(A, d))

for all A,B E X for which IA = BI, all d E DA, and all from A onto B. Neutrality
says that if the ballot response profile for B is obtained from the ballot response profile
for A under the action of a, then a is in A's choice set if and only if a(a) is in B's
choice set. It fails for the binary voting procedure in which a challenger needs a two-
thirds majority to replace an incumbent. Because the challenger has a bigger hurdle
in replacing the incumbent than the incumbent has in staying in office, they are not
treated equally.
Pareto dominance is usually defined with respect to voters' true preferences. We
consider an alternative definition based on ballot response profiles. To do this, it is
necessary to have a sense in which a ballot reveals that a voter prefers a to b even
when, because of strategic voting, the voter may actually prefer b to a. For example, if
a voter votes for a but not b on a nonranked ballot, or ranks a ahead of b on a ranked
Ch. 4: Voting Procedures 183

ballot, we may say that the voter expresses a preference for a over b. We denote this
by adib for voter i, and for a, b E A and d E DA write adb if adib for i = 1, ... , n.
Because adb indicates unanimous preference for a over b, we consider the following
dominance condition for choose-i voting procedures:

{a, b E A, (A, d) E D, adb} = b F(A, d). (2.1)

A similar condition for k > 2 may be inconsistent with IF(A, d)l > k. We can, however,
require that a C F(A, d) when adb and b C F(A, d), so that if a dominated candidate is
in the choice set, then every feasible candidate that dominates it is also in the choice
set. This reduces to the former choose-1 condition for k = 1 when F is decisive. We
will see in Section 7.3 that it is violated by the type of multistage voting procedure
defined in Example 1.1.

3. Voting procedures for two candidates

We assume throughout this section that F is a binary choose-i voting procedure with
X = {a, b} and X = {X}. For algebraic convenience, we encode the nonempty subsets
of X as follows:
1 signifies {a}
0 signifies {a,b}
-1 signifies {b}.

We take 1 = {1, 0, -1} and D = {1, 0, -1}, where 1, 0 and - denote a vote for a, an
abstention (or a vote for both a and b), and a vote for b, respectively. The social choice
function F, with F(d) = F({a, b}, d), maps D into {1, 0, -1, where F(d) equals 1, 0,
and -1 according to whether a wins, a and b tie, and b wins, respectively.
The convenience of our encoding is seen in part by expressions of potential
properties for F. Using terms defined in the preceding section, F is decisive if F(d)
never equals 0, F is anonymous if F(d) = F(pd) for all d e D and mappings p from
{1, ... , n} onto {1, ... , n}, F is neutral if F(-d) = -F(d) for all d D, and F
satisfies the dominance condition (2.1) if F(j, ... , j) =j forj E {1,-1}.
Let d > d' mean that di > d for all i, and let d > d' mean that d > d' and not
(d' > d). We then say that F is monotonic if

d > d' F(d) >F(d') for all d,d' E D,

and that F is strongly monotonic if it is monotonic and, for all d > d' in D,

F(d') = 0 = F(d) = 1; F(d) = 0 = F(d') = -1.

Monotonicity conditions are often referred to as nonnegative orpositive responsiveness

conditions. They formalize the crucial idea for voting procedures that increased support
for a candidate never hurts, and may help it to win.
184 S.. Brains and PC Fishburn

The overwhelmingly most common binary choose-i voting procedure is simple

majority, with or without a tie-breaking rule. Let s: R {1, 0, -1 } be the signfunction
defined by

s(r)=l1 if r>0, s(O)=0, s(r)=-1 if r < 0.

The simple majorityfunction is defined by

F(d)= s ( di) for all d ED.

If F is to be decisive and all tied votes are broken in a's favor, we can take

F(d) = s d + for all dD.

This violates neutrality because neutrality implies F(0, .. , 0) = 0 so long as

abstentions are allowed.
May (1952) axiomatized the simple majority function in one of the earliest
characterizations of a voting procedure by properties it possesses:
Theorem 3.1. F is the simple majorityfunction if and only if it is anonymous, neutral,
and strongly monotonic.
We next consider other binary procedures that relax one or more of May's conditions.
Monotonicity, but not necessarily strong monotonicity, will apply throughout.
The most common voting procedures that violate anonymity are weighted majority
functions. Let wi > 0 denote a weightfor voter i, let W = {w = (wl, ... , w, ) : wi > 0
for all i and _ wi > 0}, and denote by (w, d) the inner product Z widi of w and d.
We say that F is a weighted majority function if there is a w E W such that

F(d) = s((w, d)) for all d E D.

Simple majority, with wl = w2 = w,, is the only anonymous weighted majority

function. Simple majority with a tie-breaking chairperson, voter 1, can be characterized
by w = (3, 2, 2, ... , 2).
The usual setting for weighted majority is a representative body in which voters have
different sized constituencies. A common concern for such bodies, which is intimately
connected to notions of voting power and fair representation [Banzhaf (1965), Shapley
and Shubik (1954), Dubey and Shapley (1979), Balinski and Young (1982, 2001),
Felsenthal and Machover (1998), Chapter 8 in this Volume, Chapter 26 in Volume 2
of this Handbook)] and has precipitated many court cases, is what weights to assign
voters.
Ch. 4: Voting Procedures 185

Neutrality says that if d l , d2 E D and d + d2 = (0, ... , 0), then F(d') =

1 X F(d2 ) = -1. We extend this for weighted majority by saying that F is strongly
neutral if for all m > 2 and d', ... , d m E D for which d1 + + d = (0, ... , 0),
F(d j ) = 1 for some j <m F(dk) =
-
for some k<m.
The following characterization is from Fishburn (1973):
Theorem 3.2. F is a weighted majorityfunction if and only if it is monotonic, strongly
neutral, and satisfies condition (2.1).
The most common violators of neutrality are special majorities. An example is the
decisive two-thirds majority rule defined by
F(d) = s (I{i: di = 1} - 2n +

where the challenger a wins if and only if more than two-thirds of the voters
vote for a.
A generalization of weighted majority functions was introduced by Murakami (1966,
1968) and further studied by Fishburn (1971, 1973, 1979b), Fine (1972) and Keiding
(1984). Fishburn refers to these procedures as representative systems. Beginning
from the projections pi(d) = d, the family of representative systems is defined
recursively by nested hierarchies of weighted majorities under the s operator. With
s(rl, r2 , ... , r) = s(Z rj), an example is

F(d) =s(s(dl, ... , d), s(dq+ ,..., d,), s(d+, ., dn))

This characterizes a tricameral legislature in which each house votes by simple
majority. The overall winner is determined by simple majority applied to the vote
outcomes of the houses.
To be more precise, let R 0 = {dl, ...., d, } and for each positive integer t let

Z,= {s(fi,...,fK):K E{1,2, ... } and fl,...,fK E t-1}.

It is easily checked that R1Z is the set of weighted majority functions and, because
s(f) = f, we have Ro C ZRIC 72 C ... We refer to R = URi as the set of
representative systems for n voters.
Complete characterizations of representative systems first appeared in Fishburn
(1971) and Fine (1972). A key aspect of their characterizations is a relaxation of strong
neutrality that we refer to as the dual partition condition. We say that voter i is essential
if there is some (dl, ... , d,_ . .., d,) E {1,0, -1}n for which it is false that
F(dl, ...., d 1,x, d+ , ...., d) is the same for each x in {1, 0, -1}. Then F satisfies
the dualpartition condition if F(dk) = 1 for some k E {1, ... , m} whenever m is an
odd positive integer,
d > _d2, d2 > _d3 ., dm- l > -d, d > -d,
>
and Ek'= dk for some essential i.
186 S.J. Brams and PC. Fishburn

Theorem 3.3. F is a representativesystem if and only if it is monotonic, neutral and

satisfies condition (2.1) and the dualpartition condition.
A proof of Theorem 3.3 is included in Fishburn (1973, Chapter 4).
Further studies of representative systems have focused on the number of hierarchical
levels needed to express all F £ 1Z for n voters. Let Al(n) denote the smallest t for
n voters such that R, = 7R. Fishburn (1979b) notes that /l(n) = n - 1 for 1 < n < 4,
pU(5) = pu(6) = 4, u(n) < n - 2 for all n > 6, and p is unbounded. He also conjectured
that lu(n)/n- 0 as n gets large. Keiding (1984) confirmed the conjecture by proving
that p(n) < log2 (n(n - 1)) + 5.

4. Introduction to voting procedures for three or more candidates

The monotonic binary voting procedures of the preceding section are strategyproof,
because a voter can never help elect a preferred candidate by voting contrary to his
or her true preferences. A very different picture of strategic voting emerges when
there are three or more feasible candidates. While virtually all voting procedures
for multicandidate elections are susceptible to strategic manipulation, we will see in
ensuing sections that some are more manipulable than others.
In this section we indicate why elections with three or more candidates can be
problematical. We begin with issues familiar to Borda and Condorcet and then consider
Arrow's impossibility theorem from the perspective of voting procedures.
Example 4.1. Suppose ranked ballots are used in a four-candidate election among a,
b, c, and x, and there are 13 voters with a ballot response profile in which

4 voters have axbc

3 voters have caxb
6 voters have bcax.

Majority comparisons of expressed preferences show that

a has a 7 to 6 majority over b

b has a 10 to 3 majority over c
c has a 9 to 4 majority over a,

so majorities cycle among a, b, and c: there is no candidate that beats the other two
in pairwise contests. (As for x, everyone prefers a to x, c has a 9 to 4 majority over x,
but x has a 7 to 6 majority over b.) Although there is no majority candidate, making
Condorcet's choice criterion inapplicable, if we count the number of majority wins for
each candidate, a and c come out on top with two apiece. Because c is preferred to
a by nearly 70% of the voters, c would appear to be the best social choice from a
Condorcet perspective.
Ch. 4: Voting Procedures 187

What about Borda? When 3, 2, 1 and 0 points are awarded to the candidates in best-
to-worst order for each voter, the Borda point totals for a, b, c, and x are 24, 22, 21,
and 11, respectively, so a is the top Borda candidate. While everyone prefers a to x,
however, one might reasonably suspect that a's supporters nominated x to inflate a's
Borda score.
Why? Suppose x is disqualified for any of the following reasons: x is certified as a
ringer or Doppelginger; the rules edit out all Pareto-dominated candidates before the
Borda count is taken; x drops dead before the votes are counted. With x deleted, the
Borda point totals for a, b, and c are 11, 16, and 12, respectively. Not only is b the
top Borda candidate, but a also comes in last. ·
Arrow's seminal contribution was to show that there is no way around the inherent
difficulties of situations like Example 4.1. His theorem has several formulations and
interpretations. The one that follows seems most congruent with our emphasis on
voting procedures.
Let X 2 denote the family of two-candidate subsets of X with IX[> 3, and assume
that every member of X2 is a potential feasible set, so X2 C X. For every {x,y} G X2,
assume D{x,,} = {{x}, {x,y}, {y}}n, as in the preceding section. We do not assume
that the restriction of F to any {x,y} x D{,y} is anonymous, neutral, or monotonic,
and we do not presume that the same voting procedure is used for every {x,y} E X2.
We do, however, require F on {A x DA: A G X 2} to satisfy two conditions, (2.1) for
Pareto dominance and a minimal nondictatorship condition:
Al: For every {x,y} E X 2, F({x,y}, (j, ... , j)) =jforj e {{x}, y}};
A2: For every i E {1, ... , n}, there is an {x,y} c X2 and a d e D{,y} such that
d = {x} and y E F({x,y}, d).
A2 says that for every voter there is at least one binary situation in which the voter
votes for x but x is not the sole member of the social choice set, as might be expected
if all the other voters voted for y.
Conditions Al and A2 are undemanding and accommodate a wide variety of
behaviors for F on {A x DA: A X2 }. The thrust of Arrow's theorem is that all
such behaviors are incompatible with a certain transitivity constraint on relationships
among binary choices.
To formulate his constraint, let {a, b, c} be a three-candidate subset of X, and let d 1,
d , and d3 be binary ballot response profiles in D{a,b}, D{a,c}, and D{b,c}, respectively.
2

We say that the triple (dl,d 2,d 3 ) is consistent if, for every i {1, ... , n},

di = d2=( or dd=
dia} b or di = {c}.

The sense of this definition can be seen by the fact that if, for example, di' = di2 ={a},
then voter i's ballots in the three binary cases are consistent with the hypothesis that his
or her true preferences on {a, b, c} are a -i b >-i c or a >-i c >-i b or a >-i b -i c, and
that he or she votes according to these preferences in each case. Arrow's transitivity
constraint can be expressed as follows:
188 SJ Brains and PC Fishburn

A3: For all three-candidate subsets a, b, c} of X, and all consistent (d', d2, d3 ) E
D{a,b} x D{a,c} x D{bc}, if

a E F({a,b},dl) and b E F({b,c},d3 ),

then a E F({a, c}, d2 ).

Given consistency, a E F({a, b}, d l ) and b e F({b, c}, d3) say that a is socially as good
as b, and b is socially as good as c according to F: the conclusion, a E F({a, c}, d2),
completes the transitivity triad by asserting that a is socially as good as c.
Theorem 4.2. If IXI > 3 and X2 C X, then F cannot satisfy all three of Al, A2 and
A3.
Arrow's theorem has been interpreted in various ways, but the interpretation we
prefer in the voting context is that it is unrealistic to suppose that binary voting
outcomes should be transitive in the sense of A3. A corollary is that even when
voters have weak preference orders and vote nonstrategically, there is no obvious
or compelling way to ground social choices (from feasible sets of three or more
candidates) on binary comparisons, whether determined by simple majority or in other
ways.
Theorem 4.2, or rather the original versions of Arrow's theorem in Arrow (1950,
1951), have motivated a vast amount of research on multicandidate elections. There
are now several dozen Arrow-type impossibility theorems that address a wide array of
social choice situations, but all have the same theme of the collective incompatibility of
conditions which, taken separately, seem reasonable and appealing. The books by Kelly
(1978) and Fishburn (1987), and Chapters 1, 2 and 3 in this Volume and Chapter 17
in Volume 2 of this Handbook, focus on these theorems.
We conclude our discussion of impossibility theorems by recalling the theorem of
Gibbard (1973) and Satterthwaite (1975) for strategyproof social choice functions. The
question they addressed is whether it is possible to design a decisive choose-1 social
choice function F on D = {X} x V that is nondictatorial and strategyproof when
XI > 3 and V is the set of all n-tuples of weak orders on X. Unlike Theorem 4.2, X
has only one member, namely X. We let

X* = {x E X: F(X, v) = {x} for some v V}

and say that F is nondictatorialif for every i {1, ... , n} there is a u = (L, .., ,
n)
in V and x,y X* such that x >-i y and F(X, v) = {y}. In addition, F is strategyproof
if for all v = (, ... , ) and v' = (,, .. , ,) in V and all i E {1, ... , n},

for all j i) > F(X, v) i F(X, v').

This implies that a voter can never unilaterally obtain a preferred outcome by voting
contrary to his or her true preferences:
Ch. 4: Voting Procedures 189

Theorem 4.3. Suppose F is a decisive social choice function on {X} x V, where V

is the set of all n-tuples of weak orders on X, and X* > 3. Then F cannot be both
nondictatorialand strategyproof
When Example 4.1 and Theorems 4.2 and 4.3 are compounded many times over by
related examples and theorems, it takes little imagination to conclude that there is no
such thing as a completely satisfactory voting procedure for elections among three or
more candidates.
In our ensuing discussion of specific multicandidate voting procedures, we make
several simplifying assumptions with no significant loss of generality. We assume that
F is an m-ary social choice function for m > 3, that all ballot sets B(A) for IAI = m
and A E X are similar, that DA = 13(A)n for each A E X, and that the same criterion
for membership in F(A, d) is used for all A C X.
We make a further concession to notation by taking X itself as the exemplary
m-candidate set for the purpose of defining each procedure, and use A, B, C, S, and
T to denote subsets of X. The ballot set will be denoted by 13 so that F is a mapping
from {X} x n3 into subsets of X with IF(X,d)l > k for every d = (d, ... , d,) in
13".

5. Nonranked voting and dominated strategies

We assume henceforth that XI > 3 and, until Section 10, that k = 1. This section and
Section 6 focus on the strategic analysis of nonranked voting procedures with a single
balloting stage. Section 7 will consider aspects of multistage nonranked voting, and
Sections 8 and 9 then treat procedures that use ranked ballots. The following definition
applies to this section and Section 6.
Definition 5.1. F on {X} x 13n with IXI = m is a nonranked voting procedure if there
is a nonempty subset M of {0, 1, . ., m - 1} that includes at least onej > 0 such that
B =ACX: AI E M}and, forall x E X and d Bn,

xEF(X,d)l{i: xE di}l >I{i:yEdi}l forall yX. (5.1)

Most M we consider include 0, which denotes abstention. We exclude m from M

(a vote for all candidates) because it has the same effect on Equation (5.1) as an
abstention. According to Equation (5.1), x is in the choice set if and only if as many
voters vote for x as for any other candidate. An example of a single-stage nonranked
procedure that does not adhere to Definition 5.1 is given at the end of Section 6.

5.1. Examples
Although nonranked voting procedures limit the ability of voters to express their
preferences by their votes, they are the most widely used procedures for multicandidate
elections. Some examples follow.
190 S.J. Brains and PC. Fishburn

Plurality voting has M = {0, 1}, so each nonabstaining voter votes for exactly one
candidate. It is commonly used in single-winner elections and sometimes in multiple-
winner elections. The main criticisms of plurality voting concern its severe limitation
on the expression of voter preferences, the dispersion of votes that it produces across
ideologically similar candidates - rendering them vulnerable to other candidates,
particularly on the ideological extremes, who have no opposition - and the extent to
which it encourages voters to vote for candidates other than their favorites when their
favorites have no real chance of winning. Axioms for plurality voting are given by
Richelson (1978) and Roberts (1991).
Vote for no more than t has M = {0, 1,..., t}. This is sometimes used for choose-1
elections with 1 t < k, but it is more common for choose-k (often k = t) elections.
When k < t, it is called limited voting. M = {O, k} is a more restrictive procedure.
Approval voting has M = {O, 1, ... , in - 1}, so a voter can vote for any proper
subset of X. It was named by Weber (1977), axiomatized by Fishburn (1978a) and
Sertel (1978) with a variable-n formulation (see Section 9.2), and is extensively
analyzed with comparisons to other procedures, including plurality voting, in Brams
and Fishburn (1978, 1983) and Merrill (1988). The analysis in this section and the
next is adapted from Brams and Fishburn (1983) and Brams (1994). Approval voting
has been adopted by several professional societies for elections of their officers.
Negative voting, which allows each voter either to vote for one candidate or to vote
against one candidate, is tantamount to M = { 1, m - 1} or to M = {0, 1, m - 1} when
abstentions are allowed. It was proposed by Boehm (1976) and is analyzed in Brams
(1977, 1978, 1983) and Felsenthal (1989).

5.2. Voter preferences

Additional definitions and assumptions for voter preferences are needed for our
ensuing analysis of strategic voting. We denote a voter's weak preference order on
X by A, with A t B for nonempty A,B C X if a t b for all (a, b) A x B. Strict
preference >- and indifference - are defined in the usual ways, and A >- B if a >- b
for all (a, b) E A x B. As in Section 2, the ordered indifference classes determined by
are X- X2 >- - >X,-.
We say that t is unconcerned if r = 1, dichotomous if r = 2, trichotomous if
r = 3, and multichotomous if r > 4. A voter is unconcerned if r = 1, and is otherwise
concerned. In the latter case, with r > 2, we say also that t is concerned.
A subset A C X is high for if
x E A nX Xi CA forall i<j,
and is low for t if
xEAnXjXiCA forall i>j.
Every subset of X is both high and low for an unconcerned voter, but only X and the
empty set 0 are both high and low for a concerned voter. It is easily seen that A is low
Ch. 4: Voting Procedures 191

for - if and only if its complement X \ A is high for t. The high sets for trichotomous
t on {xl, ... , xs} with

X = {xl}, X 2 = {x2 ,x 3 }, and X 3 = {X4,Xs},

areX, 0, {xl}, {xl,x2}, {x l ,x 3 }, {x 1,x 2 ,x 3 }, {x,X 2,x 3,x4}, {x,x 2 ,x 3,x 5 }. These can
be interpreted as the voting strategies this voter would consider viable - they are not
dominated by any other strategies, as we will see in Section 5.3 and describe further
in Section 6.1.
Assumptions that go beyond weak order are used in our analysis. They concern
preferences between subsets viewed as vote outcomes or social choice sets. For a
voter with weak order Z on X, APB means that outcome A is strictly preferred to
outcome B, and ARB means that the voter considers A at least as good as B. We assume
without further notice that {a}P{b} = a >-b, {a}R{b} = a b, and that APB and
BRA cannot both hold. In addition, we assume the following for all a, b E X and all
A,B, CC X:
Assumption P. If a >- b, then {a}P{a, b}P{b}.

Assumption R. If A U B and B U C are not empty, and if a t b, b t c and a c for

all a E A, b c B and c C C, then (A U B)R(B U C).
Assumption P asserts that if candidate a is preferred to candidate b, then
outcome {a} is preferred to the tied outcome {a, b}, which is preferred in turn to {b}.
This is uncontroversial if the voter believes that, however a tie might be broken, each
of a and b has positive probability of winning when F(X, d) = {a, b}. Assumption R
asserts that outcome A U B is at least as good as B U C when A t B, B t C and
A - C for the nonempty pairs from {A,B, C}.

5.3. Dominance between strategies

A strategy in the present context is any A C X, and a voter chooses strategy A if he

or she votes for all a E A and no other candidate. We defer consideration of feasible
strategies in B to Section 6 and focus here on a notion of dominance that applies to
all strategies and is therefore applicable to all nonranked voting procedures.
Roughly speaking, strategy S dominates strategy T for a voter if he or she likes
the outcome of S as much as the outcome of T in every possible circumstance, and
strictly prefers S's outcome to T's outcome in at least one circumstance. To make this
precise, we define a contingency as a functionf that assigns a nonnegative integer to
each candidate. A contingency is interpreted as specifying the number of votes each
candidate receives from all voters other than the voter for whom dominance is being
defined.
Call the latter voter the focal voter. Given a contingencyf and a strategy S for the
focal voter, let F(S,f) denote the outcome of the vote when the focal voter chooses
192 SI Brains and PC. Fishburn

S, presuming that the votes in S count. Thus, with S(a) = 1 if a E S, and S(a) = 0
otherwise,

a E F(S,f) X f(a)+S(a) >f(b)+S(b) for all b X

Although different strategies may be preferred under different contingencies, some

strategies are uniformly as good as, or better than, other strategies, regardless of the
contingency. That is, one strategy may dominate another.
Definition 5.2. Given P and R for a voter with weak order t on X, strategy S
dominates strategy T for this voter, or SdomT for A, if F(S,f)RF(T,f) for all
possible contingencies f, and F(S,f)PF(T,f) for at least one contingency.
It may be noted that if n is very small, say n = 2, then a contingency which
demonstrates F(S,f)PF(T,f) might be unavailable because it presumes more than
one other voter. However, even modest values of n avoid this possibility, and we shall
ignore it in what follows.
Assumption R implies that an unconcerned voter will be indifferent between
all outcomes as well as all individual candidates. Because Definition 5.2 requires
F(S,f)PF(T,f) for some f to obtain S dom T, it follows that there is no dominance
for an unconcerned voter. The following theorem characterizes dominance between
strategies for all concerned voters.
Theorem 5.3: Dominance. Suppose is concerned and Assumptions P and R hold.
Then S dom T for t if and only if S # T, S \ T is high for a, T \ S is low for a,
and neither S \ T nor T \ S is the set of all candidates.
Proofs of Theorem 5.3 and results in the next section are given in Brams and
Fishbum (1978). The intuition behind Theorem 5.3 is that because dominance is based
on all contingencies, and the focal voter votes for all candidates in S n T when he or
she uses S or T, S dominates T for t if and only if S \ T dominates T \ S for A. That
is, dominance shows up in the nonoverlapping candidates.
Although the conclusion of Theorem 5.3 is predicated on Assumptions P and R, the
necessary and sufficient conditions for S dom T use only t and not P or R explicitly.
This greatly simplifies the identification of dominated strategies for a voter.
For example, if X = {a,b,c} and is the trichotomous linear order abc,
Theorem 5.3 says that strategy {a}, under which the voter votes only for his or her
most preferred candidate, dominates strategies {c}, {a, c}, {b, c}, {a, b, c}, and the
abstention strategy. Moreover, these are the only strategies that a} dominates, whereas
{a, b} dominates these strategies and {b} also.
Continuing with preference order abc, we illustrate the applicability of Theorem 5.3
to plurality and approval voting. Under approval voting, the theorem says that if the
voter considers voting for b (second choice), he or she should also vote for a (first
choice) because {a, b} dom {b}. That is, {a, b} is as good as, and sometimes better
than, {b}. However, under plurality voting, a vote for b alone could be the voter's
Ch. 4: Voting Procedures 193

best strategy since in this case {b} is not dominated by any other feasible strategy.
Exhaustive enumeration for approval voting shows that there is no contingency in
which {b} induces a better outcome than {a, b} [Brams (1978, pp. 199-202), Brams
(1983, pp. 38-41)]. Fortunately, Theorem 5.3 relieves one of the necessity of checking
all contingencies for which the focal voter's vote might affect the outcome.

6. Strategic analysis of nonranked voting

Let M C {0, 1, ... , m - 1} denote the nonranked voting procedure characterized by M

in Definition 5.1. We say that strategy S is feasible for M if ISI e M, i.e., if S c B for
M. We assume that, when M is used, a ballot is counted if and only if it is a feasible
strategy. All other nonabstaining ballots are thrown out or treated like abstentions.

6.1. Admissible strategies

An admissible strategy is a feasible strategy that is not dominated by another feasible

strategy.
Definition 6.1. Strategy S is admissiblefor M and t if S is feasible for M and there
is no strategy T that is also feasible for M and has T dom S for .
As seen above, a strategy such as {b} with preference ranking abc that is feasible
for two or more nonranked voting procedures can be admissible for some procedures
and inadmissible for others. Because of this, and because our analysis of strategic
voting will be based on the assumption that nonabstaining voters use only admissible
strategies, it is useful to have a theorem that characterizes all admissible strategies
for every M and every concerned t. To facilitate the statement of the admissibility
theorem to follow, let

H(-) = X1, the subset of most preferred candidates under ,

L(>) = X,, the subset of least preferred candidates under .

The admissibility theorem may seem complex, but as later corollaries will make clear,
it is not difficult to apply to particular voting procedures. Moreover, comparisons
among procedures will show that they possess striking differences that bear on their
susceptibility to strategic manipulation.
Theorem 6.2: Admissibility. Suppose t is concerned and Assumptions P and R
hold. Then strategy S is admissible for M and if and only if S is feasible for M
and either C1 or C2 (or both) holds:
C1. Every candidate in H(t) is in S, and it is impossible to partition S into nonempty
subsets SI and S2 such that S1 is feasible for M and S2 is low for .
194 Si Brains and PC. Fishburn

C2. No candidate in L(F) is in S, and there is no nonempty A C X disjoint from S

such that A U S is feasible for M and A is high for >-.
Because the abstention strategy satisfies neither Cl nor C2, it is never admissible
for a concerned voter. A vote for all m candidates, which was omitted from the formal
Definition 5.1, would likewise be inadmissible if it were permitted.
We now consider some corollaries of Theorem 6.2 for particular voting procedures,
beginning with approval voting:
Corollary 6.3. Strategy S is admissible for approval voting and concerned t if and
only if S contains all candidates in H(t) and none in L(F).
Hence, concerned voters use an admissible strategy under approval voting if and only
if they vote for every one of their most preferred candidates and never vote for a least
preferred candidate. If m = 4 and a voter has linear preference order abcx, then his or
her admissible strategies are {a}, {a, b}, a, c} and {a, b, c}.
Corollary 6.4. A voter has a unique admissible strategy under approval voting if and
only if his or her t is dichotomous. This unique strategy is the voter's subset of most
preferred candidates.
Thus, if a voter has dichotomous preferences with XI = {a, b, c} and X2 = {x,y}, then
{a, b, c} is his or her unique dominant and admissible strategy under approval voting.
It is instructive to compare approval voting with plurality voting and negative voting
with respect to feasible and admissible strategies. We assume that abstentions are
allowed in all cases, so negative voting is equivalent to approval voting when m = 3.
When in > 3, approval voting has 2"' - 1 feasible strategies, which is the number
of subsets of X, minus X itself. By contrast, plurality voting allows m + 1 different
choices (a vote for one of the m candidates or an abstention), and negative voting
allows 2m + 1 strategies (a vote for or against a candidate or an abstention). Other
'
nonranked voting procedures allow between m + 1 and 2 - 1 different strategies.
The following corollaries of Theorem 6.2 identify the admissible strategies for
plurality and negative voting. In Corollary 6.6, ii denotes the strategy in which the voter
votes for all candidates other than candidate a or, equivalently, casts a vote against a.
Corollary 6.5. Strategy {a} is admissible for plurality voting and concerned t if and
only if a is not in L(F).
Corollary 6.6. Suppose m > 4 and t is concerned. Then:
(i) strategy {a} is admissible for negative voting if and only if the voter strictly prefers
a to at least two other candidates;
(ii) strategy a is admissible for negative voting if and only if the voter strictly prefers
at least two other candidates to a.
Corollaries 6.3, 6.5 and 6.6 can be used to identify and compare sets of admissible
strategies for various preference orders under approval, plurality, and negative voting.
Suppose, for example, that X = a, b, c, d} (where d is a candidate, not a ballot
response profile) and a - b > c - d, which we write as (ab)cd with parentheses
Ch. 4. Voting Procedures 195

Table 1
Numbers of admissible voting strategies for three procedures with four candidates

Concerned weak order Number of admissible strategies for

Approval voting Negative voting Plurality voting

Dichotomous a(bcd) 1 1 1
(abc)d 1 1 3
(ab)(cd) 1 4 2
Trichotomous (ab)cd 2 4 3
ab(cd) 2 4 2
a(bc)d 4 2 3
Multichotomous abcd 4 4 3

enclosing candidates between which the voter is indifferent. The admissible strategies
for weak order (ab)cd are:
(1) Approval voting: {a, b}, {a, b, c}. These are the only feasible strategies that contain
all the voter's most preferred, and none of his or her least preferred, candidates.
(2) Plurality voting: {a}, {b}, {c}. These are the only feasible nonabstention strategies
that do not contain the voter's least preferred candidate.
(3) Negative voting: {a}, {b}, c, d. These are the only feasible strategies in which the
voter strictly prefers the candidate to at least two others, or strictly prefers at least
two others to the barred candidate.
Table 1 shows the numbers of admissible strategies for all concerned t for four
candidates for the aforementioned three voting procedures. It is clear that the relative
numbers of admissible strategies for the three procedures are very sensitive to the
specific form of . For example, approval voting offers voters more admissible
strategies than the others when t is a(bc)d but fewer when t is (ab)cd. Hence,
although the number of feasible strategies increases exponentially in m for approval
voting but only linearly in m for plurality and negative voting, the number of admissible
strategies under approval voting is comparable to that of the other procedures and
should not overwhelm voters with a wealth of viable options.

6.2. Sincere voting and strategyproofness

We use the following notions of sincere voting and strategyproofness for nonranked
voting procedures to facilitate comparisons among procedures in terms of their ability
to elicit true preferences of voters.
Definition 6.7. Let t be a concerned preference order on X. Then strategy S is
sincere for t if S is high for A; voting procedure M is sincere for if all admissible
strategies for M and t are sincere; strategy S is strategyprooffor M and t if it is the
196 SJ Brams and PC. Fishburn

only admissible strategy for M and t (in which case it must be sincere); and voting
system M is strategyprooffor t if S is strategyproof for M and Z.
Sincere strategies are essentially ballots that directly reflect the true preferences of
a voter. If t is abcd, then a, c} is not sincere because a and c are not the voter's two
most preferred candidates. Because it is desirable that democratic voting procedures
be based on true preferences, and sincere strategies foster the expression of such
preferences, voting procedures that encourage sincerity are important. They are also
important to individual voters, for if a procedure is sincere, voters will vote for all
candidates ranked above the lowest-ranked candidates they consider acceptable. Thus,
in our example, they would not vote for c without also voting for a and b, and they
would not vote for b without also voting for a.
For the seven preference orders on four candidates in Table 1, approval voting is
sincere in six cases (only abed is excluded), negative voting is sincere in four cases,
and plurality voting is sincere in only the first three cases. These results follow easily
from Corollaries 6.3, 6.5 and 6.6.
It is no accident that approval voting is "more sincere" than the others in Table 1.
The following theorem demonstrates that approval voting is the uniquely most sincere
nonranked voting procedure among those characterized in Definition 5.1.
Theorem 6.8. If - is dichotomous, then every voting procedure M is sincere for
A. If t is trichotomous, then approval voting is sincerefor A, and this is the only
procedure that is sincerefor every trichotomous 3>.If t is multichotomous, then no
M is sincerefor >.
No procedure is sincere when > is multichotomous because, for every M and every
F with indifference classes X1 >-X2 >- · · · >-Xr, r > 4, there is an admissible strategy
that is not sincere. When there are relatively few candidates, however, it is reasonable
to expect that many voters will have dichotomous or trichotomous preference orders.
Indeed, Theorem 6.8 says that when voters do not (or cannot) make finer distinctions,
approval voting is the most sincere of all nonranked voting procedures, and this result
extends to voters with multichotomous preferences [Fishburn (1978b)].
Even if a voting procedure is sincere for A, it is not strategyproof for t if it allows
more than one admissible strategy. Like sincerity, strategyprooffess seems desirable
for voting procedures. If voters have a strategyproof strategy, they will never have an
incentive to deviate from it, even when they know the result of all other votes. Such a
strategy dominates all other feasible strategies, so whatever contingency arises, a voter
cannot be hurt, and may be helped, by choosing it.
Sincerity, on the other hand, does not imply such stability but asserts instead that
whatever admissible strategy is chosen, whenever it includes voting for some candidate,
it also includes voting for all candidates preferred to that one. In effect, a voting
procedure is sincere if it never induces voters, for strategic reasons, to "abandon" a
more preferred for a less preferred candidate.
Because the demands of strategyproofness are more stringent than those for sincere
Ch. 4: Voting Procedures 197

voting, strategyproofness is less likely to obtain than sincerity. Nevertheless, as with

sincerity, approval voting is the uniquely most strategyproof of the procedures covered
by Definition 5.1.
Theorem 6.9. If t is dichotomous, then approval voting is strategyprooffor ,
and this is the only procedure that is strategyprooffor every dichotomous I. If is
trichotomous or multichotomous, then no M is strategyprooffor >.
Theorems 6.8 and 6.9 provide strong support for approval voting based on sincerity
and strategyproofness, which can be extended to k > 2 for the election of committees
[Fishbum (1981a)]. However, the limitations of these results also are important:
strategyproofness depends entirely on dichotomous preferences; sincerity extends to
trichotomous preferences, but it is a weaker criterion of nonmanipulability than
strategyproofness. We note also that the manipulability and stability of approval voting
(as measured by how sensitive outcomes, for a given voter preference profile, are to
different ballot response profiles of admissible strategies) have provoked exchanges
between Niemi (1984) and Brams and Fishburn (1985), between Saari and Van
Newenhizen (1988a,b) and Brams, Fishburn and Merrill (1988a,b), and between Brams
and Fishburn (2001) and Saari (2001a). The first paper in each pair is critical of
approval voting, saying that approval voting is too sensitive to where voters draw
the line between acceptable and unacceptable candidates, whereas the second paper
responds to this criticism, saying that this sensitivity is desirable because it makes
approval voting more responsive to voter preferences than voting procedures that allow
the voter less leeway, either by choosing or by ranking candidates.

6.3. Efficacy
Another criterion that has been used to compare voting procedures concerns the ability
of a voting strategy to change the outcome from what it would be if the voter in
question abstained. We refer to this as the efficacy of a voting strategy and define
it as the probability that a focal voter's ballot will affect the outcome, given that all
possible ways that other voters can vote are equiprobable and ties are broken randomly
[Fishburn and Brams (1981b,c), Brams and Fishbum (1983)].
In large electorates, the most efficacious approval voting strategies are for a focal
voter to vote for either the top one or two candidates in three-candidate contests, and
to vote for approximately the top half of all candidates when m > 4. When utilities
are associated with the voter's preferences according to the expected-utility model
[Fishburn (1970)], the voter's utility-maximizing strategy in large electorates is to vote
for all candidates whose utilities exceed the average utility over all candidates. Hoffman
(1982, 1983) and Merrill (1979, 1981, 1982, 1988) have independently derived similar
results; in doing so, they consider criteria other than expected-utility maximization.
A voter's utility-maximizing strategy can lead to substantially different expected-utility
gains, depending on his or her utilities for the candidates. However, it can be shown
that plurality voting gains are even more disparate [Fishburn and Brams (1981b,c,
198 S.J. Brains and PC. Fishburn

1983), Rapoport and Felsenthal (1990)], so approval voting is more equitable in the
sense of minimizing differences among voters.
As a case in point, plurality voting affords a dichotomous voter who equally likes
four candidates but despises a fifth in a five-candidate race little opportunity to express
his or her preferences, compared with a voter who greatly prefers one candidate to all
others. Approval voting, on the other hand, is equitable to both - allowing the first voter
to vote for his or her top four choices, the second to vote for his or her top choice -
despite the extreme differences in their utilities. In general, not only is a voter able to
be more efficacious under approval than plurality voting, but he or she cannot suffer
as severe utility losses under the former procedure.

6.4. Plurality ballots and the median outcome

We conclude our discussion of single-stage nonranked voting with an example that
does not fit Definition 5.1 because it has a continuum of candidates and uses a
different selection criterion. The example features single-peaked preferences, which
are discussed at greater length in Chapter 13 in Volume 2 of this Handbook..
Example 6.10. A nine-member committee is to decide how much of next year's budget
to devote to some activity. We assume that each member has an ideal amount, with
preference decreasing as one moves away from the ideal in either direction. Suppose
the committee decides as follows: each member writes down one amount on a slip of
paper, and the median of the ballot amounts becomes the collective choice. It is easily
seen that each member has a unique dominant strategy, namely to vote for his or her
ideal. The unique dominant strategies are sincere and the system itself is strategyproof
in this restricted context. A complete characterization of unique dominant strategies
for voting procedures in a general context is given in Dasgupta, Hammond and Maskin
(1979). U

7. Nonranked multistage voting: successive elimination

A multistage nonranked voting procedure is a procedure which, in a succession of

nonranked ballots, eliminates candidates at each stage, or after each ballot, until a
winner is determined. The number of ballots can be fixed or variable, depending on
the procedure's rules. While there are many such procedures in use today, most are
similar to one of the following three types.

7.1. Examples
Plurality with a runoff starts with a plurality-voting ballot; then it determines a
winner by a simple majority vote between the top two candidates from the first ballot.
The second ballot is often avoided if the top plurality candidate gets a sufficiently
Ch. 4: Voting Procedures 199

large percentage of the vote on the first ballot, say 40% or 50%, in which case that
candidate wins outright. Other nonranked procedures could be used on the first ballot,
but plurality voting is by far the most common. Requiring a maximum of two ballots,
plurality with a runoff is effective in reducing a large field of candidates quickly and
hence is popular in public elections, in which several ballots are impractical.
Plurality with successive elimination uses plurality voting on a succession of
ballots until one candidate, the winner, gets more than 50% of the vote. After each
ballot that requires a successor, some candidates may withdraw voluntarily from the
race or be eliminated by a low-vote rule such as "lowest-person out". But successive
votes are sometimes allowed without any reduction of the still-in-contention set, and
some procedures even allow new candidates to enter during the process. Plurality with
successive elimination often leads to fierce politicking during the balloting, with arm
twisting, backroom deals, and the like. Examples of the procedure are analyzed by
Brams and Fishburn (1981, 1983) and Fishburn, Fishburn and Hagy (1992). The latter
paper describes instances in which dozens of time-consuming votes were taken to elect
a candidate, leading some to refer to the procedure as "election by exhaustion"; such
contests were common in U.S. national party conventions in the 19th and early 20th
centuries [Brams (1978, Chapter 2)].
Successive majority voting uses a series of simple majority votes between
subsets of X. The first vote is between subsets A and B for which A U B = X and
both A \ B and B \ A are nonempty. If A wins the first vote, the candidates in B \ A
are eliminated and, if A > 2, a second vote is taken between Al and A2 for which
Al U A2 = A and both A \ A2 and A2 \ A are nonempty. If B wins the first vote, the
candidates in A \ B are eliminated; if B] > 2, a second vote is taken between B 1 and
B2 for which B 1 U B2 = B and both B 1 \ B 2 and B2 \ B 1 are nonempty. At each vote, the
remaining candidates not in the winning subset are eliminated. The process continues
until a single candidate, the final winner, remains. Example 1.1 gives an example.
The first vote, between al and a2, is viewed as a vote between A = {al,a 3 , .. ., am}
and B = {a 2, a3, ... , am} since the winning subset remains in contention after this
vote. Succeeding votes can be interpreted similarly as votes between two subsets of
the candidates not yet eliminated.

7.2. Binary multistage voting

Farquharson (1969) discusses a generalization of successive majority voting that we
refer to as binary multistage voting. The votes proceed between subsets as described
in the preceding paragraph with the following modifications: abstentions are not
permitted, each vote is decisive, and, when m > 4, the decision rule in each
stage need not be simple majority. However, the decision rules must be monotonic,
nondictatorial, and responsive to every voter's vote: see Farquharson (1969, p. 14) for
precise definitions. To be satisfied, his conditions require at least three voters.
The two subsets and the decision rule used for each potential binary vote are
specified in advance. A strategy, or ballot in our previous terms, says which of the
200 S.J. Brains and PC. Fishburn

two subsets the voter votes for in each case that might arise. Our previous definitions
of sincere, admissible, and strategyproof strategies are patterned after Farquharson's
definitions. He assumes that voters have linear preference orders on X. A strategy is
sincere if, in any vote between subsets A and B, the voter votes for A if the top candidate
in A \ B is preferred to the top candidate in B \ A, and votes for B when the reverse
obtains. A strategy is admissible if it is not dominated by another strategy, where
dominance is based on contingencies, as in Section 5. And a strategy is straightforward
(strategyproof by our earlier definition) if it is the unique admissible strategy.
One further definition leads to Farquharson's main theorem for straightforwardness.
When A and B are nonempty subsets of X, A,B} separates a voter's preference
order t if either the least preferred candidate in A bears > to the most preferred
candidate in B, or the least preferred candidate in B bears > to the most preferred
candidate in A.
Theorem 7.1. A binary multistage voting procedure is straightforwardfor a voter
if and only if {A, B} separates his or her linear preference orderfor every potential
binary vote between subsets A and B that might arise during the voting.
For example, if m = 3 and the first vote is between a, b} and {c}, then abc, bac,
cab and cba have straightforward strategies (the separation in each case is shown by
the slash: ab/c, ba/c, c/ba, and c/ab), but acb and bca do not because c divides a from
b (no slash can separate these orders into a, b} and {c}). If the first vote is between
{a, b} and {b, c}, then only abc and cba have straightforward strategies (in these cases,
there is a common element on both sides of the slashes: ab/bc and cb/ba). Farquharson
notes that no binary multistage voting procedure can be straightforward for all linear
orders.
He also introduced the term sophisticated voting to characterize voting strategies
arrived at by recursive analysis when every voter knows the others' preferences and it is
assumed that every voter uses an admissible strategy. A strategy is primarilyadmissible
strategy if there is no other strategy which produces at least as good an outcome in
every contingency where other voters use admissible strategies, and produces a better
outcome in some such contingency. A strategy is secondarily admissible strategy if
it is primarily admissible when all other voters use primarily admissible strategies.
Continuation leads to ultimately admissible strategies, which are called sophisticated
strategies.
Theorem 7.2. When all voters have linear preference orders and know each others
preferences, every voter has a unique sophisticatedstrategyfor every binary multistage
voting procedure.
The voting strategies described in Example 1.1 are sophisticated. As seen there,
sophisticated strategies need not be sincere. Indeed, insincere sophisticated strategies
are prime examples of a procedure's susceptibility to strategic manipulation.
Farquharson (1969, p. 43) notes that the conclusion of Theorem 7.2 does not extend
to multistage nomranked voting when votes are taken for three or more competing
Ch. 4: Voting Procedures 201

subsets in some stages. This does not, of course, mean that such systems are free from
strategic manipulation but only that voters need not have uniquely best strategies under
the type of calculation that produces sophisticated strategies. For example, plurality
with a runoff is rife with strategic possibilities: if one's favorite candidate cannot win,
there may be several ways to defeat one's worst candidate, either by helping to prevent
him or her from making the runoff, or by helping someone else win against him or
her in the runoff.

7.3. Paradoxes

Although multistage nonranked voting procedures are very popular and can serve a
group's practical needs, they are subject not only to strategic manipulation but also to
a variety of anomalies, or paradoxes, that are often not recognized by their proponents.
The paradoxes can arise under naive or sincere voting in the absence of strategic
calculations; indeed, they can subvert a procedure's purpose of electing a candidate,
in a democratic manner, that best serves the interests of a group.
An array of paradoxes for multicandidate voting procedures is described and
analyzed by Riker (1958), Fishburn (1974a, 1981b, 1982), Niemi and Riker (1976),
Doron and Kronick (1977), Doron (1979), Gehrlein (1983), Fishbum and Brams
(1983), Saari (1984, 1987, 1989, 1994), Moulin (1988a), Brams, Kilgour and Zwicker
(1998), Scarsini (1998) and Nurmi (1998a,b, 1999), among others. We illustrate five
of these for multistage nonranked procedures under sincere voting.
The dominated candidateparadox [Fishburn (1974a)] occurs when all voters prefer
another specific candidate to the winner. Suppose the 13-voter voter preference profile
of Example 4.1 holds and successive majority voting is used with voting order acbx.
The winners under sincere voting after the three votes are c, b, and x, respectively, so
x wins the election. However, all voters prefer a to x. It follows that, when m > 4,
successive majority voting does not satisfy the Pareto dominance condition (2.1).
Theorem 7.2 applies to this case and, if all voters use their sophisticated strategies, c
would be elected. Including sincere voting on the third vote, the sophisticated strategy
of the 4 voters with ranking axbc is to vote for a whenever possible, and to vote for b
if b faces c. The sophisticated strategies of the 9 voters with rankings caxb and bcax
are to vote for c whenever possible, and to vote for a if a faces b. Only the latter
voters, with ranking bcax, have insincere sophisticated strategies.
The winner-turns-loserparadox [Doron and Kronick (1977)] illustrates the failure
of monotonicity that occurs under sincere voting when the winner would have been a
loser if some voters had ranked this candidate higher in their preference orders, all else
unchanged. An example for plurality with a runoff occurs with the 93-voter preference
profile in which

27 voters have abc; 42 voters have cab; 24 voters have bca.

202 S.2. Brains and PC. Fishburn

The first-vote plurality winners are a and c, with c beating a 66-to-27 in the runoff.
If 4 of the first 27 voters had raised c from third place to first, the profile would have
been
23 voters have abc
46 voters have cab
24 voters have bca.
Now the first-vote winners are b and c, with b beating c 47-to-46 in the runoff. Thus,
c changes from a winner to a loser when it gains support.
Another anomaly, which is closely related to the winner-turns-loser paradox, is
the no-show paradox [Fishburn and Brams (1983)]. It occurs when the addition of
identical preference orders with candidate x ranked last changes the winner from
another candidate to x. This occurs in the preceding example when we begin with
the second profile, where b wins under plurality with a runoff. If we then add from
two to 42 voters to the group with ranking abc who have c in last place, c wins.
Additional analyses of the paradox are in Ray (1986), Moulin (1988b) and Holzman
(1988-1989). The key to both this paradox and its predecessor is who gets scratched
after the first vote. It does not depend on the Condorcet paradox, wherein majorities
cycle, which happens to occur in these examples. The main three-candidate example
in Fishburn and Brains (1983) has a majority or Condorcet candidate (see Section 8),
but the no-show paradox still occurs when that candidate is scratched after the first
vote for one of the voter preference profiles.
Our fourth paradox, the multiple-districtsparadox, occurs when one candidate would
win in each of a number of districts separately but loses the combined-districts election.
A two-districts example in Fishburn and Brams (1983) shows for a three-candidate
election conducted by plurality with a runoff, one candidate would win in each district
but lose the overall combined-districts election. Moreover, each of the other candidates
has a sizable majority over the candidate who would win in each separate district. We
return to this phenomenon in Section 9, where its proscription is noted to be a central
axiom of positional voting procedures like Borda's.
Our final paradox in this section (others will be discussed in Section 9.3) is
the multiple-election paradox [Brams, Kilgour and Zwicker (1998), Scarsini (1998),
Nurmi (1998b, 1999)]. Consider a referendum in which voters can vote either
yes (Y) or no (N) on each proposition on the ballot. The paradox occurs when
the set of propositions that wins, when votes are aggregated separately for each
proposition (proposition aggregation), receives the fewest votes when votes are
aggregated by combination (combination aggregation). As an example, suppose there
are 3 propositions, so there are 23 = 8 combinations because each voter can make
one of two choices (i.e., Y or N) on each proposition. Suppose further that there are
13 voters who cast the following numbers of votes for each of the eight combinations:
YYY:I YYN:1 YNY: NYY: YNN:3 NYN:3 NNY:3 NNN:0.
For example, YYN means a Y vote on the first and second propositions and an N vote
on the third.
Ch. 4: Voting Procedures 203

Notice that on each of the three propositions, N beats Y by 7 to 6 votes, so NNN

wins according to proposition aggregation. However, NNN is the only combination that
receives 0 votes, illustrating the dramatic difference that can arise between proposition
aggregation and combination aggregation (YNN, NYN, and NNY tie for first place
with 3 votes each). It turns out that the existence of this paradox implies that majorities
cycle, but not vice versa, and actually occurred in the 1990 California general election,
as did some variants of the paradox [Brams, Kilgour and Zwicker (1997)].
This paradox vividly illustrates the conflict that can occur between the two vote-
aggregation procedures. Like the other paradoxes, it does not depend on either sincere
or strategic voting: voters may be perfectly sincere in voting for their preferred position
on every proposition, or they may be strategic (in some sense). The paradox says only
that majority choices according to proposition aggregation may receive the fewest votes
according to combination aggregation.

8. Condorcet choices and ranked voting

We continue to assume that IXI > 3 and k = 1 so that F(d) = F(X, d) is a nonempty
subset ofX for every ballot response profile d e D. In addition, V with typical member
U = (i, . .. , ,,) is a set of voter preference profiles. The Condorcet set, or majority
set, for v C V is

Con(v)= {a C X: {i: a i b}l > {i: b ti a}l for all b X}

= {a EX: I{i: a >-i b}l > [{i: b >-i a}l for all b X},

and a candidate in Con(v) is a Condorcet candidate or majority candidate. Con(v) is

empty if and only if every candidate can be beaten by another candidate in a sincere
simple majority vote between the two, presuming a voter abstains if and only if he or
she is indifferent between the two. When n is odd and every >-i in v is a linear order,
ICon(v)l C {0, 1}. If all voters are indifferent among all candidates, Con(v) = X. We
also denote by >M the strict simple majority relation on X induced by a profile v, so
that a >M b if {i: a -i b} > {i: b >-i all, with Con(v) = {a C X: b >M a for no
b GX \ {a}}.
Our discussion of Con(v) is divided into three parts. The first considers combinato-
rial aspects of >M and Con(v). The second relates Con(v) to voting procedures defined
in preceding sections, and the third examines Condorcet social choice functions, which
are designed to elect a candidate which has a strict simple majority over every other
candidate when such a candidate exists.

8.1. Condorcet combinatorics

Under this heading we describe studies devoted to the structure of >M and Con(v).
As before, n is the number of voters and m is the number of candidates. In addition,
204 SJ.Brains and PC. Fishburn

VL will denote the set of all voter preference profiles for (m, n) in which every i is
a linear order or strict ranking.
Some time ago, McGarvey (1953) showed that if n is large enough compared to
m, then for every asymmetric binary relation on X there is a v G VL that has this
relation as its >M. The question then arose as to the smallest n, denoted by o(m),
for which this is true when IX = m. Stearns (1959) proved that (m) < m + 1 for
odd m, (m) < m + 2 for even m, and equality holds here when m E {3, 4, 5}. He
showed also that o(m) > [(log 3)/2]m/(log m), where log denotes the natural logarithm.
Erd6s and Moser (1964) then noted that o(m) < clm/(log m) for a fixed constant cl.
Precise values of o(m) are unknown beyond the first few m, and the question of whether
o(m)(log m)/m tends to a limit as m oc remains open.
Riker (1958, 1982) and Gehrlein (1983), among others, describe multicandidate
cases in practice that probably had no Condorcet candidate. One technical approach
to the likelihood of Condorcet's paradox focuses on the proportion p(m, n) of the
(m!)f profiles in VL that have a Condorcet candidate that bears >M to every other
candidate. If each voter independently chooses one of the m! rankings at random
(i.e., according to the uniform distribution), then p(m, n) is the probability that one
candidate has a strict majority over every other candidate. Early studies of p(m, n)
include Guilbaud (1952), Niemi and Weisberg (1968), and DeMeyer and Plott (1970),
with later refinements by Gehrlein and Fishburn (1976, 1979). It is easily seen that
p(3 , 3) = 17/18, but exact computations for m > 3 or n > 3 get complex very quickly.
The most efficient method for three candidates [Gehrlein and Fishburn (1976)] uses
n!2-( n2+ n3)
p(3, n) = 3+ E n !n2!n3!4!'

where E is a triple sum with limits {0 < nI < (n - 1)/2, 0 < n2 < (n - 1)/2 - n,
0 < n3 < (n - 1)/2 - n and n4 = n - n I - n2 - n3. The most efficient method known
for three voters [Gehrlein and Fishburn (1979)] uses
- Ii -mI l (m - 1 - ml)!(m - 1 - m2)!
' 0 m!(m - 1 -ml-m2)!(ml +m2 + 1)
rnll=0O 2=O

When m > 4 is even and n is odd, there is a nice recursion relation for p(m, n). The
simplest case [May (1971)] is
p(4, n) = 2p(3, n) - 1.
The recursion for m = 6 and n odd is
p(6, n) = 3p(5, n) - 5p( 3 , n) + 3,
and in general [Gehrlein and Fishburn (1976)]
m2/2

p(m, n) = ,cjp(2j - 1, n),

Ch. 4: Voting Procedures 205

where the cim coefficients are independent of n. No similar relationship holds for odd m.
With limiting proportion

p(m) = lim p(m, n),

n o

Guilbaud (1952) showed that

p(3) = + 23 sin-' () - 0.91226,

and Niemi and Weisberg (1968) proved that p(m) equals m times the (m - 1)-
dimensional normal positive orthant probability with all correlations equal to . An
approximation accurate within one-half of one percent for odd m < 50 is

p(m) - 9 + (0.63)(_-3)/2
m + 9.53
A more complex but more accurate approximation appears in Gehrlein (1999). Many
additional results along this line appear in Kelly (1974), Fishburn, Gehrlein and Maskin
(1979) and the extensive reviews of Gehrlein (1983, 1997).
We now turn to restrictions on voter preferences which imply that >M is acyclic, or
that Con(v) is nonempty. Based on the approach taken by Ward (1965) and Sen and
Pattanaik (1969), let T denote a subset of the m! linear orders on X = {1,2, .. ., m},
and define T to be acyclic if there do not exist a, b, c G X and three orders in T whose
restrictions on {a, b, c} are abc, cab, and bca. Interest in T stems from the following
basic proposition, where V(T) denotes the set of all nonempty finite lists (any number
of voters) of linear orders in T and vA for A C X is the restriction to A of v G V(T).
Theorem 8.1. Con(vA) is nonempty for every v V(T) and every A C X with
3 < IAI < m if and only if T is acyclic.
Several people, including Kim and Roush (1980), Abello and Johnson (1984),
Abello (1991), Craven (1996) and Fishburn (1997), have considered how large T can
be while providing the guarantee of Condorcet candidates given by Theorem 8.1. We
let

f(m) = max{lTI: T is acyclic forX = {1, ... , m}},

and remark that T is acyclic if and only if, for all a < b < c in X, the restrictions
of T's orders to {a, b, c} must exclude at least one order in each of the cyclic triples
{abc, cab, bca} and {acb, bac, cba}. Thus f(3) = 4. In addition, f(4) = 9 [Abello
(1981), Raynaud (1982)] with acyclic

T = {1234, 1324, 1342,3124,3142,3412,3421,4312,4321},

f(5) = 20 [Fishburn (1997)] and, based on an example of Bernard Monjardet and a

construction procedure in Craven (1996) and Fishburn (1997),f(6) > 45,f(7) > 100,
206 S.2. Brains and PC. Fishburn

and f(8) > 222. It is known also that the optimal pattern for max T undergoes a
paradigm shift near m = 10, thatf(m) > (2.1708)" for all large m, and, as proved by
Raz (2000), thatf (m) < c for some constant c2 and all m.
Nonmaximum but natural restrictions on voters' preferences that guarantee Con-
dorcet candidates include single-peaked preferences [Galton (1907), Arrow (1951),
Black (1958), Fishburn (1973)]. A typical case occurs when the candidates can be
ordered along a line and each voter's preferences, left-to-right, increase up to a most
preferred candidate and then decrease. With no loss of generality we use the natural
order 12 ... m for X = {1,2 ... , m}. The following definition generalizes weak order
by allowing each ji to be a partial order, which means that its asymmetric part -i
is irreflexive and transitive.
Definition 8.2. A voter preference profile = (il,..., fi) of partial orders on
X = {1,2, ... , m} is single peaked in the order 12 . m if, for each i {1, ... , n},
there are unique ai, bi e X with a < bi such that, for all x,y,z G X:
(i) x < y ai y-ix x
(ii) bi < y < x y Fi x
(iii) ai < y < bi ~ y i x
(iv) (x < y < z,x i Y,Y -i z) X x i z.
The candidates in [ai, bi] are voter i's preference plateau. The following theorem
[Fishburn (1973, p. 108)] locates Con(v) as a nonempty interval of integers:
Theorem 8.3. Suppose v is a single-peaked voter preference profile ofpartial orders
as specified in Definition 8.2. Let cl, c2, ... , c,,2 be a rearrangementof the sequence
a, ... , a,,bl,..., b with cl < c2 <... < C2. Then

Con(v) = {x E X: c < x Cn+ }i

Strategyproofness for single-peaked preferences is discussed by Moulin (1980),

Berga (1998), and Ching and Serizawa (1998). Other restrictions on voter preferences
are discussed in Chapter 3 in this Volume, and Chapter 21 in Volume 2 of this
Handbook.

8.2. Nonranked voting and Condorcet candidates

This section considers the propensities of nonranked voting procedures to elect
Condorcet candidates when Con(v) is nonempty. As in Section 6, we let M denote the
single-stage nonranked voting procedure characterized by M in Definition 5.1. We also
let M + denote the two-stage runoff procedure in which the first vote is a procedure-M
vote and the two candidates with the most votes on the first vote go against each other
in the runoff, whose outcome is determined by simple majority voting. Thus, 0,1 }+
is plurality with a runoff, and {0, 1, ... , m - 1}+ is approval voting with a runoff.
Ch. 4: Voting Procedures 207

Inada (1964) proved that Con(v) is nonempty if all preference weak orders in profile
v are dichotomous. Using previous results, which presume Assumptions P and R
in Section 5.2, we can prove more than this, namely that the use of admissible
strategies under approval voting and dichotomous preferences always yields Con(v) as
the outcome. Moreover, for every other M procedure, the use of admissible strategies
may give an outcome that contains no Condorcet candidate. The following definition, in
which preference orders are mapped into admissible strategies, will be used to express
these results more precisely.
Definition 8.4. For any nonempty list v = ( ..., .,) of preference orders, and every
nonranked voting procedure M, let v(M) be the set of all d = (dl, ... , dn) in which
di is an admissible strategy for M and i. For every d C v(M), let F(d) = F(X, d),
the outcome for ballot response profile d under procedure M.
To illustrate, suppose v = (abc, abc, c(ab)): two voters prefer a to b to c; the other
is indifferent between a and b and prefers c to both. Assume that M is the plurality
procedure. By Corollary 6.5, each of the first two voters has two admissible strategies,
fa} and {b}, and the third voter has one. The outcomes of the 4 = 2 x 2 x 1 members
of v(M) are {a}, {a, b, c}, {a, b, c} and {b}.
Theorem 8.5. If all preference orders in v are dichotomous and M is the approval
voting procedure, then F(d) = Con(v) for all d E v(M).
In other words, if all voters have dichotomous preferences, their use of admissible
strategies under approval voting invariably yields Con(v) as the outcome.
To see how plurality voting differs with dichotomous preferences, suppose X =
{a, b, c} with 2n + 1 terms in v: one is a(bc), n are b(ac), and n are (ac)b. Then
Con(v) = {a}. However, if as few as two of the n with (ac)b vote for c (voting for
either a or for c is an admissible strategy under plurality voting), then the outcome is
{b}, which is disjoint from Con(v). The following theorem shows that a similar result
holds for every procedure other than approval voting:
Theorem 8.6. Suppose M is a nonranked voting procedure other than approval voting.
Then there is a v composed entirely of dichotomous preference orders and a d E v(M)
such that F(d) and Con(v) are disjoint.
In contrast to the definitive picture for dichotomous preferences, comparisons among
approval voting and other M procedures are less clear-cut when some voters partition
X into three or more indifference classes. Reviews by Merrill (1988) and Nurmi
(1987), based primarily on computer simulations [see also Bordley (1983), Chamberlin
and Featherston (1986), Fishburn and Gehrlein (1976, 1977, 1982), Nurmi (1988),
Regenwetter and Grofman (1998)], suggest that approval voting is generally as good
as or better than other M procedures, particularly plurality voting [Nurmi and Uusi-
Heikkila (1985), Felsenthal and Maoz (1988)], in electing Condorcet candidates.
Indeed, it compares favorably with most positional scoring procedures (Section 9) not
208 S." Brams and PC. Fishburn

only in terms of its Condorcet efficiency [Merrill (1985)] but also in terms of the
"social utility" of elected candidates [Weber (1977, 1995), Merrill (1984, 1988)].
We now integrate runoff procedures of type M+ into the picture. Generally speaking,
these procedures are less sincere than M procedures, and plurality with a runoff
is less sincere than approval voting with a runoff because the former is more
restrictive when preferences are dichotomous. Neither runoff approval nor plurality
is strategyproof, even when preferences are dichotomous. Runoff approval voting,
especially, is susceptible to severe manipulation effects and is even more manipulable
when preferences are not dichotomous. For example, if a sizable minority of voters
has preference order abc and is fairly sure that a would beat c but lose to b in a runoff,
these voters may well vote {a, c} on the first vote in an attempt to engineer a runoff
between a and c. Other examples, and more precise statements of results, are given in
Fishburn and Brams (1981a) and Brams and Fishburn (1983).
Our next theorem considers Condorcet candidates under sincere voting. A sincere
strategy for M + is a strategy whose M vote is sincere according to Definition 6.7 and
whose vote in the runoff is sincere.
Theorem 8.7. For both M and M + procedures, there exist sincere strategies that
will elect a Condorcet candidate under approval voting but not necessarily under any
other procedure.
This is not to say, however, that all sincere strategies guarantee the election of a
Condorcet candidate when one exists and approval voting is used. But it is possible to
make this guarantee in some cases.
Example 8.8. Suppose m = n = 3 and v = (xab, (ax)b, (bx)a). Then Con(v) = {x}.
Under approval voting, the only admissible strategies for voters 2 and 3 are {a,x}
and {b,x}, respectively. The first voter has two sincere admissible strategies, {x} and
{x, a}, and x wins for both. Hence, approval voting must elect x when voters use sincere
admissible strategies.
Consider plurality, runoff plurality, and runoff approval voting. The following
strategies are admissible and sincere for all three procedures on the first or only vote:
1 votes for x, 2 for a, and 3 for b. These do not guarantee the election of x under
plurality, nor under a runoff procedure where the runoff pair could be {a, b}. U
A second example demonstrates that when all voters use admissible but not
necessarily sincere strategies, a Condorcet candidate's election may again be ensured
only under approval voting.
Example 8.9. Suppose m = 3, n = 4, and

v = ((xa)b, (xa)b, xba, bxa),

so Con(v) = {x}. All admissible strategies for approval voting are sincere
(Theorem 6.8) and will elect x. But if the last two voters vote for b under plurality
voting, x's election would not be ensured. Moreover, if in a runoff the first two voters
Ch. 4: Voting Procedures 209

vote for a on the first vote and the other two vote for b, neither runoff plurality nor
runoff approval voting would elect x. Thus, not only may approval voting be the only
procedure to guarantee the election of a Condorcet candidate when voters are restricted
to sincere admissible strategies, but it also may be the only procedure to provide this
guarantee when voters are less constrained and can use any admissible strategy. ·
We generalize these examples in the following theorem for plurality and approval
voting, with or without a runoff. When x beats all other candidates by simple majority
comparisons, x is a strict Condorcet candidate.
Theorem 8.10. Suppose all voters use admissible strategies on the first vote (if there
is a runoff) or only vote (if there is not) and, if there is a runoff vote on the runoff
if and only if they are not indifferent between its two candidates. Then, for all m > 3
and all voter preference profiles for m candidates,

[x must be elected under runoff plurality voting]

= [x must be elected under runoff approval voting]
= [x must be elected under plurality voting]
I[x must be elected under approval voting]
X [x must be a strict Condorcet candidate].

Because examples can be constructed to show that the converse implications are
false, the ability of a procedure to guarantee the election of a strict Condorcet candidate
is highest for approval voting, next highest for plurality voting, third highest for
runoff approval voting, and lowest (essentially nonexistent) for runoff plurality voting.
Moreover, approval voting also encourages the use of sincere strategies. Because
procedures with more complex choice criteria do not [Merrill and Nagel (1987),
Merrill (1988)], voters need not resort to insincere strategies to elect Condorcet
candidates, if they exist, under the approval voting procedure.
The primary mathematical analyses of the likelihood that a rule of type M or M + will
choose a Condorcet candidate, given that one exists, appear in Gehrlein and Fishburn
(1978a) and Gehrlein (1981, 1982, 1993, 1995), with a review in Gehrlein (1997,
pp. 190-194). Under random choices of voter rankings for VL and the presumption
of sincere voting, the limiting conditional probabilities in n that plurality and runoff
plurality will elect a Condorcet candidate for m = 3 are 0.7572 and 0.9629, respectively
[Gehrlein and Fishburn (1978a), Gehrlein (1993)]. Given m = 3 with n = 3j,
j E {1,3, 5, ... }, Gehrlein (1982) obtained closed-form expressions for the conditional
probability of electing the Condorcet candidate (given that one exists) under a different
probabilistic preference order assignment called "impartial anonymous culture". The
expression for plurality voting is

119n4 + 1348n 3 + 5486n2 + 10812n + 10395 119 .

= 0.8815,
135(n + 1)(n + 3)2(n + 5) 135
210 S.J2 Brains and PC Fishburn

and for runoff plurality is

523n4 + 6191n3 + 25117n2 + 40749n +22140 523 .
- = 0.9685.
540(n + l)(n + 3)2(n + 5) 540
Gehrlein (1995, 1997) includes tables which show how such Condorcet likelihoods
change as social homogeneity, measured by a parameter in a contagion model of
choices of preferences, varies. As one would expect, for most voting procedures social
homogeneity increases the likelihood that Condorcet candidates will be elected.

8.3. Condorcet voting procedures

A number of voting procedures have been proposed whose aim, in part, is to elect
a strict Condorcet candidate x (x >M y for all y x in X) when one exists
[Condorcet (1785), Fishbum (1977), Richelson (1979), Straffin (1980), Riker (1982),
Dummett (1984), Schwartz (1986), Tideman (1987), Nurmi (1987, 1998a), Merrill
(1988), Levin and Nalebuff (1995), Le Breton and Truchon (1997)]. Apart from the
highly manipulable successive majority procedure of Section 7.1, none is widely used
in practice. We therefore merely outline various procedures and leave most details of
their axiomatic and strategic analysis to the references.
We simplify matters considerably by assuming that voters have linear preference
orders and vote sincerely. In some cases, the ballot response profile d is taken to be
equal to the voter preference profile v, whereas other cases only require d to reveal
certain aspects of v, such as >M on X. We say that a procedure is a Condorcet voting
procedure if F(X, d) = {x} whenever x is a strict Condorcet candidate. We define an
even dozen such procedures, which are partitioned into three groups according to the
information needed to determine the social choice set F(X, d), which we denote simply
by F. The names of most procedures are explained more fully in Fishburn (1977).
A Condorcet voting procedure is a C1 procedure if >M is sufficient to determine
F. We note six C1 procedures
(1) Copeland' procedure [Copeland (1951), Goodman (1954), Henriet (1985)] has
x F if x maximizes I{y: x >M y} - I{y: y >M x}l. It always chooses a strict
Condorcet candidate when one exists, but its F can be disjoint from a nonempty
Con(v). The four-voter profile v = (xyabc,xybac, cbaxy, yacbx) has Con(v) = {x}
but F ={y}.
(2) Miller's procedure [Miller (1980, 1995), Shepsle and Weingast (1984), Epstein
(1998)] takes x E F if x is in the uncovered subset of (X, >,M), i.e., if whenever
y >M x there is a z E X such that z >M y and x >M z, where x >M z means that
not (z >M x) or that x beats or ties z.
(3) Fishburn procedure [Fishbum (1977)] defines >M by a >' b ifz >M a =} z >M b
for all z E X, and for some z, a >M z >M b. Then x E F if no y has y > x. The
relation >M is a strictpartial order (asymmetric and transitive), so (X, >M) always
has maximal candidates when X is finite. Fishburn's choice set is always included
in Miller's choice set, and the inclusion can be proper. Another Condorcet voting
procedure with this property is described by Dutta (1988).
Ch. 4: Voting Procedures 211

(4) Schwartzs procedure [Good (1971), Schwartz (1972, 1974)] defines > as the
asymmetric part of the transitive closure of >M, and takes x E F if no y has
y >M x. Like >, > is a strict partial order.
The next two definitions assume that x >M y ory >M x for all distinct x, y E X, i.e.,
that (X, >M) is a tournament. This is always true under preceding assumptions when
n is odd.
(5) Banks s procedure [Banks (1985), Miller, Grofman and Feld (1990)] takes x e F
if x is the maximum candidate of a maximal transitive subtournament of (X, >M).
In other words, if (Y, >M) is a >M-linearly ordered subset of (X, >M) and no other
linearly ordered subset (Z, >M) has Y c Z, then the candidate in Y that beats all
others in Y is in F.
(6) Slater procedure [Slater (1961), Laffond and Laslier (1991), Laslier (1997)] takes
x E F if x is the maximum candidate in a linear ordering of X which requires the
minimum number of reversals of pairs in >M to achieve linearity. Charon, Hudry
and Woirgard (1997) define a 16-vertex tournament for which the choice sets of
the Banks and Slater procedures are disjoint.
For v E VL and distinct x,y E X, let v(x,y) denote the number of voters who prefer
x to y, so v(x,y) + v(y,x) = n when there are n voters. We refer to a Condorcet voting
procedure as a C2 procedure if it is not a Cl procedure and the v(x,y) counts suffice
to determine F. We consider four C2 procedures:
(7) Blacks procedure [Black (1958)] takes F = Con(v) if Con(v) is nonempty;
otherwise, F is the set of Borda winners, so that x E F if Yz v(x, z) > jz v(y, z)
for ally eX.
(8) Nansons procedure [Nanson (1907), Hoag and Hallett (1926), Black (1958),
Nurmi (1989), McLean (1996)] is a Borda-elimination procedure that is a
Condorcet voting procedure. Let v(x,A) = >yCA (x,y). Let Al = X and for
each j > 1 let Aj+ = A \ x Aj: v(x,Aj) < v(y,Aj) for all y E Aj, and
v(x, A) < v(y,Aj) for some y C A}). The Aj decrease to a nonempty limit set A*,
which is F for Nanson's procedure.
(9) Condorcets procedure [Condorcet (1785), Black (1958)] is a maximin procedure.
Let v.(x) = min{v(x,y): y E X \ {x}}. Then x c F if x maximizes v(x) over X.
Young (1988) argues that the next procedure is more in keeping with Condorcet's
intentions when there is no Condorcet candidate.
(10) Kemeny procedure [Kemeny (1959)] takes x C F if x is the maximum candidate
in a linear ordering L of X that maximizes ,{v(a,b)L(a, b): a, b E X}, where
L(a, b) = 1 if aLb and L(a, b) = 0 otherwise. This procedure is axiomatized by
Young and Levenglick (1978); also see Young (1988, p. 1242).
Finally, we define a Condorcet voting procedure as a C3 procedure if it is neither a
C1 nor C2 procedure. We note two C3 procedures:
(11) Dodgson s procedure [Dodgson (1876), Black (1958), Fishburn (1973)], named
after the Rev. Charles Lutwidge Dodgson, a.k.a. Lewis Carroll, is based on the
minimum number of reversals in the linear orders in v by which a candidate beats
212 S.J Brants and PC. Fishburn

or ties every other candidate under simple majority comparisons. Denote this
minimum reversals number for x by r(x, v), and let r*(x, v) = limNo r(x, Nv)/N,
where Nv is v replicated N times (so Nv has Nn voters). Dodgson's procedure
takes F as the set of candidates that minimize r*(x, v).
(12) Youngs procedure [Young (1975b)] is based on the largest number of terms
(voters) in a sublist of v for which a candidate beats or ties every other candidate
under simple majority applied to the sublist. Denote this largest number for x
by l(x, v) and, if no sublist has the noted property for x, let l(x, v) = 0. Then let
l* (x, v) = limN -, l(x, Nv)/N. Young's procedure takes F as the set of candidates
that maximize l*(x, v).
Along with the strict Condorcet property, all 12 procedures are anonymous, neutral,
and homogeneous in the sense that F remains invariant to replications No of v. All were
intended to yield a "good" choice set in the absence of a strict Condorcet candidate,
but some seem better than others in this regard. Apart from the Miller, Banks, and
Slater procedures, a comparative analysis of other properties these procedures do and
do not satisfy is given in Fishburn (1977). For example, all except Schwartz's procedure
satisfy the Pareto condition (2.1), and all except Nanson's and Dodgson's procedures
are monotonic. Another appealing property is Smith s Condorcet principle Smith
(1973)], which says that if X can be partitioned into nonempty A and B such that
a >M b for all (a, b) E A x B, then F contains no candidate in B. This is violated by
procedures 7, 9, 11 and 12, but it holds for the others.

9. Positional scoring procedures and Borda choices

Along with XI = m > 3 and k = 1, we assume throughout this section that the ballot
set B is a set of linear orders or strict rankings of X, except when noted otherwise. We
take D = 3n when n is fixed. It is convenient, however, in the present setting to extend
our prior definition of a social choice function by letting n range over the positive
integers with

D + =31 BU 2 U 33 U ...

and with F+ defined on {X} x DT. The extended form is used in axiomatizations of
positional scoring procedures, which we will consider shortly.

9.1. Positionalscoringprocedures
Positional scoring procedures include Borda's method and those in which the
differences between the points awarded to candidates in successive positions on a
voter's ballot are not equal. We denote by sj the points awarded to a candidate in
position and refer to s = (sls 2, . .. , s,,) as a positional scoring vector. It is assumed
that sl > 2 > .-- > s,, and s > s,. Borda's method as described in Section 1 has
s = (m- ,m-2, ... , 1,0).
Ch. 4: Voting Procedures 213

For every x E X, every j E 1, ... , m}, and every ballot response profile
d = (dl, ... , dn) in B", let d(x,j) denote the number of voters who rank candidate x
injth position. Clearly, d(x, 1) + d(x, 2) + -.. + d(x, m) = n. The score of candidate x
for ballot response profile d with respect to positional scoring vector s is
m
s(x, d) = sjd(x,j).
j=l
The positional scoring procedure for s takes F(X, d) or F+(X, d) as the subset of
candidates that maximize s(x, d) over X for each d in D or D+.
The positional scoring procedures for s and s' are equivalent when there is a
positive number a > 0 and a real number fi such that s' = as + (, ... , /3); then
s'(x, d) = as(x, d) +/3n, and the maximizing subsets for s and s' are identical. Moreover,
if s and s' are not so related, then in the extended formulation there will be a d G D+
at which F+(X, d) differs for s and s'. Because of equivalence, we can set Sm = 0 with
no loss of generality. If, in addition, s were fixed at 1, then each different positional
scoring procedure in the extended formulation would be characterized by a unique s.
Plurality voting has s = (1,0,..., 0); the procedure which assigns 3 points to a
first-place candidate, 1 point to a second-place candidate, and zero points thereafter
has s = (3, 1, 0, ... , 0). Apart from plurality voting, positional scoring procedures are
seldom used in practice. The one exception is the use of Borda's method in elections
with small numbers of candidates and voters. We note for the Borda procedure with
s = (m - 1,m - 2, ... , 1,0) that s(x,d) is identical to the aggregate number of
candidates ranked lower than x on all ballots. We used this fact in describing Black's
procedure and Nanson's procedure in the preceding section.

9.2. Axioms
Positional scoring procedures have been axiomatized by Smith (1973) and Young
(1975a), and axiomatic characterizations of Borda's method are given in Young (1974),
Hansson and Sahlquist (1976), and Nitzan and Rubinstein (1981). We note versions
of these and then show how their extended formulation yields axioms for approval
voting.
Positional scoring procedures are not subject to the multiple-districts paradox of
Section 7.3. To express this axiomatically, we use Young's formulation in which H is
the set of all functions t that map the m! linear orders on X into nonnegative integers
with ot > 0 for some order. Each summary profile rt, which presumes anonymity, tells
how many voters have each linear order for a ballot response profile in the D+ setting.
For convenience we let C(Tr) = F+(X, d) when d generates T. Young's (1975a) axiom
that avoids the multiple-districts paradox is referred to as
consistency: C(r) n C(;r') 0 X C(rT + a') = C(t) n C(tr'),
where ( + Tr')(>-) = r(>-) + or'(>-). This says that if two disjoint groups of voters
have some candidate in common in their social choice sets, then the choice set of the
combined groups consists of the common choices of the separate groups.
214 S.J. Brains and PC. Fishburn

Young's other axioms for positional scoring procedures are neutrality, monotonicity
(to obtain sl > s2 > ... > s,,), the nonconstancy condition for C (which then gives
sl > Sm), and the following condition of

continuity: If C(Z) = {x} and AT'is any other member of H,

then C(Nat + aT') = {x} for all sufficiently large integers N.

Young shows that consistency, neutrality, and continuity characterize C as a scoring

procedure for some s = (sl ... s,,); then monotonicity and nonconstancy yield
s > ... > m with s > s,,.
Theorem 9.1. Suppose C maps H into the nonempty subsets of X. Then C is a
positional scoring procedure if and only if it is nonconstant, neutral, monotonic,
consistent, and continuous.
Myerson (1995b) generalizes this theorem by not requiring voters to have linear
preference orders. He refers to the preceding consistency and continuity conditions as
"reinforcement" and "overwhelming majority," respectively. Young (1974) specializes
the preceding theorem to Borda's procedure, which is characterized by neutrality,
consistency,faithfulness [if n = 1, C(d) contains only the voter's first place candidate],
and a "cancellation property" whose primary function is to ensure that

s s2 = S2 -3 = = Snm l - S-.

Other conditions on a C of Theorem 9.1 that imply this equal-successive-difference

property for s are noted in Section 9.4.
Young's approach motivated the axiomatization of approval voting in Fishburn
(1978a). For the nonranked context, let H'*be the set of functions at that map the
subsets of X (approval ballots) into nonnegative integers with at > 0 for some subset,
and let o[x] = Z{:(A): x E A}, the number of voters whose ballots contain
candidate x. The approval voting choice set for at E * is the subset of candidates
that maximize at[x] over X.
Theorem 9.2. Suppose C maps HI:' into the nonempty subsets of X. Then
C is the approval voting procedure if and only if it is neutral, consistent
n C(a') = • C(: +
[C(t) N0 +r) = C() C(at ')], and satisfies the disjoint equality
property which says that if t consists of exactly two ballots A and B with A X B and
A B=0, then C(r)=A U B.
A different characterization of approval voting that features strategyproofness with
dichotomous preferences (see Theorem 6.9) is included in Fishburn (1979a).

9.3. Paradoxes
Paradoxes of positional voting arise from the algebraic structure of positional scoring
procedures and their sensitivity to perturbations in ballot response profiles and
Ch. 4: Voting Procedures 215

positional scoring vectors [Zwicker (1991)]. They include choice-setparadoxes, which

focus on F or C, and rankingparadoxes, which consider the ways in which positional
scoring procedures rank candidates according to values of s(x, d) over X. The most
thorough analyses of these and many other paradoxes are included in Saari (1987, 1989,
1992, 1994, 1995a,b, 2000a,b, 2001b), Chapter 25 in Volume 2 of this Handbook, and
references cited below. We begin with examples of choice-set paradoxes.
Condorcet's "other paradox" [Condorcet (1785), Fishburn (1974a)] occurs when
there is a strict Condorcet candidate and every positional scoring procedure would
choose another candidate when sl > s2 > s3 > ... > s,,. The seven-voter response
profile in which

3 voters have xab

2 voters have abx
1 voter has axb
1 voter has bxa

yields x as the strict Condorcet candidate. However, s(a, d) - s(x, d) = s2 - s3, so x is

never in the choice set of a positional scoring procedure when S2 > 3.
Another choice-set paradox occurs when a winner turns into a loser after candidates
other than the winner are removed from X. Fishburn (1974b) constructs a profile for
any m > 3 candidates with a unique Borda winner x such that, for every Y c X with
x C Y and YI > 2 (except for one such Y with YI = 2), x is a Borda loser when the
Borda scores are recomputed on the basis of Y.
Removal of a candidate from X can affect the s-order of the remaining candidates
in specific or in arbitrary ways [Davidson and Odeh (1972), Fishburn (1974a, 1981b),
Saari (1982)]. Consider Borda's procedure applied to the seven-voter profile in which

3 voters have cbax

2 voters have baxc
2 voters have axcb.

The Borda scores for a, b, c and x are 13, 12, 11 and 6, respectively, so the Borda
order is a > b > c > x. When x is removed and Borda scores are recomputed for the
reduced profile, the Borda order is c > b > a, a complete inversion from the original.
Fishburn (1981b) generalizes this for any m > 3 by considering any s = (s, s2, ... , s)
with sl > s2 > > Sm, and any t = (tl,t 2 , ... , tn2 ) with tl > t2 > ... > t 1.
Let X = {x,x2, .. , Xm}. Given s and t, there is a profile r E H whose best-to-
worst s-order for X is x lx 2 ... Xm, whose t-order for X \ {xl } is xx,,_l
- " x 2 , and
whose t-order for X \ {x,m} is Xm, - 1 X2X. Other profiles give a complete inversion
of the remaining candidates when an intermediate member of the s-order is removed.
Saari (1982) generalizes this by allowing s and t to be any nonconstant vectors, and by
prespecifying an s-order, a candidate to be removed, and a t-order on the remainder.
Then there is a profile that produces the prespecified orders.
216 S.J Brains and PC. Fishburn

Fishburn (1981b) also considers complete inversions without removals. Let s

and s' be any two nonequivalent positional scoring vectors for m candidates with
sl > .. > s, and s > ... > s, . Then there is a C 17 with s-order
xlx2 ... x,,, and s'-order x,"" x 2xl. Saari (1984) generalizes this by considering any
h > 2 nonconstant and not necessarily monotonic scoring vectors s ... , sh and any
prespecified linear orders 1, . . , h on X. He proves that if s ... , s and (1, ... , 1)
are linearly independent, then there is a zr e II whose sJ-order is lj forj = 1, . . ., h.

9.4. In praise of Borda, mostly

Borda's procedure occupies a unique place among all positional scoring procedures
by being less susceptible than all other procedures to many unsettling possibilities and
paradoxes. For example, all positional scoring procedures are susceptible to strategic
manipulation [Nitzan (1985)], but Borda's procedure is least susceptible [Saari (1990a,
2001b)]. The next several paragraphs note other results favorable to Borda.
We begin with deterministic results. Smith (1973) showed for m > 3 that if d D+
has a strict Condorcet candidate x then SB(X, d) > SB(y, d) for some y X \ {x},
where sB denotes Borda's procedure. However, if s is not Borda's procedure, then there
is a d D+ with a strict Condorcet candidate x such that s(y, d) > s(x, d) for every
y EX\ {x}.
Saari (1987) generalizes this as follows. Let S denote a function on {A C X: AI > 2}
that assigns a positional scoring procedure s(A) to each such A: if AI = j,
s(A) = (s(A)I, ... , s(A)j), with s(A) necessarily the plurality or simple majority
procedure when A = 2. Let R on {A C X: A > 2} assign a weak order R(A) to
every such A, and let SB denote the S composed entirely of Borda procedures. Suppose
m = 3. If S = SB, then there is an R such that, for every Jr E H, the s(A)-order for r
on A is not the same as R(A) for at least one A. However, if S is not equivalent to SB,
then for every R there is a JZE H such that the s(A)-order for Jr on A equals R(A) for
all A C X with AI > 2.
Saari (1989) goes further. For m > 3, let R[S] denote the set of all R for which
there is a X E IH such that the s(A)-order for 7r on A equals R(A) for all A C X
with AI > 2. Then, for every S that is not equivalent to SB, R[SB] is a proper subset
of R[S]. In other words, if something can happen with the Borda assignment, then it
also happens to every other S assignment or, in Saari's words [Saari (1989, p. 454)],
"any fault or paradox admitted by Borda's method also must be admitted by all other
positional voting methods". For a characterization of R[SB], see Saari (1990b).
We now turn to probabilistic results under the assumption that every voter
independently selects a linear order for d or r at random. Gehrlein and Fishburn
(1978b) prove for m = 3 and n --+ oc that, among all positional scoring procedures, the
Borda procedure uniquely maximizes the probability that s elects a strict Condorcet
candidate, given that such a candidate exists. Van Newenhizen (1992) proves the same
thing for fixed n. Tataru and Merlin (1997) prove for m = 3 and n -- oc that,
among all positional scoring procedures, the Borda procedure uniquely minimizes the
Ch. 4: Voting Procedures 217

probability that the s-order has a strict Condorcet candidate in last place, given that
such a candidate exists.
We note two other results for the uniform-distribution probability model and n --* o.
First, if m E {3,4}, then the Borda procedure maximizes the probability that the
s-order between any two candidates is the same as the simple majority relation between
the two [Gehrlein and Fishbum (1980)]. For the other result, let P(s,t) be the
probability for n oc that the s-winner with s = (sl, ... , Sm) for X is also the
t-winner with t = (tl,... , t- ) for X \ {y} after one y • x is randomly removed
from X. Then Pm(s, t) is uniquely maximized when both s and t are Borda procedures
[Gehrlein, Gopinath, Lagarias and Fishburn (1982)].
Despite Borda's pre-eminence among positional scoring procedures, it does have
defects illustrated by paradoxes described earlier. Moreover, it is almost certainly
more susceptible to manipulation than approval voting. Consider, for example, a
preference profile v = (abc,abc, abc, bca, bca). Recognizing the vulnerability of their
first choice a, the first three voters might rank the candidates insincerely as acb on
their ballots, maximizing the difference between a and its closest competitor b. This
would make a the Borda winner.
Recently, Sertel and Yilmaz (1999) and Brams and Kilgour (2001) independently
proposed a procedure in which, in the 5-voter example of the preceding paragraph, a
would be chosen by sincere voters if the decision rule, or quota q, were simple majority,
but b would be chosen if q were unanimity. The procedure works by having voters rank
candidates from best to worst. If at least q voters rank a candidate first, that candidate
is chosen; if not, then one next asks if there are at least q voters who rank a candidate
either first or second - and so on, descending to lower and lower levels in the rankings
until there is agreement by at least q voters on a candidate or candidates. Thus, if q = 3
(simple majority), there is agreement on a, based only on first choices, making a the
"majoritarian compromise" [Sertel and Yilmaz (1999)]. If q = 5 (unanimity), there is
no agreement without descending to second choices, at which level all 5 voters rank b
either first or second, making b the "fallback bargaining" choice [Brams and Kilgour
(2001)]. In a voting context, Sertel and Yilmaz (1999) argue that simple majority
is sensible, whereas in a bargaining context Brams and Kilgour (2001) argue that
unanimity is sensible. Whatever the decision rule, this procedure may not select a
Condorcet candidate, but the candidate or candidates chosen by it are always Pareto-
optimal - there are no other candidates that all voters prefer - and maximizes the
minimum "satisfaction" (based on rankings) of the q most satisfied voters.
Manipulation is quite difficult under this procedure [Brams and Kilgour (2001)], as
it is under many other voting procedures. But the Borda procedure is an exception:
voters can gain by ranking the most serious rival of their favorite candidate last, which
is a relatively easy strategy to effectuate, in order to lower the rival's point total [Ludwig
(1978), Dummett (1998)].
218 S.J Brains and PC. Fishburn

10. Point distribution procedures

In this section and the next we consider choose-k social choice functions for k > 2.
Two common choose-k procedures for small k are the nonranked procedures that ask
voters to vote for exactly k candidates, or for no more than k candidates. The top k vote
getters, or more if there is a tie for kth place, are the winners. The same criterion can
be used with approval voting, positional scoring procedures, and other procedures used
primarily for choose-1 situations. A different criterion, referred to as a cutoff or quota,
does not specify k in advance but elects every candidate whose vote count exceeds
the cutoff. This is frequently used by groups to elect new members or to bestow an
honorific title on present members.
The literature for choose-k procedures is, apart from that for proportional repre-
sentation, comparatively sparse. Examples include Fishburn (1981a), Brams (1982,
1990), Gehrlein (1985), Staring (1986), Bock, Day and McMorris (1998), Barberr,
Sonnenschein and Zhou (1991), Debord (1992) and Brams and Fishburn (1992,
1993). Staring (1986) gives an example of voters with linear preference orders who
vote sincerely under the vote-for-exactly-k procedure, which illustrates an increasing-
committee-size paradox: the winners for k = 3 are disjoint from the winners for k = 2,
and the winners for k = 4 are disjoint from those for k E {2, 3}. Debord (1992) gives
an axiomatic choose-k generalization of Young's (1974) Borda axiomatization.
All voting procedures described previously use nonranked or ranked ballots that
do not allow voters to express intensities of preference in a more complete manner.
Point distribution procedures accommodate this possibility by asking each voter to
distribute a fixed number of points, say 100, to the candidates in any way he or she
please. The k candidates with the most points are the winners. The usual term for
such a procedure is cumulative voting [Glasser (1959), Brams (1975), Bolger (1983,
1985)]. It has been used by corporations to elect boards of directors, and may be
viewed as a method for proportional representation in which minorities can ensure
their approximate proportional representation by concentrating their votes on a subset
of candidates commensurate with their size in the electorate. Indeed, cumulative voting
is one of a class of voting procedures that encourage minority representation [Guinier
(1994)] and maximize majority welfare [Chwe (1999)].
To illustrate cumulative voting and the calculation of optimal strategies, suppose
there is a single minority position among the electorate favored by one-third of
the voters. The other two-thirds favor a majority position. Assume that n = 300,
six candidates are to be elected (k = 6), and each voter has six votes (points) to
distribute over the candidates. The minority controls 600 votes, and the majority
controls 1200 votes. Hence if the minority divides its votes equally between two
minority candidates (600/2 = 300 each), it can ensure their election no matter what the
majority does. If the two-thirds majority instructs its supporters to distribute their votes
equally among five candidates (1200/5 = 240), it will not match the vote totals of the
two minority candidates but can still ensure the election of four of its five candidates -
Ch. 4: Voting Procedures 219

and possibly get its fifth candidate elected if the minority splits its votes equally among
three minority candidates (600/3 = 200).
Against these majority (support five) and minority (support two) strategies, it is
easy to show that neither side can improve its position. To elect five rather than four
candidates with 301 votes each, the majority would need 1505 instead of 1200 votes;
similarly, to elect three rather than two candidates with 241 votes each, the minority
would need 723 instead of 600 votes.
It is evident that the optimal strategy for the leaders of both the majority and minority
is to instruct their members to allocate their votes as evenly as possible among a certain
number of candidates. The number to support should be proportionally about equal to
the number of their supporters in the electorate (if known).
Any deviation from this strategy - for example, by putting up a full slate of
candidates and not instructing supporters to vote for only some on this slate - offers the
other side an opportunity to capture more than its proportional "share" of the k seats.
Patently, good planning and disciplined supporters are needed to carry out an optimal
strategy.
Brams (1975) includes a systematic analysis of optimal strategies under cumulative
voting. These strategies are compared to strategies actually adopted by the Democratic
and Republican parties in elections for the Illinois General Assembly, where cumulative
voting was used until 1982. Cumulative voting was adopted by two cities in the United
States (Alamogordo, NM, and Peoria, IL) in 1987, and other small cities more recently,
to satisfy court requirements of minority representation in municipal elections.
Bolger (1983, 1985) formulates six procedures for cumulative voting in choose-k
elections and investigates their susceptibility to several paradoxes. Each procedure
allots k points to each of n voters to distribute over the candidates and uses an election
quota qo = (nk + 1)/(k + 1). Any candidate who receives at least q votes is elected
in an initial stage. The procedures differ in their vote distribution rules and in how
votes are processed after the initial stage if fewer than k are elected there. In some
procedures, a voter votes for h < k candidates, and each of the h gets k/h votes from
the voter; others allow the k points to be distributed in any way among h < k or among
any number of candidates. Vote processing after the initial stage may involve transfers
of surplus votes above qO from initial electees to others, or elimination of low-ranking
candidates. The paradoxes include violations of monotonicity and new voter and no-
show paradoxes. The new-voter paradox occurs when a new voter who votes only for
the original k electees causes one of these for whom he or she votes to become a
loser in the augmented profile. The no-show paradox occurs when an original electee
turns into a loser after a ballot involving only original losers is deleted from the ballot
response profile. All six procedures exhibit the latter two paradoxes when k > 4, and
all but two do this when k > 2.

11. Proportional representation

Unlike cumulative voting, most choose-k procedures use ballot types discussed earlier.
220 SJ Brams and PC. Fishburn

We have already noted common nonranked procedures for electing committees, and in
this section we consider other procedures designed to elect representative legislatures
and governing bodies.

11.1. The Hare system of single transferable vote

First proposed by Thomas Hare in England and Carl George Andrae in Denmark in
the 1850s, single transferable vote (STV) procedures have been adopted throughout the
world. They are used in such countries as Australia, Malta, the Republic of Ireland, and
Northern Ireland; in local elections in Cambridge, MA, and formerly in other cities in
the United States [Tideman (1995)]. John Stuart Mill (1862) placed STV "among the
greatest improvements yet made in the theory and practice of government". Although
STV violates some desirable properties of voting procedures [Kelly (1987)], it has
strengths as a method of proportional representation. In particular, minorities can elect
a number of candidates roughly proportional to their numbers in the electorate. Also,
if one's vote does not help elect a first choice, it can still count for lower choices.
To define one version of STV with IX = m, suppose k of the m candidates are
to be elected by n ballots which rank from 1 to m candidates. (In practice, voters
are encouraged to rank as many candidates as possible.) The point quota needed for
election is

q Lk 2 +1,

where zj is the integer part of z. We denote bypi the points for ballot i. Initially, pi = 1,
but pi can change during the ballot-processing stages because (1) the top candidate not
yet removed from ballot i is elected, or (2) no candidate is left on ballot i, or (3) no
candidate is left on other ballots. The initial Pi sum is n; afterj candidates have been
elected, the revised Pi sum is n -jq. Whenever points are counted to determine if new
candidates reach q, the pi points of ballot i are awarded to the top-ranked candidate
remaining on ballot i.
Let e denote the number of candidates elected thus far, and let A denote the subset
of candidates still in contention. The following steps are used to move e from 0 to k.
Step 0: Set e = 0, A = X, andpi = 1 for all i. Go to step 1.
Step 1: If e + IA < k, declare all candidates in A as elected, and if e + A I < k, choose
k - (e + IA ) of the not yet elected m (e + IA ) candidates at random, declare
them elected also, and stop. If e + A[ > k, for each x E A compute p(x) as the
sum of the Pi for all ballots that rank x first, then let E = {x e A: p(x) > q},
and declare the candidates in E as elected. If e + IEI = k, stop. Otherwise,
change e to e + El, go to step 2 if EI > 1, and go to step 3 if El = 0.
Step 2: For each x E E, let , = q/p(x), and for each ballot with x ranked first, replace
Pi by (1 - ,x)pi. This removes q points from the process for each newly elected
candidate in E. Delete all x E E from all ballots, change A to A \ E, and go
to step 4.
Ch. 4: Voting Procedures 221

Step 3: Determine the candidate in A, say y, with the minimum p(x). (If two or more
in A have the min p(x) value, choose one at random for y.) Delete y from all
ballots, change A to A \ y}, and go to step 4.
Step 4: Let P be the sum of the Pi for ballots that, because of deletions, have no
remaining candidates, set pi = 0 for these ballots and, when n' nonempty
ballots remain, increase the Pi of each by adding P/n'. Go to step 1.
Step 4 is used to maintain the current point total when all candidates ranked on
a ballot have been elected or deleted. When y is ranked first on a ballot in step 3,
its Pi at that point is transferred to the second-ranked candidate if there is one. The
surplusp(x) - q of points needed for election of a newly elected candidate in step 1 is
retained by step 2 for the ballots that rank the elected candidate first, while q points are
removed from those ballots, but if the set E of newly elected candidates exhausts all
that remain on a ballot, its adjusted points get transferred to other ballots in step 4.
The paradoxes described in Section 7.3 for plurality with a runoff apply to STV
when there are three candidates and k = 1 [Doron and Kronick (1977), Fishburn and
Brams (1983)]. The following examples [Brams (1982), Brams and Fishburn (1984c)]
illustrate the mechanics of STV and phenomena associated with truncated rankings.
Example 11.1. Assume that two of four candidates are to be elected, and there are
three classes of voters who rank the candidates as follows:

I. 6 voters have xabc

II. 6 voters have xbca
III. 5 voters have xcab.

Then n = 17, so q = [17/3] + 1 = 6. The initial point totals are 17 for x and 0 for the
others, so x is elected. The surplus of 11 = 17 - 6 points for x are redistributed in the
proportions 6: 6 : 5 to the classes, so I and II are left with 66/17 - 3.9 points each,
and III is left with 55/17 - 3.2 points. Candidate x is deleted (step 2) and, since none
of the others has q revised points, c is deleted (lowest total, step 3) to give

I. 66/17 points, ab
II. 66/17 points, ba
III. 55/17 points, ab.

Then a (7.1 points) is elected along with x.

Now suppose that two of the six class II voters had ranked only their first choice x.
As before, x is elected on the first round. Its deletion, and points reductions of step 2,
give

I. 66/17 points, abc

II.1. 22/17 points, no remaining candidates
11.2. 44/17 points, bca
III. 55/17 points, cab.
222 S.J Brans and PC. Fishburn

We now use step 4 with P = 22/17 and n' = 15:

I. 66/17 + (6/15)(22/17) = 4.4 points, abc

11.2. 44/17 + (4/15)(22/17) = 2.933 ... points, bca
III. 55/17 + (5/15)(22/17) = 3.666 ... points, cab.

Since none of a, b and c here makes q, the low candidate b is eliminated and c is
elected with 6.6 points.
Observe that the two class II voters who ranked only x induced a better second
choice (c instead of a) for themselves by submitting truncated ballots. Thus, it may
be advantageous not to rank all candidates on one's ballot, contrary to a claim made
by a professional society that "there is no tactical advantage to be gained by marking
few candidates" [Brams (1982)]. Put another way, one may do better under STV by
not expressing preferences - at least beyond first choices. U
Lest one think that an advantage gained by truncation requires allocation of surplus
votes, we give a truncation example for k = 1. Here STV is similar to plurality with
successive elimination (Section 7.1), but with the added feature of ranked ballots.
Example 11.2. Assume that one of four candidates is to be elected by 21 voters:

I. 7 voters have abcx

II. 6 voters have bacx
III. 5 voters have cbax
IV. 3 voters have xcba.

Here q = 11. No candidate makes q initially, so x is eliminated and a, b and c then

have 7, 6 and 8 votes, respectively. Because none of these makes q, b is eliminated and
a is elected with 13 = 7 + 6 votes even though b is the strict Condorcet candidate.
Now suppose the three class IV voters rank only x as their first choice. As before, x
is eliminated first, and since the ballots of IV have no other candidates, their 3 points
go to the others:

I. 7 + (7/18)3 = 49/6 points, abc

II. 6 + (6/18)3 = 7 points, bac
III. 5 + (5/18)3 = 35/6 points, cba.

Now c is eliminated and b is the winner with 7 + 35/6 - 12.8 votes. Because the
class IV voters prefer b to a, it is in their interest not to rank candidates below x. U
It is true under STV that a first choice can never be hurt by ranking a second choice,
a second choice by ranking a third choice, ... , because higher choices are eliminated
before the lower choices can affect them. However, lower choices can affect the order of
elimination and, hence, transfer of votes. Consequently, a higher choice can influence
whether a lower choice is elected.
Ch. 4: Voting Procedures 223

We do not suggest that voters would routinely make the strategic calculations in
Examples 11.1 and 11.2. Such calculations are not only complex but also might be
neutralized by counterstrategies of other voters. Rather, the point is that to rank all
candidates for whom one has preferences is not always rational under STV Additional
discussion of STV's manipulability in this regard, and its relationship to the election
of Condorcet candidates, is in Fishburn and Brams (1984).

11.2. Additional-member systems

In most parliamentary democracies, it is not candidates who run for office but political
parties that put up lists of candidates. Under party-list voting, voters vote for parties,
which receive seats in a parliament proportional to the total numbers of votes they
receive. There is often a threshold, such as 5% of the total vote, which a party must
exceed to gain any seats.
This is a rather straightforward procedure of ensuring proportional representa-
tion (PR) of parties that surpass the threshold, though it is not paradox-free with
respect to the distribution of seats that take account of the complete preference orders
of voters [Van Deemen (1993)]. More interesting are systems in which some legislators
are elected from districts, but new members may be added to ensure that parties
underrepresented on the basis of their national-vote proportions gain additional seats.
Denmark and Sweden, for example, use votes summed over each party's district
candidates as the basis for allocating additional seats. In elections to Germany's
Bundestag and Iceland's Parliament, voters vote twice, once for district representatives
and once for a party. Half of the Bundestag is chosen from party lists, on the basis
of the national party vote, with adjustments to the district results made to ensure
approximate PR of parties. Italy, New Zealand, and several Eastern European countries
and former Soviet republics have recently adopted similar systems. In Puerto Rico, if
the largest party in one house of its bicameral legislature wins more than two-thirds of
the seats in district elections, then that house can be increased by as much as one-third
to redress underrepresentation of minority parties.
We offer insight into an important strategic feature of additional-member systems
by assuming, as in Puerto Rico, that a variable number of additional members can be
added to a legislature to adjust for underrepresentation. We consider a procedure, called
adjusted district voting, or ADV [Brams and Fishburn (1984a,b)], that is characterized
by four assumptions:
(1) There is ajurisdiction divided into equal-size districts, each of which elects a single
representative to a legislature.
(2) The jurisdiction has two main factions, one majority and one minority, whose sizes
can be determined.
(3) The legislature consists of the district winners plus the largest vote-getters among
the losers - necessary to achieve PR - if PR is not realized by the district winners.
This addition would typically be minority-faction losers in district elections.
224 SJ.Brams and PC. Fishburn

(4) The legislature's size is variable, with a lower bound equal to the number of
districts (if no additions are needed to achieve PR), and an upper bound equal to
twice the number of districts (if a nearly 50% minority wins no district election).
To illustrate ADV, suppose the jurisdiction has eight districts with an 80% majority
faction and a 20% minority faction. If the minority wins no district election, then its
two biggest vote-getters could be given seats in a 10-member parliament that achieves
PR exactly.
Now suppose the minority wins one seat, so its initial representation is 8, or about
13%. If it were given an additional seat, its representation would rise to 9 (22%),
which is closer than 8 to its 20% proportion in the electorate. Assume, however, that
additions can never make its proportion in the legislature exceed its proportion in the
electorate, so the addition is not made.
Paradoxically, the minority would benefit by winning no district election. To prevent
a minority from benefiting by losing in district elections, assume the following no-
benefit constraint:the allocation of extra seats to the minority can never give it a greater
proportion in the legislature than it would obtain had it won more district elections.
Because 1 < 1 < 2 2, this implies that if the minority wins in no district, then it can
be given only one rather than two seats for a representation of ~ (11%) rather than 2
(20%).
It can be proved in the general case that the no-benefit constraint may prevent
a minority from receiving up to about half of the extra seats it would be entitled
to otherwise [Brains and Fishburn (1984a)]. This constraint can be interpreted as
a sincere-voting promoter in ADV It makes it unprofitable for a minority party
deliberately to lose district elections in order to do better with extra-seat additions.
This comes at a price, however. As our example and its generalization demonstrate,
the constraint can severely restrict the ability of ADV to satisfy PR, giving rise to the
following dilemma: under ADV, one cannot assure a close correspondence between
a party's proportion in the electorate and its representation in the legislature if one
insists on the no-benefit constraint; dropping it allows one to approximate PR, but this
may give the minority party an incentive purposely to lose in certain district contests
in order to do better after the adjustment.
It is worth noting that the "second chance" for minority candidates afforded by ADV
would encourage them to run in the first place, because even if most or all lose their
district races, their biggest vote-getters would still have a chance at extra seats. But
these extra seats might be cut by up to a factor of two from the minority's proportion
in the electorate should one want to motivate district elections with the no-benefit
constraint. Indeed, [Spafford (1980, p. 393)], anticipating this dilemma, recommended
that only an (unspecified) fraction of seats that the minority is entitled to be alloted to
it in the adjustment phase to give it "some incentive to take the single-member contests
seriously, ... , though that of course would be giving up strict PR".
Ch. 4: Voting Procedures 225

11.3. Minimizing representationalimbalance

We conclude our discussion of PR with a few comments occasioned by Monroe's

(1995) proposal to select winning candidates in an election for a legislature by
minimizing an aggregate measure of representational imbalance. Such a measure
would depend on the ballot type and how ballots are assessed with regard to
representativeness, but in any case it is a function of potential winning sets of
k candidates.
Let A = {A C X: A = k} and for A A letf map {1, ... , n} into A. If ballots are
approval ballots, the misrepresentation score for voter i underf is 0 iff(i) is in voter i's
approved set, and is 1 otherwise. If ballots are linear orders, the misrepresentation
score for voter i underf is j - 1 whenf(i) is jth-ranked in voter i's order. The total
misrepresentation of assignmentf is the sum over i of the voters' misrepresentation
scores.
Monroe (1995) suggests thatf be restricted so that approximately the same number
of voters are assigned to each candidate, or "represented" by each candidate, in A.
Subject to this restriction, one then determines the elected set to be an A A for which
the minimum total misrepresentation of an f for A is as small as possible. Potthoff
and Brams (1998) note that this is the same as a proposal of Chamberlin and Courant
(1983) when no restrictions are placed on f (a proposal rejected by Monroe), and
that if, in addition, k is unrestricted, a proposal of [Tullock (1967, Chapter 10)] is
obtained.
Potthoff and Brams (1998) demonstrate the efficiency of using integer programming
to compute a solution for Monroe's procedure as well as for a variety of related
procedures. One of these uses anf that maps {1, ... , n} into h-candidate subsets of
A with 1 < h < k, restricted so that each candidate is in the h-candidate subsets of
approximately hn/k voters. When h = k with approval ballots, the elected A consists of
the k candidates with the greatest approval votes. When h = k with fully ranked ballots,
the elected A is the set of k candidates with the most Borda points. Intermediate values
of h may be more faithful to the intention of electing a proportionately representative
legislature.

12. Conclusions

There is no perfect voting procedure [Niemi and Riker (1976), Fishburn (1984), Nurmi
(1986), Amy (2000)], but some procedures are clearly superior to others in satisfying
certain criteria.
Among nonranked voting procedures to elect one candidate, approval voting
distinguishes itself as more sincere, strategyproof, and likely to elect Condorcet
candidates than other procedures, including plurality voting and plurality with a runoff.
Its use in earlier centuries in Europe [Cox (1984, 1987a), Lines (1986)], and its recent
adoption by a number of professional societies - including the Institute of Management
226 S.J Brains and PC. Fishburn

Sciences [Fishburn and Little (1988)], the Mathematical Association of America

[Brams (1988)], the American Statistical Association [Brams and Fishburn (1988)],
the Institute of Electrical and Electronics Engineers [Brams and Nagel (1991)], the
American Mathematical Society, and the Social Choice and Welfare Society [Brams
and Fishbum (2001), Saari (2001a)] - augurs well for its more widespread use,
including possible adoption in public elections [Brams (1993), Brams and Herschbach
(2001)]. Bills have been introduced in several U.S. state legislatures for its enactment
for state primaries, and its consideration has been urged in such countries as Finland
[Anckar (1984)] and New Zealand [Nagel (1987)]. Its probable effects in the U.S.
presidential elections of 1864 [Tabarrok and Spector (1999)], 1980 [Brams and
Fishburn (1983)], 1992 [Brams and Merrill (1994), Tabarrok (2001)], and 2000 [Saari
(2001b)], along with the likely effects of other voting procedures such as the Borda
count, have been assessed.
Among ranked positional scoring procedures to elect one candidate, Borda's
method is superior in many respects, including susceptibility to strategic manipulation,
propensity to elect Condorcet candidates, and ability to minimize paradoxical
possibilities [Smith (1973), Gehrlein and Fishburn (1978b), Saari (1989, 1990a, 1994,
1995a,b, 2000a,b, 2001b), Chapter 25 in Volume 2 of this Handbook, Van Newenhizen
(1992)]. Some Condorcet voting procedures, such as the Schwartz and Kemeny
procedures, have a number of attractive properties [Fishburn (1977), Young (1988)],
but they have witnessed more theoretical than practical interest. Despite Borda's
superiority in many respects, it is easier to manipulate than many other procedures.
For example, the strategy of ranking the most serious rival of one's favorite candidate
last is a transparent way of diminishing the rival's chances.
While plurality with a runoff, and STV for elections of one or more candidates, are
commonly used, they are subject to some of the more noxious paradoxes, including
violations of monotonicity which can turn a potential winner into a loser when it
rises in the ballot response profile. Additional-member systems, and specifically ADV
that results in a variable-size legislature, provide a mechanism for approximating
proportional representation in a legislature without the nonmonotonicity of STV or
the manipulability of Borda-type procedures. Cumulative voting also offers a means
for factions or parties to ensure their proportional representation, but it requires
considerable organizational effort on the part of parties. In the face of uncertainty
about their level of support in the electorate, party leaders may well make suboptimal
choices about how many candidates their supporters should concentrate their votes
on, which weakens the argument that cumulative voting can guarantee proportional
representation in practice. But the no-benefit constraint on allocation of additional seats
to underrepresented parties under ADV - in order to deny them the incentive to throw
district races - also vitiates fully satisfying proportional representation, underscoring
the difficulties of satisfying a number of desiderata.
An understanding of these difficulties, and possible trade-offs that must be made,
facilitates the selection of procedures to meet certain needs. Over the past half century
the explosion of results in social choice theory, and the burgeoning decision-theoretic
Ch. 4. Voting Procedures 227

and game-theoretic analyses of different voting procedures, not only enhance one's
theoretical understanding of the foundations of social choice but also contribute to
the better design of practical voting procedures that satisfy the criteria that one deems
important.

References

Abello, J.M. (1981), "Toward a maximum consistent set", Tech. report TRCS 11-81 (University of
California, Santa Barbara).
Abello, J.M. (1991), "The weak Bruhat order of Sz, consistent sets, and Catalan numbers", SIAM
Journal on Discrete Mathematics 4:1-16.
Abello, J.M., and C.R. Johnson (1984), "How large are transitive simple majority domains?", SIAM
Journal on Algebraic and Discrete Methods 5:603-618.
Amy, D.J. (2000), Behind the Ballot Box: A Citizen's Guide to Voting Systems (Praeger, Westport, CT).
Anckar, D. (1984), "Presidential elections in Finland: a plea for approval voting", Electoral Studies
3:125-138.
Arrow, K.J. (1950), "A difficulty in the concept of social welfare", Journal of Political Economy
58:328-346.
Arrow, K.J. (1951), Social Choice and Individual Values (Yale University Press, New Haven, CT);
2nd Edition 1963 (Wiley, New York).
Balinski, M.L., and H.P. Young (1982), Fair Representation: Meeting the Ideal of One Man, One Vote
(Yale University Press, New Haven, CT).
Balinski, M.L., and H.P Young (2001), Fair Representation: Meeting the Ideal of One Man, One Vote,
2nd Edition (Brookings Institute, Washington, D.C.).
Banks, J.S. (1985), "Sophisticated voting outcomes and agenda control", Social Choice and Welfare
2:295-306.
Banzhaf III, J.E (1965), "Weighted voting doesn't work: a mathematical analysis", Rutgers Law Review
19:317 343.
BarberB, S., H. Sonnenschein and L. Zhou (1991), "Voting by committees", Econometrica 59:595-609.
Berga, D. (1998), "Strategy-proofness and single-plateaued preferences", Mathematical Social Sciences
35:105-120.
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, Cambridge).
Bock, H.-H., WH.E. Day and FR. McMorris (1998), "Consensus rules for committee elections",
Mathematical Social Sciences 35:219-232.
Boehm, G.A.W. (1976), "One fervent vote against Wintergreen", Mimeograph.
Bolger, E.M. (1983), "Proportional representation", in: S.J. Brams, WE Lucas and P.D. Straffin Jr, eds.,
Modules in Applied Mathematics, Vol. 2 (Springer, New York) pp. 19-31.
Bolger, E.M. (1985), "Monotonicity and other paradoxes in some proportional representation schemes",
SIAM Journal on Algebraic and Discrete Methods 6:283-291.
Borda (J.-C. de Borda) (1781), "M6moire sur les elections par scrutin", M6moires de l'Acad6mie Royale
des Sciences ann6e 1781, pp. 657 665. Translated in English in 1953 by A. de Grazia: "Mathematical
derivation of an election system", Isis 44:42-51.
Bordley, R.F (1983), "A pragmatic method for evaluating election schemes through simulation", American
Political Science Review 77:123-141.
Brams, S.J. (1975), Game Theory and Politics (Free Press, New York).
Brams, S.J. (1977), "When is it advantageous to cast a negative vote?", in: R. Henn and O. Moeschlin,
eds., Mathematical Economics and Game Theory: Essays in Honor of Oskar Morgenstern (Springer,
Berlin) pp. 564-572.
Brams, S.J. (1978), The Presidential Election Game (Yale University Press, New Haven, CT).
228 S.J. Brams and PC. Fishburn

Brams, S.J. (1982), "The AMS nomination procedure is vulnerable to truncation of preferences". Notices
of the American Mathematical Society 29:136-138.
Brams, S.J. (1983), "Comparison voting", in: S.J. Brams, W.E Lucas and PD. Straffin Jr, eds., Modules
in Applied Mathematics, Vol. 2 (Springer, New York) pp. 32-65.
Brams, S.J. (1985), Rational Politics: Decisions, Games, and Strategy (CQ Press, Washington, DC).
Brams, S.J. (1988), "MAA elections produce decisive winners", Focus (Mathematical Association of
America) 8:1-2.
Brams, S.J. (1990), "Constrained approval voting: a voting system to elect a governing board", Interfaces
20:65-80.
Brams, S.J. (1993), "Approval voting and the good society", PEGS Newsletter 3:10, 14.
Brams, S.J. (1994), "Voting procedures", in: R.J. Aumann and S. Hart, eds., Handbook of Game Theory,
Vol. 2 (Elsevier, Amsterdam) pp. 1055-1089.
Brams, S.J., and M.D. Davis (1973), "Models of resource allocation in presidential campaigning:
implications for democratic representation", Annals of the New York Academy of Sciences 219:
105-123.
Brams, S.J., and M.D. Davis (1974), "The 3/2's rule in presidential campaigning", American Political
Science Review 68:113-134.
Brams, S.J., and M.D. Davis (1982), "Optimal resource allocation in presidential primaries", Mathematical
Social Sciences 3:373-388.
Brams, S.J., and PC. Fishburn (1978), "Approval voting", American Political Science Review 72:
831-847.
Brams, S.J., and PC. Fishburn (1981), "Reconstructing voting processes: the 1976 House Majority
Leader election under present and alternative rules", Political Methodology 7:95-108.
Brams, S.J., and P.C. Fishburn (1983), Approval Voting (Birkhuser, Boston).
Brams, S.J., and P.C. Fishburn (1984a), "A note on variable-size legislatures to achieve proportional
representation", in: A. Lijphart and B. Grofman, eds., Choosing an Electoral System: Issues and
Alternatives (Praeger, New York) pp. 175-177.
Brams, S.J., and PC. Fishburn (1984b), "Proportional representation in variable-size legislatures", Social
Choice and Welfare 1:397-410.
Brams, S.J., and P.C. Fishburn (1984c), "Some logical defects of the single transferable vote", in:
A. Lijphart and B. Grofman, eds., Choosing an Electoral System: Issues and Alternatives (Praeger,
New York) pp. 147 151.
Brams, S.J., and P.C. Fishburn (1985), "Comment on 'The problem of strategic voting under approval
voting' ", American Political Science Review 79:816-818.
Brams, S.J., and PC. Fishburn (1988), "Does approval voting elect the lowest common denominator?",
PS: Political Science & Politics 21:277-284.
Brams, S.J., and P.C. Fishburn (1992), "Coalition voting", in: PE. Johnson, ed., Mathematical and
Computer Modelling 16:15-26.
Brams, S.J., and PC. Fishburn (1993), "Yes-no voting", Social Choice and Welfare 10:35-50.
Brams, S.J., and P.C. Fishburn (2001), "A nail-biting election", Social Choice and Welfare 18:409-414.
Brams, S.J., and D.R. Herschbach (2001), "The science of elections", Science 292:1449.
Brams, S.J., and D.M. Kilgour (2001), "Fallback bargaining", Group Decision and Negotiation 10:
287-316.
Brams, S.J., and S. Merrill III (1994), "Would Ross Perot have won the 1992 presidential election under
approval voting?" PS: Political Science and Politics 27:39-44.
Brams, S.J., and J.H. Nagel (1991), "Approval voting in practice", Public Choice 71:1-17.
Brams, S.J., P.C. Fishburn and S. Merrill III (1988a), "The responsiveness of approval voting: comments
on Saari and Van Newenhizen", Public Choice 59:121-131.
Brams, S.J., PC, Fishburn and S. Merrill III (1988b), "Rejoinder to Saari and Van Newenhizen", Public
Choice 59:149.
Ch. 4: Voting Procedures 229

Brams, S.J., D.M. Kilgour and WS. Zwicker (1997), "Voting on referenda: the separability problem and
possible solutions", Electoral Studies 16:359-377.
Brams, S.J., D.M. Kilgour and W.S. Zwicker (1998), "The paradox of multiple elections", Social Choice
and Welfare 15:211-236.
Chamberlin, J.R., and P.N. Courant (1983), "Representative deliberations and representation decisions:
proportional representation and the Borda rule", American Political Science Review 77:718-733.
Chamberlin, J.R., and E Featherston (1986), "Selecting a voting system", Journal of Politics 48:347-369.
Charon, I., O. Hudry and E Woirgard (1997), "A 16-vertex tournament for which Banks set and Slater
set are disjoint", Discrete Applied Mathematics 80:211-215.
Ching, S., and S. Serizawa (1998), "A maximal domain for the existence of strategy-proof rules", Journal
of Economic Theory 78:157-166.
Chwe, M.S.-K. (1999), "Minority voting rights can maximize majority welfare", American Political
Science Review 93:85-97.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Copeland, A.H. (1951), "A 'reasonable' social welfare function", Mimeograph (University of Michigan
Seminar on Applications of Mathematics to the Social Sciences).
Coughlin, P.J. (1992), Probabilistic Voting Theory (Cambridge University Press, Cambridge).
Cox, G.W (1984), "Strategic electoral choice in multi-member districts: approval voting in practice",
American Journal of Political Science 28:722-738.
Cox, G.W. (1987a), The Efficient Secret: The Cabinet and the Development of Political Parties in
Victorian England (Cambridge University Press, New York).
Cox, G.W. (1987b), "Electoral equilibrium under alternative voting institutions", American Journal of
Political Science 31: 82-108.
Cox, G.W (1997), Making Votes Count: Strategic Coordination in the World's Electoral Systems
(Cambridge University Press, New York).
Craven, J. (1996), "Majority-consistent preference orderings", Social Choice and Welfare 13:259-267.
Dasgupta, P., P.J. Hammond and E. Maskin (1979), "The implementation of social choice rules: some
general results on incentive compatibility", Review of Economic Studies 46:185-216.
Davidson, R.R., and R.E. Odeh (1972), "Some inconsistencies in judging problems", Journal of
Combinatorial Theory A 13:162-169.
de Grazia, A. (1953), "Mathematical derivation of an election system" [an English translation of"Memoire
sur les elections par scrutin", by J.-C. de Borda (1781)], Isis 44:42-51.
Debord, B. (1992), "An axiomatic characterization of Borda's k-choice function", Social Choice and
Welfare 9:337-343.
DeMeyer, E, and C.R. Plott (1970), "The probability of a cyclical majority", Econometrica 38:345 354.
Dodgson, C.L. (Lewis Carroll) (1876), A Method of Taking Votes on More than Two Issues (Clarendon
Press, Oxford). Reprinted, 1958, in: D. Black, ed., The Theory of Committees and Elections (Cambridge
University Press, Cambridge) pp. 224-234.
Doron, G. (1979), "The Hare voting system is inconsistent", Political Studies 27:283 286.
Doron, G., and R. Kronick (1977), "Single transferable vote: an example of a perverse social choice
function", American Journal of Political Science 21:303-311.
Dubey, P., and L.S. Shapley (1979), "Mathematical properties of the Banzhaf power index", Mathematics
of Operations Research 4:99-131.
Dummett, M. (1984), Voting Procedures (Clarendon Press, Oxford).
Dummett, M. (1998), "The Borda count and agenda manipulation", Social Choice and Welfare 15:
287-296.
Dutta, B. (1988), "Covering sets and a new Condorcet choice correspondence", Journal of Economic
Theory 44:63 80.
230 S.J. Brains and PC. Fishburn

Enelow, J.M., and M.J. Hinich (1984), The Spatial Theory of Election Competition: An Introduction
(Cambridge University Press, Cambridge).
Enelow, J.M., and M.J. Hinich, eds. (1990), Advances in the Spatial Theory of Voting (Cambridge
University Press, New York).
Epstein, D. (1998), "Uncovering some subtleties of the uncovered set: Social choice theory and distributive
politics", Social Choice and Welfare 15:81-93.
Erd6s, P, and L. Moser (1964), "On the representation of directed graphs as the unions of orderings",
Publications of the Mathematical Institute of the Hungarian Academy of Sciences 9:125-132.
Farquharson, R. (1969), Theory of Voting (Yale University Press, New Haven, CT).
Felsenthal, D.S. (1989), "On combining approval with disapproval voting", Behavioral Science 34:
53-60.
Felsenthal, D.S., and M. Machover (1998), The Measurement of Voting Power (Edward Elgar,
Cheltenham, UK).
Felsenthal, D.S., and Z. Maoz (1988), "A comparative analysis of sincere and sophisticated voting under
the plurality and approval procedures", Behavioral Science 33:116-130.
Fine, K. (1972), "Some necessary and sufficient conditions for representative decision on two alternatives",
Econometrica 40:1083-1090.
Fishburn, P.C. (1970), Utility Theory for Decision Making (Wiley, New York).
Fishburn, P.C. (1971), "The theory of representative majority decision", Econometrica 39:273-284.
Fishburn, P.C. (1973), The Theory of Social Choice (Princeton University Press, Princeton, NJ).
Fishburn, P.C. (1974a), "Paradoxes of voting", American Political Science Review 68:537-546.
Fishburn, P.C. (1974b), "On the sum-of-ranks winner when losers are removed", Discrete Mathematics
8:25-30.
Fishburn, P.C. (1977), "Condorcet social choice functions", SIAM Journal on Applied Mathematics
33:469-489.
Fishburn, P.C. (1978a), "Axioms for approval voting: direct proof", Journal of Economic Theory 19:
180-185; Corrigendum 45 (1988) 212.
Fishburn, P.C. (1978b), "A strategic analysis of nonranked voting systems", SIAM Journal on Applied
Mathematics 35:488-495.
Fishburn, PC. (1979a), "Symmetric and consistent aggregation with dichotomous voting", in: J.-J. Laffont,
ed., Aggregation and Revelation of Preferences (North-Holland, Amsterdam) pp. 201-218.
Fishburn, P.C. (1979b), "Heights of representative systems", Discrete Applied Mathematics 1:181 199.
Fishburn, PC. (1981a), "An analysis of simple voting systems for electing committees", SIAM Journal
on Applied Mathematics 41:499-502.
Fishburn, PC. (1981b), "Inverted orders for monotone scoring rules", Discrete Applied Mathematics
3:27-36.
Fishburn, PC. (1982), "Monotonicity paradoxes in the theory of elections", Discrete Applied Mathematics
4:119-134.
Fishburn, P.C. (1984), "Discrete Mathematics in voting and group choice", SIAM Journal on Algebraic
and Discrete Methods 5:263-275.
Fishburn, P.C. (1987), Interprofile Conditions and Impossibility (Harwood Academic, Chur,
Switzerland).
Fishburn, P.C. (1997), "Acyclic sets of linear orders", Social Choice and Welfare 14:113 124.
Fishburn, P.C., and S.J. Brams (1981a), "Approval voting, Condorcet's principle, and runoff elections",
Public Choice 36:89-114.
Fishburn, P.C., and S.J. Brams (1981b), "Efficacy, power, and equity under approval voting", Public
Choice 37:425-434.
Fishburn, P.C., and S.J. Brams (1981c), "Expected utility and approval voting", Behavioral Science
26:136-142.
Fishburn, P.C., and S.J. Brams (1983), "Paradoxes of preferential voting", Mathematics Magazine
56:207-214.
Ch. 4: Voting Procedures 231

Fishburn, PC., and S.J. Brams (1984), "Manipulability of voting by sincere truncation of preferences",
Public Choice 44:397-410.
Fishburn, PC., and W.V. Gehrlein (1976), "An analysis of simple two-stage voting systems", Behavioral
Science 21:1-12.
Fishburn, P.C., and W.V. Gehrlein (1977), "An analysis of voting procedures with nonranked voting",
Behavioral Science 22:178-185.
Fishburn, P.C., and W.V Gehrlein (1982), "Majority efficiencies for simple voting procedures: summary
and interpretation", Theory and Decision 14:141-153.
Fishburn, PC., and J.D.C. Little (1988), "An experiment in approval voting", Management Science
34:555-568.
Fishburn, PC., W.V Gehrlein and E. Maskin (1979), "Condorcet proportions and Kelly's conjectures",
Discrete Applied Mathematics 1:229-252.
Fishburn, P.C., J.E Fishburn and A. Hagy (1992), "Are there better ways to elect bishops?", Circuit
Rider, July-August: 11-12.
Galton, E (1907), "One vote, one value", Nature 75:414.
Gehrlein, W.V (1981), "Single stage election procedures for large electorates", Journal of Mathematical
Economics 8:263-275.
Gehrlein, W.V. (1982), "Condorcet efficiency and constant scoring rules", Mathematical Social Sciences
2:123-130.
Gehrlein, W.V (1983), "Condorcet's paradox", Theory and Decision 15:161-197.
Gehrlein, W.V (1985), "The Condorcet criterion and committee selection", Mathematical Social Sciences
10:199-209.
Gehrlein, W.V. (1993), "Condorcet efficiency of two stage constant scoring rules", Quality and Quantity
27:95-101.
Gehrlein, W.V. (1995), "Condorcet efficiency and social homogeneity", in: C.R. Barnett, H. Moulin,
M. Salles and N. Schofield, eds., Social Choice, Welfare and Ethics (Cambridge University Press,
Cambridge) pp. 127-143.
Gehrlein, W.V (1997), "Condorcet's paradox and the Condorcet efficiency of voting rules", Mathematica
Japonica 45:173-199.
Gehrlein, WV (1999), "Approximating the probability that a Condorcet winner exists", Proceedings of
the International Decision Sciences Institute Meeting, Athens, Greece, pp. 626 628.
Gehrlein, W.V, and P.C. Fishburn (1976), "The probability of the paradox of voting: a computable
solution", Journal of Economic Theory 13:14-25.
Gehrlein, W.V, and P.C. Fishburn (1978a), "Probabilities of election outcomes for large electorates",
Journal of Economic Theory 19:38-49.
Gehrlein, W.V, and P.C. Fishburn (1978b), "Coincidence probabilities for simple majority and positional
voting rules", Social Science Research 7:272-283.
Gehrlein, W.V, and P.C. Fishburn (1979), "Proportions of profiles with a majority candidate", Computers
and Mathematics with Applications 5:117-124.
Gehrlein, W.V, and PC. Fishburn (1980), "Robustness of positional scoring over subsets of alternatives",
Applied Mathematics and Optimization 6:241-255.
Gehrlein, W.V, B. Gopinath, J.C. Lagarias and P.C. Fishburn (1982), "Optimal pairs of score vectors for
positional scoring rules", Applied Mathematics and Optimization 8:309-324.
Gibbard, A.E (1973), "Manipulation of voting schemes: a general result", Econometrica 41:587-601.
Glasser, G.J. (1959), "Game theory and cumulative voting for corporate directors", Management Science
5:151-156.
Good, I.J. (1971), "A note on Condorcet sets", Public Choice 10:97-101.
Goodman, L.A. (1954), "On methods of amalgamation", in: R.M. Thrall, C.H. Coombs and R.L. Davis,
eds., Decision Processes (Wiley, New York) pp. 39-48.
Guilbaud, G.Th. (1952), "Les theories de l'int6rt general et la problem logique de l'agr6gation",
Economic Appliquee 5:501-584.
232 S.J. Brains and PC Fishburn

Guinier, L. (1994), The Tyranny of the Majority: Fundamental Fairness in Representative Democracy
(Free Press, New York).
Hansson, B., and H. Sahlquist (1976), "A proof technique for social choice with variable electorate",
Journal of Economic Theory 13:193 200.
Hare, T. (1861), The Election of Representatives, Parliamentary and Municipal: A Treatise (Longman,
Green, London).
Henriet, D. (1985), "The Copeland choice function: an axiomatic characterization", Social Choice and
Welfare 2:49 63.
Hoag, C.G., and G.H. Hallett (1926), Proportional Representation (Macmillian, New York).
Hoffman, D.T. (1982), "A model of strategic voting", SIAM Journal on Applied Mathematics 42:
751-761.
Hoffman, D.T. (1983), "Relative efficiency of voting systems", SIAM Journal on Applied Mathematics
43:1213-1219.
Holzman, R. (1988-1989), "To vote or not to vote: What is the quota?", Discrete Applied Mathematics
22:133-141.
Inada, K. (1964), "A note on the simple majority decision rule", Econometrica 32:525-531.
Keiding, H. (1984), "Heights of representative systems: a proof of Fishburn's conjecture", Social Choice
and Welfare 1:149-158.
Kelly, J.S. (1974), "Voting anomalies, the number of voters, and the number of alternatives", Econometrica
42:239-251.
Kelly, J.S. (1978), Arrow Impossibility Theorems (Academic, New York).
Kelly, J.S. (1987), Social Choice Theory: An Introduction (Springer, New York).
Kemeny, J. (1959), "Mathematics without numbers", Daedalus 88:577-591.
Kim, K.H., and EW. Roush (1980), Introduction to Mathematical Consensus Theory (Marcel Decker,
New York).
Laffond, G., and J.-F Laslier (1991), "Slater's winners of a tournament may not be in the Banks set",
Social Choice and Welfare 8:365-369.
Lake, M. (1979), "A new campaign resource allocation model", in: S.J. Brams, A. Schotter and
G. Schw6diauer, eds., Applied Game Theory: Proceedings of a Conference at the Institute for
Advanced Studies, Vienna, June 13 16, 1978 (Physica-Verlag, Wiirzburg) pp. 118 132.
Laslier, J.-E (1997), Tournament Solutions and Majority Voting (Springer, Berlin).
Le Breton, M., and M. Truchon (1997), "A Borda measure for social choice functions", Mathematical
Social Sciences 34:249-272.
Levin, J., and B. Nalebuff (1995), "An introduction to vote-counting schemes", Journal of Economic
Perspectives 9:3-26.
Lines, M. (1986), "Approval voting and strategy analysis: a Venetian example", Theory and Decision
20:155 172.
Ludwig, W.G. (1978), "Strategic voting and the Borda method", Public Choice 33:85-90.
May, K.O. (1952), "A set of independent necessary and sufficient conditions for simple majority decision",
Econometrica 20:680-684.
May, R.M. (1971), "Some mathematical results on the paradox of voting", Behavioral Science 16:
143-151.
McGarvey, D.C. (1953), "A theorem on the construction of voting paradoxes", Econometrica
21:608-610.
McLean, I. (1996), "E.J. Nanson, social choice and electoral reform", Australian Journal of Political
Science 31:369-385.
McLean, I., and A.B. Urken, eds. (1995), Classics of Social Choice (University of Michigan Press, Ann
Arbor).
Merrill, S. (1979), "Approval voting: a 'best buy' method for multicandidate elections", Mathematics
Magazine 52:98-102.
Ch. 4: Voting Procedures 233

Merrill, S. (1981), "Strategic decision under one-stage multicandidate voting systems", Public Choice
36:115-134.
Merrill, S. (1982), "Strategic voting in multicandidate elections under uncertainty and under risk", in:
M. Holler, ed., Power, Voting, and Voting Power (Physica-Verlag, Wiirzburg) pp. 179-187.
Merrill III, S. (1984), "A comparison of efficiency of multicandidate electoral systems", American
Journal of Political Science 28:23-48.
Merrill III, S. (1985), "A statistical model for Condorcet efficiency using simulation under spatial model
assumptions", Public Choice 47:389-403.
Merrill III, S. (1988), Making Multicandidate Elections More Democratic (Princeton University Press,
Princeton, NJ).
Merrill III, S., and J. Nagel (1987), "The effect of approval balloting on strategic voting under alternative
decision rules", American Political Science Review 81:509-524.
Mill, J.S. (1862), Considerations on Representative Government (Harper, New York).
Miller, N. (1980), "A new solution set for tournaments and majority voting: further graph-theoretical
approaches to the theory of voting", American Journal of Political Science 24:68-96.
Miller, N.R. (1995), Committees, Agendas, and Voting (Harwood Academic, Chur).
Miller, N.R., B. Grofman and S.L. Feld (1990), "The structure of the Banks set", Public Choice
66:243-251.
Monroe, B.L. (1995), "Fully proportional representation", American Political Science Review
89:925-940.
Moulin, H. (1980), "On strategy-proofness and single peakedness", Public Choice 35:437 455.
Moulin, H. (1988a), Axioms of Cooperative Decision Making (Cambridge University Press, Cambridge).
Moulin, H. (1988b), "Condorcet's principle implies the no show paradox", Journal of Economic Theory
45:53-64.
Moulin, H. (1994), "Social choice", in: R.J. Aumann and S. Hart, eds., Handbook of Game Theory,
Vol. 2 (North-Holland, Amsterdam) pp. 1091-1125.
Murakami, Y. (1966), "Formal structure of majority decision", Econometrica 34:709-718.
Murakami, Y. (1968), Logic and Social Choice (Routledge and Kegan Paul, London).
Myerson, R.B. (1995a), "Analysis of democratic institutions: structure, conduct and performance", Journal
of Economic Perspectives 9:77-90.
Myerson, R.B. (1995b), "Axiomatic derivation of scoring rules without the ordering assumption", Social
Choice and Welfare 12:59-74.
Nagel, J. (1987), "The approval ballot as a possible component of electoral reform in New Zealand",
Political Science 39:70-79.
Nanson, E.J. (1883), "Methods of elections", Transactions and Proceedings of the Royal Society of
Victoria 19:197-240.
Nanson, E.J. (1907), Methods of Elections (British Government Blue Book Miscellaneous No. 3).
Niemi, R.G. (1984), "The problem of strategic voting under approval voting", American Political Science
Review 78:952 958.
Niemi, R.G., and WH. Riker (1976), "The choice of voting systems", Scientific American 234:21-27.
Niemi, R.G., and H.E Weisberg (1968), "A mathematical solution to the problem of the paradox of
voting", Behavioral Science 13:317-323.
Nitzan, S. (1985), "The vulnerability of point-voting schemes to preference variation and strategic
manipulation", Public Choice 47:349-370.
Nitzan, S., and A. Rubinstein (1981), "A further characterization of Borda ranking method", Public
Choice 36:153-158.
Nurmi, H. (1986), "Mathematical models of elections and their relevance for institutional design",
Electoral Studies 5:167-182.
Nurmi, H. (1987), Comparing Voting Systems (Reidel, Dordrecht).
Nurmi, H. (1988), "Discrepancies in the outcomes resulting from different voting schemes", Theory and
Decision 25:193-208.
234 S.J Bratns and PC. Fishburn

Nurmi, H. (1989), "On Nanson's method", in: J. Paastela, ed., Democracy in the Modern World: Essays
for Tatu Vanhanen, Acta Univ. Tamperensis A 260:199-210.
Nurmi, H. (1998a), Rational Behavior and the Design of Institutions: Concepts, Theories and Models
(Edward Elgar, Cheltenham).
Nurmi, H. (1998b), "Voting paradoxes and referenda", Social Choice and Welfare 15:333-350.
Nurmi, H. (1999), Voting Paradoxes and How to Deal with Them (Springer, Berlin).
Nurmi, H., and Y. Uusi-Heikkili (1985), "Computer simulations of approval and plurality voting: the
frequency of weak Pareto violations and Condorcet loser choices in impartial cultures", European
Journal of Political Economy 2(1):47-59.
Peleg, B. (1984), Game-Theoretical Analysis of Voting in Committees (Cambridge University Press,
Cambridge).
Potthoff, R.E, and S.J. Brams (1998), "Proportional representation: broadening the options", Journal of
Theoretical Politics 10:147-178.
Rapoport, A., and D.S. Felsenthal (1990), "Efficacy in small electorates under plurality and approval
voting", Public Choice 64:57-71.
Ray, D. (1986), "On the practical possibility of a 'no show paradox' under the single transferable vote",
Mathematical Social Sciences 11:183-189.
Raynaud, H. (1982), "The individual freedom allowed by the value restriction conditions", Tech. report 360
(Center for Research in Organizational Efficiency, Stanford University).
Raz, R. (2000), "VC-dimension of sets of permutations", Combinatorica 20:1-15.
Regenwetter, M., and B. Grofman (1998), "Approval voting, Borda winners, and Condorcet winners:
evidence from seven elections", Management Science 44:520 533.
Richelson, J.T. (1978), "A characterization result for the plurality rule", Journal of Economic Theory
19:548-550.
Richelson, J.T. (1979), "A comparative analysis of social choice functions I, II, 111: a summary",
Behavioral Science 24:355.
Riker, W.H. (1958), "The paradox of voting and Congressional rules for voting on amendments",
American Political Science Review 52:349-366.
Riker, W.H. (1982), Liberalism Against Populism: A Confrontation Between the Theory of Democracy
and the Theory of Social Choice (Freeman, San Francisco).
Riker, W.H. (1986), The Art of Political Manipulation (Yale University Press, New Haven).
Riker, W.H. (1996), The Strategy of Rhetoric: Campaigning for the American Constitution (Yale
University Press, New Haven).
Roberts, ES. (1991), "Characterization of the plurality function", Mathematical Social Sciences 21:
101-127.
Saari, D.G. (1982), "Inconsistencies of weighted summation voting systems", Mathematics of Operations
Research 7:479-490.
Saari, D.G. (1984), "The ultimate of chaos resulting from weighted voting systems", Advances in Applied
Mathematics 5:286-308.
Saari, D.G. (1987), "The source of some paradoxes from social choice and probability", Journal of
Economic Theory 41:1-22.
Saari, D.G. (1989), "A dictionary for voting paradoxes", Journal of Economic Theory 48:443-475.
Saari, D.G. (1990a), "Susceptibility to manipulation", Public Choice 64:21-41.
Saari, D.G. (1990b), "The Borda dictionary", Social Choice and Welfare 7:279-317.
Saari, D.G. (1992), "Millions of election rankings from a single profile", Social Choice and Welfare
9:277-306.
Saari, D.G. (1994), Geometry of Voting (Springer, New York).
Saari, D.G. (1995a), Basic Geometry of Voting (Springer, New York).
Saari, D.G. (1995b), "A chaotic exploration of voting paradoxes", SIAM Review 37:37-52.
Saari, D.G. (2000a), "Mathematical structure of voting paradoxes I: pairwise vote", Economic Theory
15:1-53.
Ch. 4: Voting Procedures 235

Saari, D.G. (2000b), "Mathematical structure of voting paradoxes II: positional voting", Economic Theory
15:55-101.
Saari, D.G. (2001a), "Analyzing a nail-biting election", Social Choice and Welfare 18:415 430.
Saari, D.G. (2001b), Chaotic Elections! A Mathematician Looks at Voting (American Mathematical
Society, Providence).
Saari, D.G., and J. Van Newenhizen (1988a), "The problem of indeterminacy in approval, multiple, and
truncated voting systems", Public Choice 59:101 120.
Saari, D.G., and J. Van Newenhizen (1988b), "Is approval voting an 'unmitigated evil'?: a response to
Brams, Fishburn, and Merrill", Public Choice 59:133-147.
Satterthwaite, M.A. (1975), "Strategy-proofness and Arrow's conditions: existence and correspondence
theorems for voting procedures and social welfare functions", Journal of Economic Theory 10:
187-218.
Scarsini, M. (1998), "A strong paradox of multiple elections", Social Choice and Welfare 15:237 238.
Schwartz, T. (1972), "Rationality and the myth of the maximum", Nofs 6:97-117.
Schwartz, T. (1974), "Serial collective choice", Mimeograph (Carnegie-Mellon University).
Schwartz, T. (1986), The Logic of Collective Choice (Columbia University Press, New York).
Sen, A.K., and PK. Pattanaik (1969), "Necessary and sufficient conditions for rational choice under
majority decision", Journal of Economic Theory 1:178-202.
Sertel, M.R. (1978), "Characterizing approval voting", Journal of Economic Theory 45:207-211.
Sertel, M.R., and B. Yilmaz (1999), "The majoritarian compromise is majoritarian-optimal and subgame-
perfect implementable", Social Choice and Welfare 16:615-627.
Shapley, L.S., and M. Shubik (1954), "A method for evaluating the distribution of power in a committee
system", American Political Science Review 48:787-792.
Shepsle, K.A., and B.R. Weingast (1984), "Uncovered sets and sophisticated voting outcomes with
implications for agenda control", American Journal of Political Science 28:49-74.
Slater, P. (1961), "Inconsistencies in a schedule of paired comparisons", Biometrica 48:303-312.
Smith, J.H. (1973), "Aggregation of preferences with variable electorate", Econometrica 41:1027-1041.
Snyder, J.M. (1989), "Election goals and the allocation of political resources", Econometrica
57:637-660.
Spafford, D. (1980), "Book review", Canadian Journal of Political Science 11:392-393.
Staring, M. (1986), "Two paradoxes of committee elections", Mathematics Magazine 59:158-159.
Stavely, E.S. (1972), Greek and Roman Voting and Elections (Cornell University Press, Ithaca, NY).
Steams, R. (1959), "The voting problem", American Mathematical Monthly 66:761-763.
Straffin Jr, PD. (1980), Topics in the Theory of Voting (Birkhiiuser, Boston).
Tabarrok, A. (2001), "President Perot or fundamentals of voting theory illustrated with the 1992 election",
Public Choice 106:275 297.
Tabarrok, A., and L. Spector (1999), "Would the Borda count have avoided the Civil War?", Journal of
Theoretical Politics 11:261-288.
Tataru, M., and V Merlin (1997), "On the relationship of the Condorcet winner and positional voting
rules", Mathematical Social Sciences 34:81-90.
Tideman, T.N. (1987), "Independence of clones as a criterion for voting rules", Social Choice and
Welfare 4:185-206.
Tideman, T.N. (1995), "The single transferable vote", Journal of Economic Perspectives 9:27-38.
Tullock, G. (1967), Toward a Mathematics of Politics (University of Michigan Press, Ann Arbor, MI).
Van Deemen, A. (1993), "Paradoxes of voting in list systems of proportional representation", Electoral
Studies 12:234-241.
Van Newenhizen, J. (1992), "The Borda method is most likely to respect the Condorcet principle",
Economic Theory 2:69-83.
Ward, B. (1965), "Majority voting and alternative forms of public enterprise", in: J. Margolis, ed., The
Public Economy of Urban Communities (Johns Hopkins University Press, Baltimore, MD) pp. 112-
126.
236 S.J Brains and PC Fishburn

Weber, R.J. (1977), "Comparison of voting systems", Discussion Paper 498A (Cowles Foundation, Yale
University).
Weber, R.J. (1995), "Approval voting", Journal of Economic Perspectives 9:37-49.
Young, H.P. (1974), "An axiomatization of Borda's rule", Journal of Economic Theory 9:43-52.
Young, H.P (1975a), "Social choice scoring functions", SIAM Journal on Applied Mathematics 28:
824-838.
Young, H.P. (1975b), "Extending Condorcet's rule", Mimeograph (The City University of New York).
Young, H.P. (1988), "Condorcet's theory of voting", American Political Science Review 82:1231-1244.
Young, H.P. (1995), "Optimal voting rules", Journal of Economic Perspectives 9:51-64.
Young, H.P, and A. Levenglick (1978), "A consistent extension of Condorcet's election principle", SIAM
Journal on Applied Mathematics 35:285--300.
Zwicker, W.S. (1991), "The voters' paradox, spin, and the Borda count", Mathematical Social Sciences
22:187-227.
Chapter 5

IMPLEMENTATION THEORY*
ERIC MASKIN
Institute for Advanced Study, Princeton, NJ, USA

TOMAS SJOSTROM
Department of Economics, Pennsylvania State University, University Park, PA, USA

Contents

Abstract 238
Keywords 238
1. Introduction 239
2. Definitions 245
3. Nash implementation 247
3.1. Definitions 248
3.2. Monotonicity and no veto power 248
3.3. Necessary and sufficient conditions 250
3.4. Weak implementation 254
3.5. Strategy-proofness and rich domains of preferences 254
3.6. Unrestricted domain of strict preferences 256
3.7. Economic environments 257
3.8. Two agent implementation 259
4. Implementation with complete information: further topics 260
4.1. Refinements of Nash equilibrium 260
4.2. Virtual implementation 264
4.3. Mixed strategies 265
4.4. Extensive form mechanisms 267
4.5. Renegotiation 269
4.6. The planner as a player 275
5. Bayesian implementation 276
5.1. Definitions 276
5.2. Closure 277
5.3. Incentive compatibility 278
5.4. Bayesian monotonicity 279

* We are grateful to Sandeep Baliga, Luis Corch6n, Matt Jackson, Byungchae Rhee, Ariel Rubinstein,
Ilya Segal, Hannu Vartiainen, Masahiro Watabe, and two referees, for helpful comments.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K. Sen and K. Suzumura
( 2002 Elsevier Science B. V All rights reserved
238 E. Maskin and T: Sj'str6m

5.5. Non-parametric, robust and fault tolerant implementation 281

6. Concluding remarks 281
References 282

Abstract

The implementation problem is the problem of designing a mechanism (game form)

such that the equilibrium outcomes satisfy a criterion of social optimality embodied in
a social choice rule. If a mechanism has the property that, in each possible state of the
world, the set of equilibrium outcomes equals the set of optimal outcomes identified
by the social choice rule, then the social choice rule is said to be implemented by
this mechanism. Whether or not a social choice rule is implementable may depend on
which game-theoretic solution concept is used. The most demanding requirement is
that each agent should always have a dominant strategy, but mainly negative results are
obtained in this case. More positive results are obtained using less demanding solution
concepts such as Nash equilibrium. Any Nash-implementable social choice rule must
satisfy a condition of "monotonicity". Conversely, any social choice rule which
satisfies monotonicity and "no veto power" can be Nash-implemented. Even non-
monotonic social choice rules can be implemented using Nash equilibrium refinements.
The implementation problem can be made more challenging by imposing additional
requirements on the mechanisms, such as robustness to renegotiation and collusion. If
the agents are incompletely informed about the state of the world, then the concept of
Nash equilibrium is replaced by Bayesian Nash equilibrium. Incentive compatibility
is a necessary condition for Bayesian Nash implementation, but in other respects the
results closely mimic those that obtain with complete information.

Keywords

social choice, implementation, mechanism design

JEL classification: D71

Ch. 5. Implementation Theory 239

1. Introduction

The problem of social decision making when information is decentralized has occupied
economists since the days of Adam Smith. An influential article by Hayek crystallized
the problem. Since "the data from which the economic calculus starts are never for
the whole society given to a single mind", the problem to be solved is "how to secure
the best use of resources known to any of the members of society, for ends whose
relative importance only these individuals know" [Hayek (1945)]. A resource allocation
mechanism is thus essentially a system for communicating and processing information.
A mathematical analysis of these issues became possible after the contributions of
Leo Hurwicz. Hurwicz (1960, 1972) provided a formal definition of a resource
allocation mechanism that is so general that almost any conceivable method for making
social decisions is a possible mechanism in this framework. Hurwicz (1972) also
introduced the fundamental notion of incentive compatibility.
The theory of mechanism design provides an analytical framework for the design
of institutions, with emphasis on the problem of incentives . A mechanism, or game
form, is thought of as specifying the rules of a game. The players are the members
of the society (the agents). The question is whether the equilibrium outcomes will
be, in some sense, socially optimal. Formally, the problem is formulated in terms
of the implementation of social choice rules. A social choice rule specifies, for each
possible state of the world, which outcomes would be socially optimal in that state.
It can be thought of as embodying the welfare judgements of a social planner. Since
the planner does not know the true state of the world, she must rely on the agents'
equilibrium actions to indirectly cause the socially optimal outcome to come about.
If a mechanism has the property that, in each possible state of the world, the set
of equilibrium outcomes equals the set of socially optimal outcomes identified by
the social choice rule, then the social choice rule is said to be implemented by this
mechanism. By definition, implementation is easier to accomplish the smaller is the
set of possible states of the world. For example, if the social planner knows that each
agent's true utility function belongs to the class of quasi-linear utility functions, then
her task is likely to be simpler than if she had no such prior information.
To be specific, consider two kinds of decision problems a society may face. The first
is the economic problem of producing and allocating private and/or public goods. Here,
a state of the world specifies the preferences, endowments, and productive technology
of each economic agent (normally, certain a priori restrictions are imposed on the
preferences, e.g., non-satiation). For economies with only private goods, traditional
economic theory has illuminated the properties of the competitive price system. In our
terminology, the Walrasian rule is the social choice rule that assigns to each state of
the world the corresponding set of competitive (Walrasian) allocations. A mechanism

1 Other surveys that cover much of the material we discuss here include Maskin (1985), Groves and
Ledyard (1987), Moore (1992), Palfrey (1992, 2001), Corch6n (1996) and Jackson (2001).
240 E. Maskin and I Sj6strom

might involve agents announcing prices and quantities, or perhaps only quantities (the
appropriate prices could be calculated by a computer). To solve the implementation
problem we need to verify that the set of equilibrium outcomes of the mechanism
coincides with the set of Walrasian allocations in each possible state of the world. In
public goods economies, we may instead be interested in implementing the Lindahl
rule, i.e., the social choice rule that assigns to each state of the world its corresponding
set of Lindahl allocations (these are the competitive equilibrium allocations in the
fictitious price system where each consumer has a personalized price for each
public good). Of course, the Walrasian and Lindahl rules are only two examples of
social choice rules in economic environments. More generally, implementation theory
characterizes the full class of implementable social choice rules.
A second example of a social decision problem is the problem of choosing one
alternative from a finite set (e.g., selecting a president from a set of candidates). In
this environment, a social choice rule is often called a voting rule. No restrictions
are necessarily imposed on how the voters may rank the alternatives. When the
feasible set consists of only two alternatives, then a natural voting rule is the ordinary
method of majority rule. But with three or more alternatives, there are many plausible
voting rules, such as Borda's rule 2 and other rank-order voting schemes. Again,
implementation theory characterizes the set of implementable voting rules.
Whether or not a social choice rule is implementable may depend on which game
theoretic solution concept is invoked. The most demanding requirement is that each
agent should have a dominant strategy. A mechanism with this property is called a
dominant strategy mechanism. By definition, a dominant strategy is optimal for the
agent regardless of the actions of others. Thus, in a dominant strategy mechanism
agents need not form any conjecture about the behavior of others in order to know
what to do. The revelation principle, first stated by Gibbard (1973), implies that
there is a sense in which the search for dominant strategy mechanisms may be
restricted to "revelation mechanisms" in which each agent simply reports his own
personal characteristics (preferences, endowments, productive capacity ... ) to the
social planner. The planner uses this information to compute the state of the world
and then chooses the outcome that the social choice rule prescribes in this state. (To
avoid the difficulties caused by tie-breaking, assume the social choice rule is single-
valued.) Of course, the chosen outcome is unlikely to be socially optimal if agents
misrepresent their characteristics. A social choice rule is dominant strategy incentive
compatible, or strategy-proof, if the associated revelation mechanism has the property
that honestly reporting the truth is always a dominant strategy for each agent.
Unfortunately, in many environments no satisfactory strategy-proof social choice
rules exist. For the classical private goods economy, Hurwicz (1972) proved that no

2 If there are m alternatives, then Borda's rule assigns each alternative m points for every agent who
ranks it first, m - 1 points for every agent who ranks it second, etc.; the winner is the alternative with
the biggest point total.
Ch. 5 Implementation Theory 241

Pareto optimal and individually rational social choice rule can be strategy-proof if the
space of admissible preferences is large enough 3. An analogous result was obtained for
the classical public goods economy by Ledyard and Roberts (1974). It follows from
these results that neither the Walrasian rule nor the Lindahl rule is strategy-proof.
These results confirmed the suspicions of many economists. In particular, Vickrey
(1961) conjectured that if an agent was not negligibly small compared to the whole
economy, then any attempt to allocate divisible private goods in a Pareto optimal way
would imply "a direct incentive for misrepresentation of the marginal-cost or marginal-
value curves". Samuelson (1954) argued that no resource allocation mechanism could
generate a Pareto optimal level of public goods because "it is in the selfish interest of
each person to givefalse signals, to pretend to have less interest in a given collective
activity than he really has, etc" 4.
If only quasi-linear utility functions are admissible (utility functions are additively
separable between the public decision and money and linear in money), then there
does exist an attractive class of mechanisms, the Vickrey-Groves-Clarke mechanisms,
with the property that truth-telling is a dominant strategy [Vickrey (1961), Groves
(1970), Clarke (1971)]. But a Vickrey-Groves-Clarke mechanism will in general fail
to balance the budget (the monetary transfers employed to induce truthful revelation
do not sum to zero), and so Vickrey's and Samuelson's pessimistic conjectures were
formally correct even in the quasi-linear case [Green and Laffont (1979), Walker
(1980), Hurwicz and Walker (1990)]5.
The search for dominant strategy mechanisms in the case of voting over a finite set
of alternatives turned up even more negative results. Gibbard (1973) and Satterthwaite
(1975) showed that if the range of a strategy-proof voting rule contains at least three
alternatives then it must be dictatorial, assuming the set of admissible preferences
contains all strict orderings. Again, this impossibility result confirmed the suspicions
of many economists, notably Arrow (1963), Vickrey (1960) and Dummett and
Farquharson (1961). It follows that the Borda rule, for example, is not strategy-proof.
In fact, Borda himself knew that his scheme was vulnerable to insincere voting and
had intended it to be used only by "honest men" [Black (1958)].
If we drop the requirement that each agent should have a dominant strategy then
the situation is much less bleak. The idea of Nash equilibrium is fundamental to much
of economic theory. In a Nash equilibrium, each agent's action is a best response to
the actions that he predicts other agents will take, and in addition these predictions
are correct. Formal justifications of this concept usually rely on each agent having
complete information about the state of the world. If agents have complete information

3 Hurwicz's (1972) definition of incentive compatibility was essentially a requirement that truthful
reports should be a Nash equilibrium in a game where each agent reports his own personal characteristics
(at a minimum, an agent's "personal characteristics" determine his preferences). This implies that truth-
telling is a dominant strategy.
4 An early discussion of the incentives to manipulate the Lindahl rule can be found in Bowen (1943).
5 But see Groves and Loeb (1975) for a special quadratic case where budget balance is possible.
242 E. Maskin and T Sj6strdm

in this sense, then the planner can ask each agent to report the complete state of
the world, not just his own characteristics 6 . With at least three agents, and with the
planner disregarding a single dissenting opinion against a consensus, it is a Nash
equilibrium for all agents to announce the state truthfully (each agent is using a best
response because he cannot change the outcome by deviating unilaterally). However,
this kind of revelation mechanism would also have many non-truthful Nash equilibria.
This highlights a general difficulty with the revelation principle: although incentive
compatibility guarantees that truth-telling is an equilibrium, it does not guarantee that
it is the only equilibrium. The implementation literature normally requires that all
equilibrium outcomes should be socially optimal (an exception is the dominant-strategy
literature, where the possibility of multiple equilibria, i.e., multiple dominant strategies,
is typically much less worrisome).
Nash implementation using mechanisms with general message spaces was first
studied by Groves and Ledyard (1977), Hurwicz and Schmeidler (1978) and Maskin
(1999) 7. For a class of economic environments, Groves and Ledyard (1977) discovered
that non-dictatorial mechanisms exist such that all Nash equilibrium outcomes are
Pareto optimal. Hurwicz and Schmeidler (1978) found a similar result for the case of
social choice from a finite set of alternatives. General results applicable to both kinds
of environments were obtained by Maskin (1999). He found that a "monotonicity"
condition is necessary for a social choice rule to be Nash-implementable. With at
least three agents, monotonicity plus a condition of "no veto power" is sufficient.
The monotonicity condition says that if a socially optimal alternative does not fall
in any agent's preference ordering relative to any other alternative, then it remains
socially optimal. In economic environments, the Walrasian and Lindahl rules satisfy
monotonicity (strictly speaking, the Walrasian and Lindahl rules have to be modified
slightly to render them monotonic). Since no veto power is always satisfied in economic
environments with three or more non-satiated agents, these social choice rules can be
Nash-implemented. In the case of voting with a finite set of alternatives, a monotonic
single-valued social choice rule must be dictatorial if the preference domain consists of
all strict orderings, and there are (at least) three different alternatives such that for each
of them there is a state where that alternative is socially optimal. However, the (weak)
Pareto correspondence is a monotonic social choice correspondence that satisfies no
veto power in any environment, and hence it can be Nash-implemented.

6 Such a mechanism requires transmission of an enormous amount of information to the social planner.
In practice, this may be costly and time-consuming. However, in this survey we do not focus on the issue
of informational efficiency, but rather on characterization of the set of implementable social choice rules.
The mechanisms are not intended to be "realistic", and in applications one would look for much simpler
mechanisms. It is worth noticing that in Hurwicz's (1960) original "decentralized mechanism", messages
were simply sets of net trade vectors. Important theorems concerning the informational efficiency of
price mechanisms were established by Mount and Reiter (1974) and Hurwicz (1977).
7 Maskin's article was circulated as a working paper in 1977.
Ch. 5: Implementation Theory 243

If agent i's strategy si is a best response against the strategies of others, and the
resulting outcome is a, then si remains a best response if outcome a moves up in
agent i's preference ordering. Thus, such a change in agent i's preferences cannot
destroy a Nash equilibrium (which is why monotonicity is a necessary condition
for Nash implementation). However, it can make si a weakly dominated strategy for
agent i, and so can destroy an undominated Nash equilibrium (i.e., a Nash equilibrium
where each agent is using a weakly undominated strategy). Hence monotonicity is not
a necessary condition for implementation in undominated Nash equilibria.
This insight was exploited by Palfrey and Srivastava (1991), who found that many
more social choice rules can be implemented in undominated Nash equilibria than in
Nash equilibria. A similar result was found by Sj6str6m (1993) for implementation in
trembling-hand perfect Nash equilibria8. Moreover, rather different paths can lead to
the implementation of non-monotonic social choice rules. Moore and Repullo (1988)
showed that the set of implementable social choice rules can be dramatically expanded
by the use of extensive game forms. This development was preceded by the work by
Farquharson (1969) and Moulin (1979) on sequential voting mechanisms. Abreu and
Sen (1991) and Matsushima (1988) considered "virtual" implementation, where the
socially optimal outcome is required to occur only with probability close to one, and
found that the set of virtually implementable social choice rules is also very large.
Despite this plethora of positive results, it would not be correct to say that any
social choice rule can be implemented by a sufficiently clever mechanism together
with a suitable refinement of Nash equilibrium. Specifically, only ordinal social choice
rules can be implemented 9 . This is a significant restriction since many well-known
social welfare criteria depend on cardinal information about preferences (for example,
utilitarianism and various forms of egalitarianism). On the other hand, if there are
at least three agents, then, with suitable equilibrium refinement, not much more than
ordinality is required for implementation 0. The mechanisms that are used to establish
these most general "possibility theorems" sometimes have a questionable feature, viz.,
out-of-equilibrium behavior may lead to highly undesirable outcomes (for example,
worthwhile goods may be destroyed). If the agents can renegotiate such bad outcomes
then such mechanisms no longer work [Maskin and Moore (1999)]. In fact, the

8 Nash equilibrium refinements help implementation by destroying undesirable equilibria, but they also
make it harder to support a socially optimal outcome as an equilibrium outcome. In practice, refinements
seem to help more often than they hurt, but it is not difficult to come up with counter-examples. Sjdstrbm
(1993) gives an example of a social choice rule that is implementable in Nash equilibria but not in
trembling-hand perfect Nash equilibria.
9 An ordinal social choice rule does not rely on cardinal information about the "intensity" of preference.
Thus, if the social choice rule prescribes different outcomes in two different states, then there must exist
some agent i and some outcomes a and b such that agent i's ranking of a versus b is not the same in
the two states (i.e., there is preference reversal).
10 Sometimes the no veto power condition is part of the sufficient condition. Although no veto power
is normally trivially satisfied in economic environments with at least three agents, it is not always an
innocuous condition in other environments.
244 E. Maskin and T Sj6stroni

possibility of renegotiation can make the implementation problem significantly more

difficult when there are only two agents. However, the general "possibility theorems"
seem to survive renegotiation in economic environments with three or more agents
[Sj6str6m (1999)].
Obviously, the social planner cannot freely "choose" a solution concept (such as
undominated Nash equilibrium) to suit his purposes. In some sense, the solution
concept should be appropriate for the mechanism and environment at hand, but it is
hard to make this requirement mathematically precise [for an insightful discussion,
see Jackson (1992)]. Harsanyi and Selten (1988) argue that game theoretic analysis
should lead to an ideal solution concept that applies universally to all possible games,
but experiments show that behavior in practice depends on the nature of the game
(even on "irrelevant" aspects such as the labelling of strategies). How the mechanism
is explained to the agents may be an important part of the design process (e.g.,
"please notice that strategy si is dominated"). Hurwicz (1972) argued in terms of a
dynamic adjustment toward Nash equilibrium: each agent would keep modifying his
strategy according to a fixed "response function" until a Nash equilibrium was reached.
However, Jordan (1986) showed that equilibria of game forms that Nash-implement the
Walrasian rule will in general not be stable under continuous-time strategy-adjustment
processes. Muench and Walker (1984), de Trenqualye (1988) and Cabrales (1999)
also discuss the problem of how agents may come to coordinate on a particular
equilibrium. Cabrales and Ponti (2000) show how evolutionary dynamics may lead to
the "wrong" Nash equilibrium in mechanisms which rely on the elimination of weakly
dominated strategies. Best-response dynamics do converge to the "right" equilibrium in
the particular mechanism they analyze. But these kinds of naive adjustment processes
are difficult to interpret, because behavior is not fully rational along the path: a fully
rational agent would try to exploit the naivete of other agents, especially if he knew
(or could infer something about) their payoff functions. In experiments where a game
is played repeatedly, treatments in which players are uninformed about the payoff
functions of other players appear more likely to end up at a Nash equilibrium (of
the one-shot game) than treatments where players do have this information [Smith
(1979)]. Perhaps it is too difficult to even attempt to manipulate the behavior of an
opponent with an unknown payoff function. It was precisely because he did not want
to assume that agents have complete information that Hurwicz (1972) introduced
the dynamic adjustment processes. But the problem of how agents can learn to play
a Nash equilibrium is difficult [for a good introduction, see Fudenberg and Levine
(1998)].
If we discount the possibility that incompletely informed agents will end up at a
Nash equilibrium, then the results of Maskin (1999) and the literature that followed
him can be interpreted as drawing out the logical implications of the assumption that
agents have complete information about the state of the world. In some cases this
assumption may be reasonable, and many economic models explicitly or implicitly rely
on it. But in other cases it makes more sense to assume that agents assign positive
Ch. 5: Implementation Theory 245

probability to many different states of the world, and behave as Bayesian expected
utility maximizers.
Bayesian mechanism design was pioneered by D'Aspremont and Grard-Varet
(1979), Dasgupta, Hammond and Maskin (1979), Myerson (1979) and Harris and
Townsend (1981). If an agent has private information not shared by other agents,
then a Bayesian incentive compatibility condition is necessary for him to be willing
to reveal it. But not every Bayesian incentive compatible social choice rule is
Bayesian Nash-implementable, because a revelation mechanism may have undesirable
equilibria in addition to the truthful one. Postlewaite and Schmeidler (1986), Palfrey
and Srivastava (1989a) and Jackson (1991) have shown that the results of Maskin
(1999) can be generalized to the Bayesian environment. A Bayesian monotonicity
condition is necessary for Bayesian Nash implementation. With at least three agents,
a condition that combines Bayesian monotonicity with no veto power is sufficient for
implementation, as long as Bayesian incentive compatibility and a necessary condition
called closure are satisfied [Jackson (1991)].
Mechanisms can also be used to represent rights [Giirdenfors (1981), Gaertner,
Pattanaik and Suzumura (1992), Deb (1994), Hammond (1997)]. Deb, Pattanaik and
Razzolini (1997) introduced several properties of mechanisms that correspond to
"acceptable" rights structures. For example, an individual has a say if there exists
at least some circumstance where his actions can influence the outcome . The
notion of rights is important but will not be discussed in this survey. Our notion
of implementation is consequentialist: the precise structure of a mechanism does not
matter as long as its equilibrium outcomes are socially optimal.

2. Definitions

The environment is (A, N, O), where A is the set of feasible alternatives or outcomes,
N = {1,2, .. , n} is the finite set of agents, and O is the set of possible states of
the world. For simplicity, we suppose that the set of feasible alternatives is the same
in all states [see Hurwicz, Maskin and Postlewaite (1995) for implementation with
a state-dependent feasible set]. The agents' preferences do depend on the state of
the world. Each agent i N has a payoff function ui: A x O - R. Thus, if the
outcome is a E A in state of the world 0 E , then agent i's payoff is ui(a, 0).
His weak preference relation in state 0 is denoted Ri = Ri(O), the strict part of his
preference is denoted Pi = Pi(0), and indifference is denoted Ii = Ij(0). That is,
xRiy if and only if ui(x, 0) > ui(y, 0), xPiy if and only if ui(x, 0) > ui(y, 0), and
xiiy if and only if ui(x, 0) = ui(y, 0). The preference profile in state 0 G O is denoted

11 Gaspart (1996, 1997) proposed a stronger notion of equality (or symmetry) of attainable sets: all
agents, by unilaterally varying their actions, should be able to attain identical (or symmetric) sets of
outcomes, at least at equilibrium.
246 E. Maskin and T Sjistrodm

R = R(O) = (Rl(O), ... , R,(0)). The preference domain is the set of preference profiles
that are consistent with some state of the world, i.e., the set

R(O) _ R: there is 0 E O such that R = R(0)}.

The preference domain for agent i is the set

7Ri(0) {Ri: there is R i such that (Ri,R i) RZ(O)}.

When is fixed, we can write R and 7Ri instead of 1Z(O) and Ri(0).
Let ZA be the set of all profiles of complete and transitive preference relations on A,
the unrestricted domain. It will always be true that Z(O) C RA. Let PA be the set of
all profiles of linear orderings of A, the unrestricted domain of strict preferences 12
For any sets X and Y, let X - Y _ {x E X: x Y}, let yX denote the set of all
functions from X to Y, and let 2x denote the set of all subsets of X. If X is finite,
then IXI denotes the number of elements in X.
A social choice rule (SCR) is a function F: (9 2A - {0} (i.e., a non-empty
valued correspondence). The set F(O) C A is the set of socially optimal (or F-optimal)
alternatives in state 0 E . The image or range of the SCR F is the set

F(O) _ {a A: a E F(O) for some 0 C O}.

A social choice function (SCF) is a single-valued SCR, i.e., a functionf: O ) A.

Some important properties of SCRs are as follows.
- Ordinality: for all (0, 0') e O x O, if R(0) = R(O') then F(O) = F(O').
- Weak Pareto optimality: for all 0 E 6 and all a E F(0), there is no b E A such that
ui(b, 0) > ui(a, 0) for all i E N.
Pareto optimality: for all 0 C O and all a E F(0), there is no b E A such that
ui(b, 0) > ui(a, 0) for all i E N with strict inequality for some i.
- Pareto indifference: for all (a, 0) A x O and all b F(0), if ui(a, 0) = ui(b, 0)
for all i N then a E F(O).
- Dictatorship: there exists i C N such that for all 0 E and all a E F(0),
ui(a, 0) > ui(b, 0) for all b C A.
Unanimity: for all (a, 0) C A x ), if ui(a, 0) > ui(b, 0) for all i E N and all b C A
then a F(O).
- Strong unanimity: for all (a, 0) A x , if ui(a, 0) > ui(b, 0) for all i E N and all
b • a then F(O) = {a}.

12 A preference relation Ri is a linear ordering if and only if it is complete, transitive and antisymmetric
(for all (a, b) E A x A, if aRib and bRia then a = b).
Ch. 5: Implementation Theory 247

- No veto power: for all (a,j, 0) C A x N x O, if ui(a, 0) > ui(b, 0) for all b C A and
all i #j then a C F(O).
A mechanism (or game form) is denoted F = (x = 1M i, h) and consists of a message
space Mi for each agent i E N and an outcome function h: x=l Mi A. Let mi E Mi
denote agent i's message. A message profile is denoted m = (ml, ... , m,) E M _
xI= Mi. All messages are sent simultaneously, and the final outcome is h(m) E A. This
kind of mechanism is sometimes called a normal form mechanism (or normal game
form) to distinguish it from extensive form mechanisms in which agents make choices
sequentially [Moore and Repullo (1988)]. With the exception of Section 4.4, nearly all
our results relate to normal form mechanisms, so merely calling them "mechanisms"
should not cause confusion.
The most common interpretation of the implementation problem is that a social
planner or mechanism designer (who cannot observe the true state of the world)
wants to design a mechanism in such a way that in each state of the world the set of
equilibrium outcomes coincides with the set of F-optimal outcomes. Let S equilibrium
be a game theoretic solution concept and let F be an SCR. For each mechanism r
and each state 0 E 0, the solution concept specifies a set of S equilibrium outcomes
denoted S(F, 0) A. A mechanism F implements F in S equilibria, or simply
S-implements F, if and only if S(F, 0) = F(O) for all 0 E . Thus, the set of
S equilibrium outcomes should coincide with the set of F-optimal outcomes in each
state. If such a mechanism exists then F is implementable in S equilibria or simply
S-implementable. This notion is sometimes referred to as full implementation. Clearly,
whether or not an SCR F is S-implementable may depend on the solution concept S.
If solution concept S2 is a refinement of SI, in the sense that for any F we have
S 2 (F,0) C Sl(F, 0) for all 0 cE , then it is not a priori clear whether it will be easier
to satisfy S 1(F, 0) = F(O) or S 2(F, 0) = F(O) for all 0 E O. However, as discussed in
the Introduction, the literature shows that refinements "usually" make things easier.
Most of this survey deals with full implementation in the above sense, but we will
briefly deal with the notions of weak and double implementation. A mechanism F
weakly S-implements F if and only if 0 S(F, 0) C F(O) for all 0 O. That
is, every S equilibrium outcome must be F-optimal, but every F-optimal outcome
need not be an equilibrium outcome. Weak implementation is actually subsumed by
the theory of full implementation, since weak implementation of F is equivalent to
full implementation of a subcorrespondence of F [Thomson (1996)]. If S1 and S2
are two solution concepts, then r doubly S1- and S 2-implements F if and only if
S 1(F, 0) = S 2(F, 0) = F(O) for all 0 E 0.

3. Nash implementation

We start by assuming that the true state of the world is common knowledge among
the agents. This is the case of complete information. We will consider mechanisms in
normal form. (Extensive form mechanisms are discussed in Section 4.4.)
248 E. Mlaskin and 7 Sjidstrtm

3.1. Definitions
Given a mechanism F = (M, h) for any m E M and i E N, let mi = {mj}j i Mi -
xjiMj denote the messages sent by agents other than i. For message profile
m = (m i, mi) e M, the set
h(m i,M) -{a E A: a = h(mi, m') for some m' E Mi}

is agent i's attainable set at m. Agent i's lower contour set at (a, 0) A x is
Li(a, 0) -- b E A: ui(a, 0) > ui(b, 0)}. A message profile m E M is a (pure strategy)
Nash equilibrium at state 0 e O if and only if h(m-i,Mi) C Li(h(m), 0) for all i C N.
(For now we neglect mixed strategies: they are discussed in Section 4.3.) The set of
Nash equilibria at state 0 is denoted Nr(O) C M, and the set of Nash equilibrium
outcomes at state 0 is denoted h(Nr(0)) = {a A: a = h(m) for some m E Nr(O)}.
The mechanism F Nash-implements F if and only if h(Nr(O)) = F(O) for all 0 e 9.

3.2. Monotonicity and no veto power

If Li(a, 0) C Li(a, 0') then we say that Ri(0') is a monotonic transformation of Ri(0)
at alternative a. The SCR F is monotonic if and only if for all (a, 0, 0') G A x O x O
the following is true: if a c F(O) and Li(a, 0) C Li(a, 0') for all i E N, then a e F(O').
Thus, monotonicity requires that if a is optimal in state 0, and when the state changes
from 0 to 0' outcome a does not fall in any agent's preference ordering relative to
any other alternative, then a remains optimal in state '. Clearly, if F is monotonic
then it must be ordinal. But many ordinal social choice rules are not monotonic 13.
Whether a particular SCR is monotonic may depend on the preference domain R(O).
For example, in an exchange economy, the Walrasian correspondence is not monotonic
in general, but it is monotonic on a domain of preferences such that all Walrasian
equilibria occur in the interior of the feasible set [Hurwicz, Maskin and Postlewaite
(1995)]. There is no monotonic and Pareto optimal SCR on the unrestricted domain RA
[Hurwicz and Schmeidler (1978)] 14. However, the weak Pareto correspondence 15 is
monotonic on any domain. A monotonic SCF on RA must be a constant fmction 16, but
there are important examples of monotonic non-constant SCFs on restricted domains.

13 If F is not monotonic then an interesting problem is to find the minimal monotonic extension, i.e.,
the smallest monotonic supercorrespondence of F [Sen (1995), Thomson (1999)].
14 Let 0 E ® be a state where the agents do not unanimously agree on a top-ranked alternative, and let
a E F(O). There must exist j N and b C A such that bPj(O) a. Let state ' be such that preferences
over alternatives in A {b} are as in state 0, but each agent i j has now become indifferent between
a and b. Agentj still strictly prefers b to a in state )' so b Pareto dominates a. But Li(a, 0) C Li(a, 0')
for all i so a E F(O') if F is monotonic, a contradiction of Pareto optimality.
15 The weak Pareto correspondence selects all weakly Pareto optimal outcomes: for all 0 E 9,
F(O) = {a C A: there is no b E A such that u(b, 0) > ui(a, 0) for all i C N}.
16 That is, f(O) = {a} for some a A. For if f(0) = a a' = f(0') then monotonicity implies
{a, a'} C f(O") if a and a' are both top-ranked by all agents in state 0", but this contradicts the fact
thatf is single-valued. See Saijo (1987).
Ch. 5: Implementation Theory 249

Maskin (1999) proved that for any mechanism , the Nash equilibrium outcome
correspondence h o N: 0 - A is monotonic.
Theorem 1: [Maskin (1999)]. If the SCR F is Nash-implementable, then F is
monotonic.
Proof: Suppose F = (M, h) Nash-implements F. Then if a E F(O) there is m Nr()
such that a = h(m). Suppose Li(a, 0) C Li(a, 0') for all i E N. Then, for all i C N,

h(m_i,Mi) C Li(a, 0) C Li(a, 0').

Therefore, m C Nr(O'), and so a C h(Nr(O')) = F(0'). ]

Theorem 1 has a partial converse. It was originally stated by Maskin in 1977, but
without a complete proof [see Maskin (1999)]. Rigorous proofs were given by Williams
(1986), Repullo (1987) and Saijo (1988). Recall that F satisfies no veto power if
an alternative is F-optimal whenever it is top-ranked by at least n - 1 agents. In
economic environments, no veto power is usually vacuously satisfied (because two
different agents will never share the same top-ranked alternative). However, in other
environments no veto power may not be a trivial condition. If, for example, A is
a finite set, 7Z(O) = PA and the number of alternatives is strictly greater than the
number of agents, then even the Borda rule does not satisfy no veto power 17 . If
T¢(O) = RA then no Pareto optimal SCR can satisfy no veto power 18. Still, the weak
Pareto correspondence satisfies no veto power on any domain.
Theorem 2: [Maskin (1999)]. Suppose n > 3. If the SCR F satisfies monotonicity
and no veto power; then F is Nash-implementable.
Proof: The proof is constructive. Let each agent i E N announce an outcome, a state
of the world, and an integer between 1 and n. Thus, M = A x O x { 1,2, ... , n} and a
typical message for agent i is denoted mi = (a', 0', z) E Mi. Let the outcome function
be as follows.
Rule 1: If (a', Oi) = (a, 0) for all i E N and a E F(O), then h(m) = a.
Rule 2: Suppose there exists j E N such that (a', 0i ) = (a, 0) for all i • j but
(aJ, Oi) (a, 0). Then h(m) = a if aJ E Lj(a, 0) and h(m) = a otherwise.
Rule 3: In all other cases, let h(m) = aj forj C N such thatj = (i N zi) (mod n) 19
We need to show that, for any 0* e 0, h(Nr(O*)) = F(O*).
Step 1: h(Nr(O*)) C F(O*). Suppose m E Nr(O*). If either rule 2 or rule 3 applies
to m, then there is j N such that any agent k ; j can get his top-ranked alternative,

17 Suppose agent 1 ranks a first and b last. All other agents rank b first and a second. If Al > n then
b gets a lower Borda score than a and hence is not selected.
18 If Ul (b, 0) > ul (a, 0), and ui(b, 0) = ui(a, ) > u(x, 0) for all i 1 and all x CA - {a, b}, then no
veto power implies a E F(O) even though b Pareto dominates a.
19 a = /3 (mod n) denotes that integers a and 3 are congruent modulo n.
250 E. iMaskin and TISjdstr6nm

via rule 3, by announcing an integer zk such that k = (L z) (mod n). Therefore, we

must have uk(h(m), 0*) > uk(x, 0*) for all k j and all x e A, and hence h(m) e F(8*)
by no veto power. If instead rule 1 applies, then (a', 0i) = (a, 0) for all i E N, and
a e F(). The attainable set for each agentj is L(a, 0) by rule 2. Since m e Nr(o*),
we have Lj(a, 0) c Lj(a, 0*). By monotonicity, a e F(O*). Thus, h(Nr()O*)) C F(0*).
Step 2: F(O*) C h(Nr(o*)). Suppose a F(O*). If mi = (a, 0%, 1) for all i N,
then h(m) = a. By rule 2, h(mj,M) = Lj(a, 0*) for all j e N, so m Nr(0*). Thus,
F(0*) C h(Nr(0*)). []
The mechanism in the proof of Theorem 2 is the canonical mechanism for Nash
implementation. Rule 3 is referred to as a "modulo game".
The canonical mechanism can be simplified in several ways even in this abstract
framework. Since any Nash-implementable F is ordinal, it clearly suffices to let the
agents announce a preference profile R e R(O0) rather than a state of the world 0 e O.
In fact, it suffices if each agent i e N announces a preference ordering for himself
and one for his "neighbor" agent i + 1, where agents 1 and n are considered neighbors
[Saijo (1988)]. Lower contour sets could be announced instead of preference orderings
[McKelvey (1989)]. Much less information is needed when F is the Walrasian rule
[Chakravorty (1991)].
More generally, given any message process that "computes" (or "realizes") an SCR,
Williams (1986) considered the problem of embedding the message process into a
mechanism which Nash-implements the SCR. If the original message process encodes
information in an efficient way, then the same will be true for Williams' mechanism
for Nash implementation.

3.3. Necessary and sufficient conditions

The no veto power condition is not necessary for Nash implementation with n > 3. On
the other hand, monotonicity on its own is not sufficient [see Maskin (1985, 1999) for
a counterexample]. The necessary and sufficient condition was given by Moore and
Repullo (1990). It can be explained by considering how the canonical mechanism of
Section 3.2 must be modified when no veto power is violated.
Suppose we want to Nash-implement a monotonic SCR F using some mechanism
F = (M, h). Let a F(O). There must exist a Nash equilibrium m* Nr(0) such
that h(m*) = a. Agent j's attainable set must satisfy h(m*j, M) C Lj(a, 0). Alternative
c e Lj(a, 0) is an awkward outcome for agentj in Lj(a, 0) if and only if there is 0' e O
such that: (i) Lj(a, 0) C L(c, 0'); (ii) for each i j, Li(c, 0') = A; (iii) c F(O').
Notice that there are no awkward outcomes if F satisfies no veto power, since in that
case (ii) and (iii) cannot both hold. But suppose no veto power is violated and (i),
(ii) and (iii) all hold for 0' so c is awkward in Lj(a, 0). If c e h(m*j,Mj) then there is
mj e M such that h(m*j, mj) = c. Then (m*/, mj) c Nr(O') since (i) implies c is the best
outcome for agentj in his attainable set h(m*, M) in state 0', and (ii) implies c is the
best outcome in all of A for all other agents. By (iii), c F(0'), so h(Nr(0)) F(0'),
Ch. 5: Implementation Theory 251

contradicting the definition of implementation. Thus, the awkward outcome c cannot

be in agent j's attainable set. We must have h(m*, Mj) C Cj(a, 0), where C(a, 0)
denotes the set of outcomes in Lj(a, 0) that are not awkward for agent j in Lj(a, 0).
That is, Cj(a, 0) {c E Lj(a, 0): for all 0' e O, if Lj(a, 0) C L(c, 0') and for
each i : j, Li(c, O') = A, then c F(O')}. But if h(m*,,M,) C Ci(a, 0) for all
i E N, then for any 0' E such that Ci(a, 0) C Li(a, O') for all i C N we will
have m* Nr(6o), so Nash implementation requires a = h(m*) E F(O'). The SCR F
is strongly monotonic if and only if for all (a, 0, 0') E A x x O the following is
true: if a E F(O) and Ci(a, 0) C Li(a, O') for all i N, then a E F(O'). Notice
that strong monotonicity implies monotonicity, and monotonicity plus no veto power
implies strong monotonicity. We have just shown that strong monotonicity is necessary
for Nash implementation. In certain environments, it is also sufficient.
In the canonical mechanism of Section 3.2, if m* is a "consensus" message profile
such that rule 1 applies, i.e., all agents announce (a, 0) with a E F(O), then agent j's
attainable set is Lj(a, 0). We have just seen why this may not work if no veto power is
violated. The obvious solution is to modify rule 2 so that Cj(a, 0) becomes agent j's
attainable set. If n > 3 and any linear ordering of A is an admissible preference relation
('PA C 1Z(O)) then this solution does work and strong monotonicity is sufficient
for Nash implementation. A version of this result appears in Danilov (1992) [see
also Moore (1992)]. It is instructive to prove it by comparing strong monotonicity
to condition M, which is a necessary and (when n > 3) sufficient condition for Nash
implementation in any environment [Sjistr6m (1991)] 20. The definition of condition M
can be obtained from the definition of strong monotonicity by replacing the set Ci(a, 0)
by a set Ci*(a, 0) defined by Sj6str6m (1991). Since Ci*(a, 0) C Ci(a, 6) always holds,
condition M implies strong monotonicity. But if PA C 7Z(O) and F is strongly
monotonic, then Ci*(a, 0) = Ci(a, 0). Thus, if PA C R(O) then strong monotonicity
implies condition M, i.e., the two conditions are equivalent in this case.
There are two ways in which the definition of Ci*(a, 0) differs from the definition
of Ci(a, 0). The first difference is due to the fact that if F does not satisfy unanimity,
then there are alternatives that must never be in the range of the outcome function h.
Alternative a is a problematic outcome if and only if a F(O) for some state 0 such
that Li(a, 0) = A for all i E N. The problematic outcome a would clearly be a non-
F-optimal Nash equilibrium outcome in state 0 if a = h(m) for some m E M. After
removing all problematic outcomes from A (several iterations may be necessary), what
remains is some set B* C A. Since we must have h(m) C B* for all m E M, Sj6strim
(1991) in effect treats B* as the true "feasible set". His analogue of part (ii) of the
definition of "awkward outcome" is therefore: for each i •j, B* C Li(c, 0'). However,
it turns out that this difference is irrelevant if PA C R(O) 21

20 Condition M is equivalent to Moore and Repullo's (1990) condition P. But it is easier to check.
21 Suppose PA C R(O) and let F be strongly monotonic. Let a E F(O), and let Cj(a, 0) be the set
of outcomes in Lj(a, 0) that are not awkward according to the new definition (using B* in (ii)). We
252 E. Maskin and 7: Sfistr6m

The second difference is due to the fact that, after removing the awkward outcomes
from Lj(a, 0), we may discover a second-order awkward outcome c C1(a, 0) such
that for some 0' O: (i) C(a, 0) Lj(c, 0'); (ii) for each i j, Li(c, 0') = A;
(iii) c F(O'). Again, this would contradict implementation, so we must remove
all second-order awkward outcomes from the attainable set, too. Indeed, Sj6str6m's
(1991) algorithm may lead to iterated elimination of even higher-order awkward
outcomes. When there are no more iterations to be made, what remains is the
set Cj*(a, 0) C C(a, 0). It turns out that if PA C R(O) and F is strongly monotonic,
then there are no second-order awkward outcomes: the algorithm terminates after one
step with Cj (a, 0) = Cj(a, 0) 22. In this case, strong monotonicity implies condition M,
which is sufficient for Nash implementation 23 . Thus, if n > 3 and PA C TZ(O) then
the SCR F is Nash-implementable if and only if it is strongly monotonic, as claimed.
Consider two examples due to Maskin (1985). First, suppose N = 1,2,3},
A = {a,b,c} and R(O) = PA. The SCR F is defined as follows. For any 0 e O,
a C F(O) if and only if a majority prefers a to b, and b F(O) if and only if
a majority prefers b to a, and c F(O) if and only if c is top-ranked in A by
all agents. This SCR is monotonic and satisfies unanimity but not no veto power.
Fix j E N and suppose 0 is such that bPj()aPj(0)c, and aPi(0)b for all i X j.
Then F(O) = a}. Now suppose 0' is such that bPj(O')cPj(O')a and Li(c, 0') = A
for all i j. Since L(a, 0) = L(c, O') = {a,c} but c d: F(0'), c is awkward
in Lj(a, 0). Removing c, we obtain Cj(a, 0) = {a}. By the symmetry of a and b,
Cj(b, 0) = {b} whenever aPj(0)bPj(O)c and bPi(0)a for all i j. There are no other
awkward outcomes and it can be verified that F is strongly monotonic, hence Nash-
implementable. For a second example, consider any environment with n > 3, and let
a0 A be a fixed "status quo" alternative. The individually rationalcorrespondence,
defined by F(O) = {a A: aRi(0)ao for all i N}, satisfies monotonicity and
unanimity but not no veto power. If a F(O) then ao E Lj(a, 0) for all j E N.

claim Cj(a, 0) = Ci(a, 0). Clearly, C(a, 0) C Cj(a, 0) since B* C A. Thus, we only need to show
Cj(a, 0) C Cj(a, 0). Suppose c E Lj(a, 0) but c C Cj(a, 0). Then there is 0' such that L(a, 0) C Lj(c, 0')
and B C Li(c, 0') for each i j, and c F(O'). We claim c C(a, 0). Suppose, in order to get a
contradiction, that c E Cj(a, 0). Then, if 0" O is a state where L(a, 0) Lj(c, 0") and Li(c, 0") = A
for each i •j, we have c C F(O"). It is easy to check that strong monotonicity implies Ci(c, 0") C B* for
all i. Thus, Cj(c, 0") C Lj(c, 0") C Lj(c, 0') and Ci(c, 0") C B* C Li(c, 0') for each i j, so c e F(O')
by strong monotonicity. This is a contradiction. Thus, C/(a, 0) C Cj(a, 0).
22 We claim that there are no second-order awkward outcomes if PA C- R(O) and F is strongly
monotonic. Suppose a F(0), c Cj(a, 0) c Lj(c, 0'), and for each i j, Li(c, 0') = A. Since
PA C R(9) there exists 0" C O such that Lj(c, 0") = Lj(a, 0) and Li(c, 0") = A for all i j. Since
c r Cj(a, 0), we have c F(O"). Now, Cj(c, 0") = Cj(a, 0) C Lj(c, 0') and Li(c, 0') = A for all i Xj, so
c e F(O') by strong monotonicity.
23 Actually, since C*(a, 0) is supposed to be agent i's attainable set at a Nash equilibrium nm*such that
h(m*) = a E F(8), Sjdstr6m (1991) explicitly required a E C?(a, 0). Such a requirement is not explicit
in strong monotonicity. But if 'PA C R(O) and F is strongly monotonic then it is easy to check that
a E Ci(a, 0) = C(a, 0).
Ch. 5: Implementation Theory 253

If c E L(a, 0) C Lj(c, 0') and Li(c, 0') = A for each i j, then cRi(O') ao for all
i E N so c C F(O'). Therefore, there are no awkward outcomes, and condition M and
strong monotonicity both reduce to monotonicity. Since F is monotonic, it is Nash-
implementable.
If R(O) = TZA then any monotonic F which satisfies Pareto indifference is strongly
monotonic 24 . This fact is useful because if F is Nash-implementable when 1(O&) = TRA
then implementation is possible (using the same mechanism) when the domain of
preferences is restricted in an arbitrary way. In the context of voting, an even stronger
symmetry condition called neutrality is often imposed. Neutrality requires that the SCR
never discriminates among alternatives based on their labelling. Suppose a F(0)
and c E Lj(a, 0), and state 0' E O is such that Lj(a, 0) C Lj(c, 0') and for each i •j,
Li(c, 0') = A. Let 0" C 6 be a state where preferences are just as in 0' except for
a permutation of alternatives a and c in the ranking of each agent 2 5 . Then Ri(O")
is a monotonic transformation of Ri(O) at a for each agent i C N, so monotonicity
would imply a F(0"). The neutrality condition then requires that, in view of the
symmetry of the two states 0' and 0", c E F(O') so c is not awkward. But with no
awkward outcomes monotonicity is equivalent to strong monotonicity. This yields a
nice characterization of Nash-implementable neutral social choice rules.

Theorem 3: [Moulin (1983)]. Suppose n > 3, and R(O) = 7PA or ZR(O) = ZA. Then
a neutral SCR is Nash-implementable if and only if it is monotonic.
Let a E F(O). Alternative c E Li(a, 0) is an essential outcome for agent i in Li(a, 0)
if and only if there exists 0 E 09 such that c E F(0) and Li(c, 0) C Li(a, 0). Let
Ei(a, 0) C Li(a, 0) denote the set of all outcomes that are essential for agent i in
Li(a, 0). An SCR F is essentially monotonic if and only if for all (a, 0, 0') E A x 0 x 0
the following is true: if a E F(O) and Ei(a, 0) C Li(a, 0') for all i E N, then
a E F(O'). If F is monotonic then Ei(a, 0) C Ci(a, 0) 26. If PA C TZ(O) then
Ci(a, 0) C Ei(a, 0) 27. Thus, while essential monotonicity is in general stronger than
strong monotonicity, the two conditions are equivalent if PA C ZR(0).
Theorem 4: [Danilov (1992)]. Suppose n > 3 and PA C 7Z(O). The SCR F is
Nash-implementable if and only if it is essentially monotonic.

24 There are no awkward outcomes in this case. Indeed, let a E F(O), and suppose c C Lj(a, 0) C Lj(c, 0')
and for each i -j, Li(c, 0') = A. We claim c C F(O'). Let 0" be such that for all i E N, cIi(0") a and for
all x,y C A {c}, xRi(O")y if and only if xRi(O)y. Since a E F(O), monotonicity implies a E F(O").
Pareto indifference implies c E F(O"). But Li(c, 0") = Li(a, 0) U {c} C Li(c, 0') for all i, so c E F(O')
by monotonicity.
25 To make use of the neutrality condition we need to assume that the preference domain R(O) is rich
enough that such permutations are admissible. Of course, this is true if Z(O9) = PA or R(O) = RA .
26 If C E Ej(a, 0) then there is 0 C O such that c C F(0) and Lj(c, 6) C Lj(a, 0). If Lj(a, 0) C Lj(c, 0')
and Li(c, 0') = A for each i #j, then c C F(O') by monotonicity. Hence, c E Cj (a, 0).
27 If c E Ci(a, 0) then c E F(O) for 0 E (O such that Lj(c, 0) = Lj(a, 0) and Li(c, O) = A for all i #j. So
c G Ej(a, 0).
254 E. iMaskin and T Sj&str6in

Yamato (1992) has shown that essential monotonicity is a sufficient condition for Nash
implementation in any environment (when n > 3), but it is a necessary condition only
if 1Z(O) is sufficiently large.

3.4. Weak implementation

If F(O) C F(O) for all 0 E O then F is a subcorrespondence ofF, denoted F C F. To
weakly implement the SCR F is equivalent to fully implementing a non-empty valued
subcorrespondence of F. Fix an SCR F, and for all 0 E O define

F*(O) _ {a E A: a F(O) for all 0 C O such that Li(a, 0) C Li(a, 0) for all i E N}.

Theorem 5. If F*(O) # 0 for all 0 0 then F* is a monotonic SCR.

Proof: Suppose a F*(O) and Li(a, 0) C Li(a, 0') for all i N. Suppose 0) a is
such that Li(a, 0') C Li(a, 0) for all i E N. Then Li (a, 0) C Li(a, 0') C Li(a, 0) for
all i. Since a E F*(O) we must have a E F(0). Therefore, a E F*(0'). l
If F*(0) = 0 for some 0 E O then F does not have any monotonic subcorrespondence,
but if F*(0) 0 for all 0 E O then F* is the maximal monotonic subcorrespondence
of F. Moreover, F is monotonic if and only if F* = F. Now, suppose that n > 3. If
F *(O) 0 for all 0 O and F satisfies no veto power, then F* satisfies no veto power
too and is Nash-implementable by Theorem 2, hence F is weakly implementable.
Conversely, if F is weakly Nash-implementable, then Theorem 1 implies that F has a
monotonic non-empty valued subcorrespondence F C F. Then F C F* so F*(O) : 0
for all 0 . Summarizing, we have the following.
Theorem 6. If F can be weakly Nash-implemented then F*(0) • 0 for all 0 E 0.
Conversely, if n > 3 and F satisfies no veto power and F*(0) • 0 for all 0 C O,
then F can be weakly Nash-implemented (and F* is the maximal Nash-implementable
subcorrespondence ofF).

3.5. Strategy-proofness and rich domains of preferences

We next show that there is an intimate connection between Nash-implementability and
strategy-proofness of an SCF, when the preference domain has a "product structure"
and is either "rich" or consists of strict orderings.
The preference domain RZ(O) has a product structure if it takes the form
RZ(O) = x= 1Zi. For any coalition C C N and any R E 1Z(0), we write R = (Rc, R c)
where Rc {Ri}ic C RZc(0) _ Xiec7R i and R c E xic'Zi. We also define
Rc(O) _ {Ri(O)}icc and R_c(O) - {Ri(0)},ic for any 0 O. If the SCF f
is ordinal, as it will be if it is monotonic, then the mapping f: R(O) A such
that f(R(O)) = f(0) for all 0 O is well defined. An ordinal SCF f on a domain
with a product structure is strategy-proof if, for all i E N, all 0 0, and all
Ch. 5: Implementation Theory 255

RI E RiT(O),ui(f(R,Ri), 0) > ui(f(RJ',R_i), 0), where (Ri,R i) = (Ri(0),Ri(0)). An

ordinal SCF f on a domain with a product structure is coalitionally strategy-proofif,
for all 0 E O, all non-empty coalitions C C N, and all preferences R'c CE c(O),
there exists i E C such that
us (f (Rc,R c), 0) > ui ( (R, )0) , (1)
where (R, R-c) = (Rc(O),R-c()). Note that coalitional strategy-proofness implies
ordinary strategy-proofness. If the SCF f is strategy-proof, then the revelation
mechanism r = (xi=7Ri,f) has the property that, for any i N and any 0 E O,
truthfully reporting Ri = Ri(O) is agent i's dominant strategy in state 0. If in addition
f is coalitionally strategy-proof, then no coalitional deviation from truth-telling can
make all members of the deviating coalition strictly better off.
To define "rich domain", we first introduce the concept of "improvement". If
ui(a, 0) > ui(b, 0) and ui(a, 0') < ui(b, O') and at least one inequality is strict, then b
improves with respect to afor agent i as the state changesfrom 0 to 6'. The following
condition was introduced by Dasgupta, Hammond and Maskin (1979).
Definition. Rich domain: For any a, b A and any 0, 0' E , if, for all i E N, b
does not improve with respect to a for when the state changes from 0 to ', then there
exists 0" E 0 such that Li(a, 6) C Li(a, 0") and Li(b, 6') C Lj(b, 0") for all i C N.

Theorem 7: [Dasgupta, Hammond and Maskin (1979)]. Supposef is a monotonic

SCE the domain is rich, and the preference domain has a product structure
Z(O) = x'= 17Zi. Then f is coalitionally strategy-proof
Proof: Let f be as hypothesized. Let C C N be any coalition. Suppose that the true
preference profile in state 0 is R = (Rc, R-c) = R(O). Consider a preference profile
R' = R(O') = (R ,Rc), with R Ri for i E C and R = Ri for i C. Let
a = f(0) = f(Rc,Rc) and b = f(0') = f(R',R_c). If a = b then Inequality (1)
holds trivially for all i E C, so suppose a b.
We claim that there exists i E C such that b improves with respect to a for agent i
as the state changes from 0 to '. Notice that because R = R i for i C, b cannot
improve with respect to a for any such agent. Hence, if the claim is false, the definition
of rich domain implies that there exists 0" such that Li(a, 0) C Li(a, 0") and
Li(b, 6') C Li(b, 0") for all i E N. But then, from monotonicity, we have a = f(O")
and b = f(0"), a contradiction off's single-valuedness. Hence the claim holds after
all.
But b improving with respect to a for agent i E C implies that
ui ( (Rc, R-c), ) > ui ( (R, Rc), ),
and so f is coalitionally strategy-proof as claimed. D

Theorem 8: [Dasgupta, Hammond and Maskin (1979)1. Suppose that n > 3. If

7ZR() has a product structure and consists of strict orderings ((0) C PA) andf is
a strategy-proofSCF satisfying no veto power, then f is Nash-implementable.
256 E. Maskin and T1Sjstrni

Proof: Let f be as hypothesized. We claim that f is monotonic. Suppose that, for

some 0, 0' E O and a A, we have a =f() and Li(a, 0) C Li(a, 0') for all i E N.
Let R = R(O) and R' = R(O'). Because R(O) has a product structure, there exists
a state 0" E such that (R,,R 2, ... , R,) = R(O"). Let c = f(0"). If c a, then
becausef is strategy-proof, ul (a, 0) > ul (c, 0) and ul (c, 0') > ul (a, 0'). But the former
inequality implies that c C L(a, 0) and, hence, from hypothesis, c E Li(a, O'), a
contradiction of the latter inequality. Thus, a = c, after all. We conclude that a C f(0"),
and, repeating the same argument for i = 2, ... , n, that a E f(0'). Thus, f is indeed
monotonic. Theorem 2 then implies that f is Nash-implementable. [

3.6. Unrestricteddomain of strict preferences

Suppose society has to make a choice from a finite set A. The set of admissible
preferences is the set of all linear orderings, R(O) = PA. This domain is rich, and
so Theorem 7 applies. The SCR F is dictatorial on its image if and only if there
exists i N such that for all 0 O and all a F(O), ui(a, 0) > ui(b, 0) for all
b c F(o).
Theorem 9: [Gibbard (1973), Satterthwaite (1975)]. Suppose that A is afinite set,
R(O) = PA, and f is a strategy-proofSCF such that f(OJ) contains at least three
alternatives. Then f is dictatorial on its image.

Theorem 10: [Muller and Satterthwaite (1977), Dasgupta, Hammond and Maskin
(1979), Roberts (1979)]. Suppose the SCFf is Nash-implementable, A is afinite set,
f(O) contains at least three alternatives, and R(0) = PA. Then f is dictatorialon its
image.
Proof: By Theorem if is monotonic. By Theorem 7f is strategy-proof. By Theorem 9
it must be dictatorial on its image. []
Theorem 10 is false without the hypothesis of single-valuedness. For example, the weak
Pareto correspondence is monotonic and satisfies no veto power in any environment,
so by Theorem 2, it can be Nash-implemented (when n > 3). Theorems 9 and 10 are
also false without the hypothesis that the image contains at least three alternatives.
To see this, let N(a, b, 0) denote the number of agents who strictly prefer a to b
in state 0. Suppose A = {x,y} and define the method of majority rule as follows:
F(O) = {x} if N(x,y, ) > N(y,x, 0), F(O) = y} if N(x,y, 0) < N(y,x, 0), and
F(O) = {x,y} if N(x,y, 0) = N(y,x, ). If n is odd and R(O) = PA then F is
single-valued, monotonic, and satisfies no veto power. By Theorem 2 it can be Nash-
implemented and by Theorem 7 it is coalitionally strategy-proof.
When A contains at least three alternatives the results are mainly negative. The
plurality rule (which picks the alternative that is top-ranked by the greatest number
of agents) is not monotonic, and neither are other well-known voting rules such
as the Borda and Copeland rules. None of these social choice rules can be even
Ch. 5: Implementation Theory 257

weakly Nash-implemented when IAI > 3. Peleg (1998) showed that all monotonic and
strongly unanimous SCRs violate Sen's (1970) condition of minimal liberty. Indeed,
if 1Z(O) = ;PA then monotonicity and strong unanimity imply Pareto optimality 2 8, but
Sen showed that no Pareto optimal SCR can satisfy minimal liberty.

3.7. Economic environments

An interesting environment is the L-good exchange economy (AE,N, OE). In this
environment no veto power is automatically satisfied when n > 3, since n - 1 non-
satiated agents can never agree on the best way to allocate the social endowment.
Thus, monotonicity will be both necessary and sufficient for implementation when
n > 3. The feasible set is

AE = a=(al,a
2 ,.... an) E RLx L X
... x L: ai < o
where ai E IRL is agent i's consumption vector, and wo E I4L+ is the aggregate
endowment vector2 9 . Let A' = {a C AE: ai 0 for all i N} denote the set of
allocations where no agent gets a zero consumption vector. Each agent cares only
about his own consumption and strictly prefers more to less. Although preferences
are defined only over feasible allocations in AE, it is conventional to introduce utility
functions defined on IR. Thus, in each state 0 E OE, for each agent i E N there
is a continuous, increasing 30 and strictly quasi-concave function vi(., 0): RL . IR
such that ui(a, 0) = vi(ai, 0) for all a A. Moreover, for any function from RL to
1R satisfying these standard assumptions, there is a state 0 OE such that agent i's
preferences are represented by that function. The domain RE - R(OE), which
consists of all preference profiles that can be represented by utility functions satisfying
these standard assumptions, is rich [Dasgupta, Hammond and Maskin (1979)]. By
Theorem 7, monotonicity implies strategy-proofness for single-valued social choice
rules. If n = 2 then strategy-proofness plus Pareto optimality implies dictatorship in this
environment [Zhou (1991)] 31. Strategy-proof, Pareto optimal and non-dictatorial social

28 For suppose ui(a, 0) > u(b, 0) for all i E N but b E F(O). Consider the state 0' where preferences
are as in state 0 except that a has been moved to the top of everybody's preference. Then, Ri(0') is
a monotonic transformation of R,(0) at b for all i so b C F(0') by monotonicity, but F(O') = {a} by
strong unanimity, a contradiction.
29 RL is L-dimensional Euclidean space, IRL = {x c IRL: Xk > 0, for k = 1, ... , L and R = x E L:
x, > 0, for k = 1, .. , L}.
30 A function vi(, 0) is increasing if and only if ui(ai, 0) > vi(a, 0) whenever ai > al, ai X a'.
31 Of course, these results depend on the assumptions we make about admissible preferences. Suppose
n = L = 2 and let O* C OE be such that in each state 0 E 6' both goods are normal for both agents.
Let e be a fixed "downward sloping line" that passes through the Edgeworth box. For each 0 E 9O*let
f(0) be the unique Pareto optimal and feasible point on . Then f: (O* AE is a monotonic, Pareto
optimal and non-dictatorial SCF which (using the mechanism described in Section 3.8) can be Nash
implemented in the environment (AE, { 1,2}, 0*).
258 E. Maskin and T Sjdstrim

choice functions exist when n > 3, but they are not very attractive [Satterthwaite and
Sonnenschein (1981)]. More positive results are obtained if the requirement of single-
valuedness is relaxed. Hurwicz (1979a) and Schmeidler (1980) constructed simple
"market mechanisms" where each agent proposes a consumption vector and a price
vector, and the set of Nash equilibrium outcomes coincides with the set of Walrasian
outcomes. Reichelstein and Reiter (1988) showed (under certain smoothness conditions
on the outcome function) that the minimal dimension of the message space M of any
such mechanism is approximately n(L - 1) + L/(n - 1) 32. However, the mechanisms
in these articles violated the feasibility constraint h(m) E A for all m E M. In fact,
the Walrasian correspondence W is not monotonic, hence not Nash-implementable,
in the environment (AE,N, OE). The problem occurs because a change in preferences
over non-feasible consumption bundles can eliminate a Walrasian equilibrium on the
boundary of the feasible set. The minimal monotonic extension of the Walrasian
correspondence W is the constrained Walrasian correspondence W c [Hurwicz, Maskin
and Postlewaite (1995)]. For simple, feasible and continuous implementation of the
constrained Walrasian correspondence, see Postlewaite and Wettstein (1989) and Hong
(1995). Under certain assumptions, any Nash-implementable SCR must contain Wc as
a sub-correspondence [Hurwicz (1979b), Thomson (1979)].
Hurwicz (1960, 1972) discussed "proposed outcome" mechanisms where each
agent i's message mi is his proposed net trade vector. "Information smuggling" can
be ruled out by requiring that in equilibrium h(m) = m. In exchange economies, a
proposed trade vector does not in general contain enough information about marginal
rates of substitution to ensure a Pareto efficient outcome [Saijo, Tatamitani and
Yamato (1996) and Sj6str6m (1996a)], although the situation may be rather different in
production economies with known production sets [Yoshihara (2000)]. Dutta, Sen and
Vohra (1995) characterized the class of SCRs that can be implemented by "elementary"
mechanisms where agents propose prices as well as trade vectors. This class contains
the Walrasian correspondence (on their preference domain, W = W).
For public goods economies, Hurwicz (1979a) and Walker (1981) constructed simple
mechanisms such that the set of Nash equilibrium outcomes coincides with the set
of Lindahl outcomes. Again, however, h(m) A was allowed out of equilibrium. In
Walker's mechanism each agent announces a real number for each of the K public
goods, so the dimension of M is nK, the minimal dimension of any smooth Pareto
efficient mechanism in this environment [Sato (1981), Reichelstein and Reiter (1988)].
Like the Walrasian correspondence, the Lindahl correspondence is not monotonic in
general. The minimal monotonic extension is the constrainedLindahl correspondence,
nicely implemented by Tian (1989).

32 The first term n(L - 1) is due to each agent proposing an (L - I)-dimensional consumption vector
for himself, and the second term L/(n - 1) comes from the need to also allow announcements of price
variables. Smoothness conditions are needed to rule out "information smuggling" [Hurwicz (1972),
Mount and Reiter (1974), Reichelstein and Reiter (1988)].
Ch. 5: Implementation Theory 259

In many economic environments a single crossing condition holds which makes

monotonicity rather easy to satisfy. For example, suppose there is a seller and a buyer,
a divisible good and "money". Let q denote the transfer of money from the buyer to the
seller (which can be positive or negative), and x > 0 the amount of the good delivered
from the seller to the buyer. The feasible set is A = {(q,x) C R2: x > 0}. The state of
the world is denoted 0 = (0s, Ob) e [0, 1] x [0, 1]- O. The seller's payoff function is
u( q,x, 0s), with ou/dq > 0, &u/0x < O0.The buyer's payoff function is u( q,x, Ob), with
Ov/dq < 0, v/ox > 0. An increase in 0 represents an increase in the seller's marginal
production cost, and an increase in b represents an increase in the buyer's marginal
valuation. More formally, the single crossing condition states that
>
Os 0,
Ou/Oq
&u/Oq 0 and v/q >0.
,9Ob cv/m9q
Under this assumption, a monotonic transformation can only take place at a boundary
allocation where x = 0. Monotonicity says that if (q, 0O)E F(O, Ob), Q> Os and
Ob < b, then ( q, 0) E F(O, ).

3.8. Two agent implementation

The necessary and sufficient condition for two-agent Nash implementation in general
environments was given by Moore and Repullo (1990) and Dutta and Sen (1991b). To
see why the case n = 2 may be more difficult than the case n > 3 note that rule 2
of the canonical mechanism for Nash implementation singles out a unique deviator
from a "consensus". However, with n = 2 this is not possible. Let a F(O) and
a' E F(O'). If F Nash-implements F then there are message profiles (ml, m 2) C Nr(O)
and (m',m') C Nr(0') such that h(ml,m2 ) = a and h(ml',m') = a'. Since agent 1
should have no incentive to deviate to message ml in state 0' and agent 2 should
have no incentive to deviate to message m2 in state 0, a property called weak non-
empty lower intersection must be satisfied: there exists an outcome b = h(ml, m')
such that a'R1(O')b and aR2(0)b. In most economic environments this condition
automatically holds, so the case n = 2 is similar to the case n > 3. In the two-agent
exchange economy (AE, {1, 2}, OE) (defined in Section 3.7) an SCR F can be Nash-
implemented if and only if it is monotonic and satisfies a very weak boundary condition
[Sj6str6m (1991)]. In particular, suppose F is monotonic and never recommends a
zero consumption vector to any agent. That is, F(OE) C AO. It is easy to check that
the following mechanism Nash-implements F. Each agent i {1,2} announces an
outcome a' = (a', a) AE, where aj is a proposed consumption vector for agent j,
and a state 0' e 9E. Thus, mi = (a', 0') Mi - A x E. Let hi(m) denote agent i's
consumption vector. Set hi(m) = a if
ml = m2 and ai E F(Oi),
or if
Rj(O') = Rj(O 1), Ri(OJ) • R(O') and aRi(Oj ) a'.
Otherwise, set hi(m) = 0.
260 E. Maskin and 7: Sjostrn

Such positive results for the case n = 2 do rely on restrictions on the domain of
preferences, as the following result shows.

Theorem 11: [Maskin (1999), Hurwicz and Schmeidler (1978)]. Suppose n = 2

and PA C Z(O9). If the SCR F is weakly Pareto optimal and Nash-implementable,
then F is dictatorial.
Proof: Suppose a weakly Pareto optimal SCR F is implemented by F = (M, h). For
any a C A, there is an agent i = i(a) E { 1,2} such that a is always in his attainable
set, i.e., a C h(mj,iMi) for all m ECMj (j i). For if not, then there is m E M such
that when m is played neither agent 1 nor agent 2 can attain a, but then x = h(m) is
a Pareto dominated Nash equilibrium outcome whenever both agents rank a first and
x second. In fact, for any two outcomes a and b we must have i(a) = i(b), for otherwise
there is no Nash equilibrium when agent i(a) ranks a first and agent i(b) ranks b first.
So there exists a dictator, i.e., an agent i such that h(mj,Mi) =A for all m E Mj. D

4. Implementation with complete information: further topics

4.1. Refinements of Nash equilibrium

Message mi Mi is a dominated strategy in state 0 E 0 for agent i N if and only

if there exists m' E Mi such that ui(h(mi, m'), 0) > ui(h(m i, mi), 0) for all m-i C M_i,
and ui(h(m-i, m), 0) > ui(h(mi,mi), 0) for some m i E M_i. A Nash equilibrium
is an undominated Nash equilibrium if and only if no player uses a dominated
strategy 3 3 . Notice that we are considering domination in the weak sense. It turns
out that "almost anything" can be implemented in undominated Nash equilibria. Of
course, a mechanism that implements a non-monotonic SCR F in undominated Nash
equilibria must have non-F-optimal Nash equilibria involving dominated strategies.
The assumption here, however, is that dominated strategies will never be used.
An SCR F satisfies property Q if and only if, for all (0, 0') x such that
F(O) Z F(0'), there exists an agent i C N and two alternatives (a, b) C A x A such that
b improves with respect to a for agent i as the state changes from 0 to 0', and moreover
this agent i is not indifferent over all alternatives in A in state 0'. Property Q is a
very weak condition because it only involves a preference reversal over two arbitrary
alternatives a and b, neither of which has to be F-optimal. If no agent is ever indifferent
over all alternatives in A, then property Q is equivalent to ordinality.

33 The Nash equilibria of the the canonical mechanism for Nash implementation are not necessarily
undominated, because if a E F(O) is the worst outcome in A for agent i in state 0 then it may be a
(weakly) dominated strategy for him to announce a. However, Yamato (1999) modified the canonical
mechanism so that all Nash equilibria are undominated. He showed that if n > 3 then any Nash
implementable SCR is doubly implementable in Nash and undominated Nash equilibria.
Ch. 5: Implementation Theory 261

Theorem 12: [Palfrey and Srivastava (1991)]. If the SCR F is implementable in

undominated Nash equilibria, then it satisfies property Q. Conversely, if n > 3 and F
satisfies property Q and no veto power, then F is implementable in undominatedNash
equilibria.
Proof: It is not difficult to see the necessity of property Q. To prove the sufficiency part,
we will simplify by assuming that (i) R(O) has a product structure, R(O) = x= R,
and (ii) value distinction holds: for all i G N and all ordered pairs (Ri, R') CGRi x Ri, if
R' Ri then there exist outcomes b and c in A such that cRib and bP'c. Let F satisfy
property Q and no veto power. Then F is ordinal, so we can suppose it is defined
directly on the set of possible preference profiles, F: Z _ x1 i i-
-- A. Consider the
following mechanism. Agent i's message space is

Mi =A x R x Ri x Z x Z x Z,

where Z is the set of all positive integers. A typical message for agent i is
mi = (a', R',r, z', , y') e Mi, where a' E A is an outcome, R = (Ri , R, . . ., R) E R
is a statement about the preference profile, ri E 1Zi is another report about agent i's own
preference, and (z i, ~i, y') are three integers. The outcome function is as follows 34.
Rule 1: If there exists j E N such that (a',R') = (a,R) for all i j, and a F(R),
then h(m) = a.
Rule 2: If rule 1 does not apply then:
(a) if there is j E N such thatj = (Zl= l z) mod(2n), set

h(m) = a';

(b) if there is j E N such that n +j = ( = l zk) mod(2n)

l, and yj >
set

aJ-1 ifaj lrja.+l.

h(m) =- a j+ otherwise

(c) if there is j N such that n +j = (n= 1 zk) mod(2n) and yj < - 1,

set
h() { aJ- t if a-' 1 Rjjaj+1
aj+ otherwise

Notice that rule 1 includes the case of a consensus, (a',R') = (a,R) for all i, as well
as the case where a single agent j differs from the rest. Rule 2a is a modulo game
similar to rule 3 of the canonical mechanism for Nash implementation. Rule 2b chooses

34 References to agentsj - I andj + 1 are always "modulo n" (ifj = 1 then agent j - 1 is agent n; if
j = n then agentj + is agent 1).
262 E. Maskin and T Sjostr6m

agent j's most preferred outcome among a- and a- 1 according to preferences r,

and rule 2c chooses agentj's most preferred outcome among a - and al + 1 according
to preferences Rj.
Let R* = (R*, ... , R) denote the true preference profile. Let Ur(R*) denote the set
of undominated Nash equilibria when the preference profile is R*. The proof consists
of several steps.
Step 1. If mj is undominated for agentj then rJ = R.'. Indeed, rJ only appears in
rule 2b, where "truthfully" announcing r = R* is always at least as good as any false
announcement. By value distinction there exists a - and a + t such that the preference
is strict.
Step 2. If mj is undominated for agent j then R' = R*. For, if R • R* then (since
rJ = Rj* by step 1) if n +j = (k= I z') mod(2n), agentj always weakly prefers rule 2b
to rule 2c, and by value distinction there exists a.I - and a j+ such that this preference
is strict. But increasing y j increases the chance of rule 2b at the expense of rule 2c,
without any other consequence, so m1 cannot be undominated.
Step 3. If m is a Nash equilibrium then either (a', R) = (a, R) for all i N
and a E F(R), or there is j such that for all i j, h(m)R'a for all a C A. This
follows from rule 2a (the same argument was used in the canonical mechanism for
Nash implementation).
Step 4. h(Ur(R*)) c F(R*). For, ifm E Ur(R*), then by steps 1 and 2, R' = r- = R*
for allj. By step 3, either rule 1 applies, in which case (a',R') = (a,R*) for all i C N
and h(m) = a E F(R*), or else h(m) G F(R*) by no veto power.
Step 5. F(R*) C h(Ur(R*)). Each agent j announcing (R',rJ) = (R*,Rj)
"truthfully" and aJ = a c F(R*) (and three arbitrary integers) is an undominated Nash
equilibrium. (Notice that if RI = ri then there is no possibility that y j can change the
outcome).
Steps 4 and 5 imply h(Ur(R*)) = F(R*). D
A similar possibility result was obtained for trembling-handperfectNash equilibriaby
Sj6str6m (1991). If agents have strict preferences over an underlying finite set of basic
alternatives B, and A = A(B) as discussed in Section 4.2, then a sufficient condition
for F to be implementable in trembling-hand perfect equilibria is that F satisfies no
veto power as well as its "converse": if all but one agent agree on which alternative
is the worst, then this alternative is not F-optimal. Yamato (1993) considers double
implementation in Nash and undominated Nash equilibrium.
A mechanism is bounded if and only if each dominated strategy is dominated by
some undominated strategy [Jackson (1992)]. The mechanism used by Sjo str6m (1991)
for trembling hand perfect Nash implementation has a finite message space, hence it
is bounded. But Palfrey and Srivastava's (1991) mechanism for undominated Nash
implementation contains infinite sequences of strategies dominating each other, hence
it is not bounded. This is illustrated by step 2 of the proof of Theorem 12. However,
in economic environments satisfying standard assumptions, any ordinal SCF which
Ch. 5: Implementation Theory 263

never recommends a zero consumption vector to any agent can be implemented in

undominated Nash equilibria by a very simple bounded mechanism which does not
use integer or modulo games.
Theorem 13: [Jackson, Palfrey and Srivastava (1994), Sjostrim (1994)]. Consider
the economic environment (AE,N, OE) with n > 2. Iff is an ordinal SCF such that
f (OE) C AO then f can be implemented in undominatedNash equilibriaby a bounded
mechanism.
Proof: We prove this for n = 2 using a mechanism due to Jackson, Palfrey and
Srivastava (1994)35. Iff is ordinal then without loss of generality we may assumef is
defined on IZE instead of on OE. Thus, considerf: RE --- A. Letfj(R) denote agentj's
f-optimal consumption vector when the preference profile is R. Each agent i {1, 2}
announces either a preference profile R' = (R,R) E ZE, or a pair of outcomes
(a', b) A x A' . Notice that a = (a, a2) is a pair of consumption vectors, and
i
b = (b , b) is another pair. Let hi(m) denote agent j's consumption.
Rule 1: Suppose both agents announce a preference profile. If Rj Rj, then hi(m) =O.
If Rj = RS, then hi(m) =f (RJ).
Rule 2: Suppose agent i announces a preference profile R' and agent j announces
outcomes (a, b). Then, h(m) = 0. If ajiPib then hi(m) = a, otherwise
hi(m) = bi.
Rule 3: In all other cases, hi(m) = h2 (m) = 0.
Suppose the true preference profile is R* = (R*,R*). It is a dominated strategy
to announce outcomes, since that guarantees a zero consumption bundle. Moreover,
truthfully announcing R = R* dominates lying since the only effect lying about
his own preferences can have on agent i's consumption is to give him an inferior
allocation under rule 2 36. Now, if agentj is announcing preferences, any best response
for agent i must involve R = RJ. (Since utility functions are increasing, getting
f(R) O0is strictly better than getting no consumption at all). Therefore, in the unique
undominated Nash equilibrium both agents announce the true preference profile, so this
mechanism implementsf. D1
The most disturbing feature of the mechanism in the proof of Theorem 13 is that
agent i's only reason to announce R' = R* truthfully is that it will give him a preferred
outcome in case agent j i uses the dominated strategy of announcing outcomes.
This problem does not occur in Sj6str6m's (1994) mechanism. In that mechanism,
each agent reports a preference ordering for himself and two "neighbors", and the only
dominated strategies are those where an agent does not tell the truth about himself.

35 Sj6str6m's (1994) mechanism is similar but works only for n > 3.

36 The allocation can be strictly inferior because value distinction holds in this environment. Indeed,
since preferences are defined over feasible outcomes, if Ri Ri then there is (a', bi) E A x A° such
that aJP*bj but b/Riaj .
264 E. Miaskin and T Sjsitrmn

When these dominated strategies have been removed, a second round of elimination of
strictly dominated strategies leads each agent to match what his neighbors are saying
about themselves.
The iteratedremoval of dominated strategies was considered by Farquharson (1969)
and Moulin (1979) in their analyses of dominance solvable voting schemes. Abreu and
Matsushima (1994) showed that if the feasible set consists of lotteries over a set of
basic alternatives, strict value distinction holds, and the social planner can use "small
fines", then any SCF can be implemented using the iterated elimination of dominated
strategies (without using integer and modulo games). It does not matter in which order
dominated strategies are eliminated, but many rounds of elimination may be required
[see Glazer and Rosenthal (1992) and Abreu and Matsushima (1992b)].
A Nash equilibrium is strong if and only if no group S C N has a joint deviation
which makes all agents in S better off. Monotonicity is a necessary condition for
implementation in strong Nash equilibria [Maskin (1979b, 1985)]. A necessary and
sufficient condition for strong Nash implementation was found by Dutta and Sen
(1991a), and an algorithm for checking it was provided by Suh (1995). Moulin and
Peleg (1982) established the close connection between strong Nash implementation
and the notion of effectivity function. For double implementation in Nash and strong
Nash equilibria, see Maskin (1979a, 1985), Schmeidler (1980) and Suh (1997). In the
environment AE,N, OE) with n > 2, any monotonic and Pareto optimal SCR F such
that F(OE) C A ° can be doubly implemented in Nash and strong Nash equilibria,
even if joint deviations may involve ex post trade of goods "outside the mechanism"
[Maskin (1979a), Sj6strbm (1996b)]. Further results on implementation with coalition
formation are contained in Peleg (1984) and Suh (1996).

4.2. Virtual implementation

Virtual implementation was first studied by Abreu and Sen (1991) and Matsushima
(1988). Let B be a finite set of "basic alternatives", and let the set of feasible outcomes
be A = A(B), the set of all probability distributions over B. The elements of A(B) are
called lotteries. Let A°(B) denote the subset of A(B) which consists of all lotteries that
give strictly positive probability to all alternatives in B. Let d(a, b) denote the Euclidean
distance between lotteries a, b E A(B). Two SCRs F and G are -close if and only if
for all 0 E there exists a bijection ao: F(O) G(O) such that d(a, ao(a)) < E
for all a E F(O). An SCR F is virtually Nash-implementable if and only if for all
E > 0 there exists an SCR G which is Nash-implementable and -close to F. If F is
virtually implemented, then the social planner accepts a strictly positive probability that
the equilibrium outcome is some undesirable element of B. However, this probability
can be made arbitrarily small.
Theorem 14: [Abreu and Sen (1991), Matsushima (1988)]. Suppose n > 3. Let
B be a finite set of "basic alternatives" and let the set of feasible alternatives be
A = A(B). Suppose for all 0 E , no agent is indifferent over all alternatives in
Ch. 5: Implementation Theory 265

B, and preferences over A satisfy the on Neumann-Morgenstern axioms. Then any

ordinal SCR F. 0 - A is virtually Nash-implementable.
Proof: Since any ordinal SCR F: 0 -- A(B) can be approximated arbitrarily closely
by an ordinal SCR G such that G(O) C A°(B), it suffices to show that any such G
is Nash-implementable. So let G: 0 - A°(B) be an ordinal SCR. In the environment
(A°(B), N, 0) the SCR G satisfies no veto power because no agent has a most preferred
outcome in A°(B). If a G(O) but a G(0'), then since G is ordinal there is
i N such that Ri(O) 7 Ri(O'). The von Neumann-Morgenstem axioms imply that
indifference surfaces are hyperplanes, so Ri(O') cannot be a monotonic transformation
of R i(O) at a C A°(B). Thus, G is monotonic. By Theorem 2, G is Nash-implementable
in environment (AO(B),N, 0). But then G is also Nash-implementable when the
feasible set is A(B), since we can always just disregard the alternatives that are not
in A°(B).
Of course, if an SCR is not ordinal then it cannot be virtually Nash-implemented,
so ordinality is both necessary and sufficient under the hypothesis of Theorem 1437.
The proof of Theorem 14 does not do justice to the work of Abreu and Sen (1991)
and Matsushima (1988), since their mechanisms are better behaved than the canonical
mechanism. For virtual implementation using iterated elimination of strictly dominated
strategies, see Abreu and Matsushima (1992a).

4.3. Mixed strategies

A mixed strategy i for agent i E N is a probability distribution over Mi. For
simplicity, we restrict attention to mixed strategies that put positive probability on
only a finite number of messages. Let lti(mi) denote the probability that agent i
sends message mi, let pu(m) - x=lui(mi) and Mj(mj) xijigi(mi). In most
of the implementation literature, only the pure strategy equilibria of the mechanism
are verified to be F-optimal, leaving open the possibility that there may be non-F-
optimal mixed strategy equilibria 38. In particular, in the proof of Theorem 2 we did
not establish that all mixed strategy Nash equilibria are F-optimal. In fact they need not
be. To see the problem, consider a mixed strategy Nash equilibrium = (l, ... , u)
for the canonical mechanism in state 0*. Suppose /u(m) > 0 for m such that rule 2
applies, that is,

(ai , 0i) = (a, 0) for all i Xj, (2)

but (aJ, 0j ) • (a, 0). If 8(m) = 1 then h(m) must be top-ranked by each agent i • j.
Otherwise, agent i j could induce his favorite alternative i ' via rule 3. Thus, no

37 Recall that ordinality says that only preferences over A matter for the social choice. Here, A = A(B).
38 Exceptions include Abreu and Matsushima (1992a), Jackson, Palfrey and Srivastava (1994) and
Sj6str6m (1994).
266 E. Maskin and 7TSjostr6dn

veto power guarantees h(m) F(O-). But suppose u_i(m'i) > 0 for some mn'i such
that m = (a', /, z') for all k i, where a' E F(O') and

u, (a, 0') > u (a', o') > u (a, 0') (3)

Then, although agent i can induce ai when the others play m , Inequality (3) and
rule 2 of the canonical mechanism imply that he cannot induce ai' when the others
play m'i. Indeed, if he tries to do so the outcome will be a', which in state 0* may
be much worse for him than a (the outcome that, from Inequality (3) and rule 2, he
would get by sticking to mi). Hence, he may prefer not to try to induce a' even if he
strictly prefers it to h(m). And so we cannot infer that h(m) is F-optimal. The difficulty
arises because which message is best for agent i to send depends on the messages that
the other agents send, but if the other agents are using mixed strategies then agent i is
unable to forecast (except probabilistically) what these messages will be. Nevertheless,
the canonical mechanism can be readily modified to take account of mixed strategies.
Suppose n > 3. The following is a version of a modified canonical mechanism
proposed by Maskin (1999). A typical message for agent i is mi = (ai , O',z, a'), where
ai E A is an outcome, 0' C O is a state, zi E Z is a positive integer, and a': A x 9 -- A
is a mapping from outcomes and states to outcomes satisfying a'(a,0) Li(a, 0) for
all (a, 0). Let the outcome function be defined as follows.
Rule 1: Suppose there exists j E N such that (ai ', O,z i ) = (a, 0, 1) for all i j and
z = 1. Then h(m) = a.
Rule 2: Suppose there exists j E N such that (a', O',z') = (a, 0, 1) for all i j and
zi > 1. Then h(m) = aJ(a, 0).
Rule 3: In all other cases let h(m) = a' for i such that z' > z j for all j E N (if there
are several such i, choose the one with the lowest index i).
Notice that rule 1 encompasses the case of a consensus, (a', 0i,z') = (a, 0, 1)
for all i E N. The mapping a' enables agent i, in effect, to propose a contingent
outcome, which eliminates the difficulty noted above. Indeed, for any mixed Nash
equilibrium A, agent i has nothing to lose from setting a(a,0) equal to his favorite
outcome in Li(a, 0), a equal to his favorite outcome in all of A, and z' larger than
any integer announced with positive probability by any other agent 39. Such a strategy
guarantees that he gets his favorite outcome in his attainable set Li(a, 0) whenever
(ak, ok,zk) = (a, 0, 1) for all k • i, and for all other m-i such that u_i(mi) > 0 it will
cause him to win the integer game in rule 3. Thus, in Nash equilibrium, if P(m) > 0
and rule 1 applies to m, so (ai , Oi) = (a, 0) for all i, then h(m) = a must be the most
preferred alternative in Li(a, 0) for each agent i. But if instead rule 2 or rule 3 applies
to m then h(m) must be top-ranked in all of A by at least n - 1 agents. Thus, if F

39 If such favorite outcomes do not exist, the argument is more roundabout but still goes through. The
same is true if the other agents use mixed strategies with infinite support. In that case, agent i cannot
guarantee that he will have the highest integer, but he can make the probability arbitrarily close to one
and that is all we need.
Ch. 5: Implementation Theory 267

is monotonic and satisfies no veto power then (m) > 0 implies h(m) is F-optimal.
Conversely, if a E F(O) then there is a pure strategy Nash equilibrium in state 0
where (ai, Oi,z i) = (a, 0, 1) for all i N 40. So this mechanism Nash-implements F
even when we take account of mixed strategies.
Maskin and Moore (1999) show that the extensive form mechanisms considered by
Moore and Repullo (1988) and Abreu and Sen (1990) can also be suitably modified
for mixed strategies. We conjecture that analogous modifications can be made for
mechanisms corresponding to most of the other solution concepts that have been
considered in the literature.

4.4. Extensive form mechanisms

An SCR F is implementable in subgame-perfect equilibria if and only if there exists
an extensive form mechanism such that in each state 0 E 0, the set of subgame-
perfect equilibrium outcomes equals F(O). Extensive form mechanisms were studied
by Farquharson (1969) and Moulin (1979). Moore and Repullo (1988) obtained a
partial characterization of subgame-perfect implementable SCRs. Their result was
improved on by Abreu and Sen (1990).
To illustrate the ideas that are involved, consider a quasi-linear environment with
two agents, N = 1, 2}. There is an underlying set B of "basic alternatives", which can
be finite or infinite. In addition, a good called "money" can be used to freely transfer
utility between the agents. Let yi denote the net transfer of money to agent i, which
can be positive or negative. However, we assume social choice rules are bounded: they
do not recommend arbitrarily large transfers to or from any agent. A typical outcome
is denoted a = (b,yl,y 2). The feasible set is
A = {(b,yl,y 2) e B x kR x R: y +Y2 < 0}.
Notice that yl + Y2 < 0 is allowed (money can be destroyed or given to some
outside party). In all states, each agent i's payoff function is of the quasi-linear
form ui(a, 0) = vi(b, 0) + yi, where vi is bounded. Assume strict value distinction
in the sense that we can select (b(O, O'),y(O, 0')) E B x R, for each ordered pair
(0, 0') E 0 x 0, such that the following is true. Whenever 0 # 0', there exists a "test
agent" j = j(O, 0') = j(O', 0) E N that experiences a strict preference reversal of the
form:
vj (b (0, '), 0) +y (0, 0') > vj (b (0', 0), 0) +y (0', 0), (4)
and
i (b (, 0'), ') +y (, ') < uj (b (', ), ') +y (', ) . (5)
In this environment, any bounded SCF f: 0 -- A can be implemented in subgame-
perfect equilibria by the following simple two-stage mechanism. [See Moore and

40 The Nash equilibrium strategies are undominated as long as a is neither the best nor the worst
outcome in A for any agent.
268 E. Maskin and IT Sjdstrdom

Repullo (1988) and Moore (1992) for similar mechanisms.] Stage consists of
simultaneous announcements of a state: each agent i E N announces i' E . If
01 = 2 = 0 then the game ends with the outcomef(0). If 0'1 02, then go to stage 2.
Let j(1) = j(0 l , 02) denote the "test agent" for (01, 02), let 0 = 0i(i) denote the test
agent's announcement in stage 1 and let 0' = 0 j( o) denote the announcement made by
the other agent, agent j(O) #j(1). Let a(1) = (b(O, 0'),yl,y2) with Yj(l) = y(0, 0') - z
and yj(o) = -z where z > 0. Let a(2) = (b(0', O),yl ,Y2) with Yj(1) = y(0', 0) - z and
Y(0) = r > 0. In stage 2, agent j(1) decides the outcome of the game by choosing
either a(1) or a(2). By Inequalities (4) and (5), agent j(1) prefers a(2) to a(l) if 0' is
the true state, but he prefers a(1) to a(2) if 0 is the true state. In effect, agent j(O)'s
announcement 0' is "confirmed" if agentj(l) chooses a(2), and then agentj(0) receives
a "bonus" r. But if agentj(l) chooses a(1l), then agentj(0) pays a "fine" z. Agentj(1)
pays the fine whichever outcome he chooses in stage 2 (this does not affect his
preference reversal over a(1) and a(2)).
If the agents disagree in stage 1, then at least one agent must pay the fine z. This is
incompatible with equilibrium if z is sufficiently big, because any agent can avoid the
fine by agreeing with the other agent in stage 141. Thus, in equilibrium both agents will
announce the same state, say 01 = 02 = 0, in stage 1. Suppose the true state is 0' 0.
Let j(l) =j(, 0') be the test agent for (0, 0'). Suppose agent j(0) j(1) deviates in
stage 1 by announcing (°0) = 0' truthfully. In stage 2, agent j(1) will choose a(2) so
agent j(0) will get the bonus r which makes him strictly better off if r is sufficiently
big. Thus, if z and r are big enough, in any subgame-perfect equilibrium both agents
must announce the true state in stage 1. Conversely, both agents announcing the true
state in stage 1 is part of a subgame-perfect equilibrium which yields the f-optimal
outcome (no agent wants to deviate, because he will pay the fine if he does). Thus,f is
implemented in subgame-perfect equilibria. The reader can verify that the sequences
a(0) =f(0), a(1), a(2) in A, and j(0),j(1) in N, fulfil the requirements of the following
definition (with e = 1 and A' = A).
Definition. Property a: There exists a set A', with F(O) C A' C A, such that for all
(a, 0, O') E A x x O the following is true. If a F(O) F(0') then there exists a
sequence of outcomes a(O) = a, a(l), ... , a(f), a(f + 1) in A' and a sequence of agents
j(O),j(1), ... , j(f) in N such that:
(i) for k = 0,1, ... , ,
Uj(k) (a(k), 0) > u(k) (a(k + 1), 0);

(ii)

Uj(e) (a(), 0') < ut(t) (a(f + 1), 0');

41 As long as f and vu are bounded, each agent prefers any f(0) to paying a large fine. Without
boundedness, z and r would have to depend on (0, 0').
Ch. 5: Implementation Theory 269

(iii) for k = 0, 1, ... , , in state 0' outcome a(k) is not the top-ranked outcome in A'
for agent j(k);
(iv) if in state 0', a(f + 1) is the top-ranked outcome in A' for each agent i j(),
then either = 0 orj(t- 1) j(f).
If F is monotonic then a E F(O) - F(0') implies the existence of (a(l),j(O)) C A x N
such that uj(o)(a, 0) > uj(o)(a(l), 0) and uj(o)(a, 0') < uj(o)(a(l), O'), so sequences
satisfying (i)-(iv) exist (with = 0). Hence, property a is weaker than monotonicity.
Recall that property Q requires that someone's preferences reverse over two arbitrary
alternatives. Since property a requires a preference reversal over two alternatives a(t)
and a(f + 1) that can be connected to a by sequences satisfying (i)-(iv), property a is
stronger than property Q.
Theorem 15: [Moore and Repullo (1988), Abreu and Sen (1990)]. If the SCR F is
implementable in subgame-perfect equilibria, then it satisfies property a. Conversely,
if n > 3 and F satisfies property a and no veto power then F is implementable in
subgame-perfect equilibria.
Recently, Vartiainen (1999) found a condition which is both necessary and sufficient
for subgame-perfect implementation when n > 3 and A is a finite set. Herrero
and Srivastava (1992) derived a necessary and sufficient condition for an SCF to be
implementable via backward induction using a finite game of perfect information. An
interesting connection between extensive and normal form implementation is drawn
by Glazer and Rubinstein (1996).

4.5. Renegotiation

So far we have been assuming implicitly that the mechanism F is immutable. In this
section we shall allow for the possibility that agents might renegotiate it. Articles
on implementation theory are often written as though an exogenous planner simply
imposes the mechanism on the agents. But this is not the only possible interpretation
of the implementation setting. The agents might choose the mechanism themselves,
in which case we can think of the mechanism as a "constitution", or a "contract"
that the agents have signed. Suppose that when this contract is executed (i.e., when
the mechanism is played) it results in a Pareto inefficient outcome. Presumably,
if the contract has been properly designed, this could not occur in equilibrium:
agents would not deliberately design an inefficient contract. But inefficient outcomes
might be incorporated in contracts as "punishments" for deviations from equilibrium.
However, if a deviation from equilibrium has occurred, why should the agents accept
the corresponding outcome given that it is inefficient? Why can't they "tear up"
their contract (abandon the mechanism) and sign a new one resulting in a Pareto
superior outcome? In other words, why can't they renegotiate? But if punishment
is renegotiated, it may no longer serve as an effective deterrent to deviation from
equilibrium. Notice that renegotiation would normally not pose a problem if all that
270 E. Maskin and T SjJstri6m

mattered was that the final outcome should be Pareto optimal. However, a contract
will in general try to achieve a particular distribution of the payoffs (for example,
in order to share risks), and there is no reason why renegotiation would lead to the
desired distribution. Thus, the original contract must be designed with the possibility of
renegotiation explicitly taken into account. Our discussion follows Maskin and Moore
(1999). A different approach is suggested by Rubinstein and Wolinsky (1992).
Consider the following example, drawn from Maskin and Moore (1999). Let
N = {1, 2}, O = {0, 0'}, and A = {a, b, c}. Agent 1 always prefers a to c to b. Agent 2
has preferences cP2() aP2(0)b in state 0 and bP2 (0') aP2(0') c in state '. Suppose
f(0) = a and f(0') = b. If we leave aside the issue of renegotiation for the moment,
there is a simple mechanism that Nash-implements this SCF, namely, agent 2 chooses
between a and b. He prefers a in state 0 and b in state 0' and sof will be implemented.
But what if he happened to choose b in state 0? Since b is Pareto dominated by a and c
the agents will be motivated to renegotiate. If, in fact, b were renegotiated to a, there
would be no problem since whether agent 2 chose a or b in state 0, the final outcome
would be a =f(0). However, if b were renegotiated to c in state 0, then agent 2 would
intentionally choose b in state 0, anticipating the renegotiation to c. Then b would not
serve to punish agent 2 for deviating from the choice he is supposed to make in state 0,
and the simple mechanism would no longer work. Moreover, from Theorem 16 below,
no other mechanism can implement either. Thus, renegotiation can indeed constrain
the SCRs that are implementable. But the example also makes clear that whether or not
f is implementable depends on the precise nature of renegotiation (if b is renegotiated
to a, implementation is possible; if b is renegotiated to c, it is not). Thus, rather than
speaking merely of the "implementation off", we should speak of the "implementation
off for a given renegotiation process".
In this section the feasible set is A = A(B), the set of all probability distributions
over a set of basic alternatives B. We identify degenerate probability distributions that
assign probability one to some basic alternative b with the alternative b itself. The
renegotiation process can be expressed as a function r: B x - B, where r(b, 0)
is the (basic) alternative to which the agents renegotiate in state 0 E O if the fall-
back outcome (i.e., the outcome prescribed by the mechanism) is b B. Assume
renegotiation is efficient (for all b and 0, r(b, 0) is Pareto optimal in state 0) and
individually rational (for all b and 0, r(b, 0)Ri(0)b for all i)4 2. For each 0 C O,
define a function r: B -- B by ro(b) = r(O, b). Let x E A, assume for the moment
that B is a finite set, and let x(b) denote the probability that the lottery x assigns to
outcome b E B. Extend ro to lotteries in the following way: let r(x) C A be the lottery
which assigns probability Ex(a) to basic alternative b B, where the sum is over

42 Jackson and Palfrey (2001) propose an alternative set of assumptions. If in state 0 any agent can
ueto the outcome of the mechanism and instead enforce an alternative a(&), renegotiation will satisfy
r(b, 0) = b if bRi(0) a(O) for all i E N, and r(b, 0) =a(O) otherwise. In an exchange economy, a(O) may
be the endowment point, inwhich case the constrained Walrasian correspondence is not implementable
[Jackson and Palfrey (2001)].
Ch. 5: Implementation Theory 271

the set {a: ro(a) = b}. For B an infinite set, define r(x) in the obvious analogous
way. Thus, we now have ro: A A for all 0 E . Finally, given a mechanism
F = (M,h) and a state 0 E , let r o h denote the composition of r and h.
That is, for any m M, (ro o h)(m) = ro(h(m)). The composition ro o h: M - A
describes the defacto outcome function in state 0, since any basic outcome prescribed
by the mechanism will be renegotiated according to ro. Notice that if the outcome h(m)
is a non-degenerate randomization over B, then renegotiation takes place after the
uncertainty inherent in h(m) has been resolved and the mechanism has prescribed a
basic alternative in B. Let S((M, ro o h) , 0) denote the set of S-equilibrium outcomes
in state 0, when the outcome function h has been replaced by r o h. A mechanism
F = (M, h) is said to S-implement the SCR F for renegotiationfunction r if and only
if S((M, r o h), 0) = F(O) for all 0 E 0. In this section we restrict our attention to
social choice rules that are essentially single-valued: for all 0 G O, if a F(O) then
F(O) = {b E A: bIi(0)a for all i C N}.
Much of implementation theory with renegotiation has been developed for its
application to bilateral contracts. With n = 2, a simple set of conditions are necessary
for implementation regardless of the refinement of Nash equilibrium that is adopted
as the solution concept.
Theorem 16: [Maskin and Moore (1999)]. The two-agent SCR F can be implemented
in Nash equilibria (or any refinement thereof)for renegotiationfunction r only if there
exists a random function i: 0 x 0 - A such that,
(i) for all 0 C O,

ro (a (0, 0)) E F(0);

(ii) andfor all (0, 0') E 0 x O,

ro (a (0, 0)) R (0) r (a (0', 0));

(iii) and

ro (a(0,0)) R2(0) r (a (0, 0'))

If (0, 0) is the (random) equilibrium outcome of a mechanism in state 0, then

condition (i) ensures that the renegotiated outcome is F-optimal, and conditions (ii)
and (iii) ensure that neither agent 1 nor agent 2 will wish to deviate and act as though
the state were 0'.
The reason for introducing randomizations over basic alternatives in Theorem 16
and the following results is to enhance the possibility of punishing agents for deviating
from equilibrium. By assumption, agents will always renegotiate to a Pareto optimal
alternative. Thus, if agent 1 is to be punished for a deviation (i.e., if his utility is to
be reduced below the equilibrium level), then agent 2 must, in effect, be rewarded for
272 E. Maskin and 7:Sjdstrdmn

this deviation (i.e., his utility must be raised above the equilibrium), once renegotiation
is taken into account. But as we noted in Section 3.8, determining which agent has
deviated may not be possible when n = 2, so it may be necessary to punish both
agents. However, this cannot be done if one agent is always rewarded when the other is
punished. That is where randomization comes in. Although, for each realization b e B
of the random variable E A, ro(b) is Pareto optimal, the random variable ro(a)
need not be Pareto optimal (if the Pareto frontier in utility space is not linear). Hence,
deliberately introducing randomization is a way to create mutual punishments despite
the constraint of renegotiation.
In the case of a linear Pareto frontier 4 3 randomization does not help. In that case,
the conditions of Theorem 16 become sufficient for implementation.
Theorem 17: [Maskin and Moore (1999)]. Suppose that the Paretofrontier is linear
for all 0 C . Then the two-agent F can be implemented in Nash equilibriafor
renegotiationfunction r if there exists a random function : 0 x 0 -- A satisfying
conditions (i), (ii) and (iii) of Theorem 16.
Under the hypothesis of Theorem 17, a mechanism in effect induces a two-person zero-
sum game (renegotiation ensures that outcomes are Pareto efficient, and the linearity
of the Pareto frontier means that payoffs sum to a constant). In zero-sum games, any
refined Nash equilibrium must yield both players the same payoffs as all other Nash
equilibria. Theorems 16 and 17 show that using refinements will not be helpful for
implementation in such a situation.
With "quasi-linear preferences" the Pareto frontier is linear, and Segal and Whinston
(2002) have shown that Theorem 17 can be re-expressed in terms of first-order
conditions 44
Theorem 18: [Segal and Whinston (2002)]. Assume
(i) N = {1,2},
(ii) the set of alternatives is

A = {(b,yi,y 2) E B x IR x IR: Yl +Y2 = 0},

where B is a connected compact space;

(iii) = [0, 0] is a compact interval in IR; and
(iv) in each state 0 E 0, each agent i s post-renegotiationpreferences take the form:
for all (b,y 1 ,y 2) E A,

ui (ro (b,yl,y 2 ), 0)= vi(b, ) +Yi,

where vi is C'.

43 Formally, the frontier is linear in state 0 if, for all b, b' E B that are both Pareto optimal in state 0,
the lottery 2.b + (1 - A) b' is also Pareto optimal, where ; is the probability of b.
44 Notice that their feasible set is different from what we otherwise assume in this section.
Ch. 5: Implementation Theory 273

If the SCR F: 9 -- A is implementable in Nash equilibria (or any refinement thereof)

for renegotiationfunction r, then there exists b: - B such that, for all 0 e 0 and
all i E N,

ui(F(O),0)= f (b(t),t) dt+ui(F(_),O_. (6)

Furthermore, if there is i C N such that (0 2 vi/a00b)(b, 0) > 0 for all b B and

all 0 E (O, then the existence of b satisfying Inequality (6) is sufficient for F S Nash-
implementability by a mechanism where only agent i sends a message.
Notice that as F is essentially single-valued, we may abuse notation and write
ui(F(O), 0) in Inequality (6).
When the Pareto frontier is not linear it becomes possible to punish both agents
for deviations from equilibrium. We obtain the following result for implementation in
subgame-perfect equilibria.
Theorem 19: [Maskin and Moore (1999)]. The two-agent SCR F can be implemented
in subgame-perfect equilibria with renegotiationfunction r if there exists a random
function i: --* A such that
(i) for all 0 E O, r(a(0), 0) E F(O);
(ii) for all (0, 0') E 0 x such that r(a(O), 0') X F(0') there exists an agent k and
a pair of random alternatives b(O, 0'), (O, 0') in A such that

r (0,0 '), 0) Rk(0)r ( (0,0'), 0),

and

r( (, '), 0') Pk (0') r (b (0, '), 0');

(iii) ifZ C A is the union of all ii()for 0 cE together with all b(0, 0') and c(0, 0')
for 0, 0' E , then no alternative z Z is maximal for any agent i in any
state 0 cE even after renegotiation (that is, there exists some di(0) A such
that di(0)Pi(0)r(z, 0)); and
(iv) there exists some random alternative A such that, for any agent i and any
state 0 C O, every alternative in Z is strictly preferred to after renegotiation
(that is, r(z, 0) Pi(O)r(, 0) for all z C Z).
The definition of implementation with renegotiation suggests that characterization
results should be r-translations of those for implementation when renegotiation is
ruled out. That is, for each result without renegotiation, we can apply r to obtain the
corresponding result with renegotiation. This is particularly clear if Nash equilibrium is
the solution concept. From Theorems 1 and 2 we know that monotonicity is the key to
Nash implementation. By analogy, we would expect that some form of "renegotiation-
monotonicity" should be the key when renegotiation is admitted. More precisely, we
274 E. Maskin and T Sjistr6m

say that the SCR F is renegotiation monotonic for renegotiationfunction r provided

that, for all 0 O and all x E F(O) there is a C A such that r(a, 0) = x, and if
Li(r(a, 0), 0) C Li(r(a, 0'), 0') for all i E N then r(a, 0') e F(O').
Theorem 20: [Maskin and Moore (1999)]. The SCR F can be implemented in Nash
equilibria with renegotiationfunction r only if F satisfies renegotiation monotonicity
for r. Conversely, if n > 3 and no alternative is maximal in A for two or more agents,
then F is implementable in Nash equilibriawith renegotiationfunction r if F satisfies
renegotiation monotonicity for r.
By analogy with Section 4.1, Nash equilibrium refinements should allow the
implementation of social choice rules that do not satisfy renegotiation monotonicity.
Theorem 16 has in fact put substantial limits on what can be achieved when n = 2.
But the situation when n > 3 is very different, at least in economic environments.
Introducing a third party into a bilateral economic relationship makes it possible to
simultaneously punish both original parties by transferring resources to the third party,
which makes the problem of renegotiation much less serious 45. Before stating this
result formally, we need a definition. A renegotiation function r: AE X OE -- AE
satisfies disagreementpointmonotonicity if for all i C N, all 0 e OE and all a, b E AE
such that all agents except i get no consumption (aj = b = 0 for all j • i), it holds
that r(a, 0)Ri(0)r(b, 0) if and only if aRi(0)b. That is, if two fall-back outcomes
a and b both give zero consumption to everyone except agent i, then agent i prefers
to renegotiate from whichever fall-back outcome gives him higher utility. Standard
bargaining solutions such as the Nash solution and the Kalai-Smorodinsky solution
satisfy this property.
Theorem 21: [Sjstrom (1999)]. Consider the environment (AE,N, OE) with n > 3.
Let r be any renegotiation function that satisfies disagreement point monotonicity
and individual rationality. If f is an ordinal and Pareto optimal SCF such that
f(OE) C A, then f can be implemented in undominated Nash equilibria with
renegotiationfunction r.
Sj6str6m's (1999) mechanism is "non-parametric" in the sense that it does not depend
on r. Moreover, it is both bounded and robust to collusion. It is sometimes argued that
introducing a third party into a bilateral relationship may lead to collusion between the
third party and one of the original parties. However, all undominated Nash equilibria
of Sj6str6m's (1999) mechanism are coalition-proof, which is the appropriate solution
concept when agents can collude but cannot write binding side-contracts ex ante
(allowing binding ex ante agreements would take the analysis into the realm of
n-person cooperative game theory). A possibility result similar to Sj6str6m's (1999)

45 What is important is not that the third person knows the true state of the world, only that he is willing
to accept transfers of goods from the original parties.
Ch. 5: Implementation Theory 275

was obtained by Baliga and Brusco (2000) for implementation using extensive form
mechanisms.

4.6. The planner as a player

The canonical mechanism for Nash implementation can be given the following intuitive
explanation. Rule 1 states that if (a, 0) is a consensus among the agents, where
a E F(O), then the outcome is a. Rule 2 states that agent j's attainable set at the
consensus is the lower contour set L(a, 0). By "objecting" against the consensus,
agent j can induce any a e Lj(a, 0). Monotonicity is the condition that makes
such objections effective. For if 0' 0 is the true state and a F(0'), then by
monotonicity some agent j strictly prefers to deviate from the consensus with an
objection aJ E Lj(a, 0) - Lj(a, 0'). Agent j would have no reason to propose a in
state 0 since a L(a, 0), but he does have such an incentive in state ' since
aj Lj(a, 0').
Now suppose the mechanism is controlled by a social planner who does not know
the true state of the world. She gets payoff uo(a, 0) from alternative a in state 0, and
the SCR F she wants to implement is

F(O)- argmax uo(a, 0). (7)

aEA

Suppose F is Nash-implementable and the planner uses the canonical mechanism to

implement it. By Inequality (7), the equilibrium outcome maximizes her payoff in each
state of the world. But out of equilibrium, she faces a credibility problem similar to
the one discussed in the previous section. After hearing out of equilibrium messages,
she may want to change the rules that she herself has laid down. Specifically, consider
the "objection" made by agentj which was described in the previous paragraph. Let
O' = {0' c O: a Lj(a, 0')} be the set of states where agent j strictly prefers a to
a. If player j tries to induce a via rule 2, when all the other agents are announcing
(a, 0), then [following the logic of Farrell (1993) and Grossman and Perry (1986)] the
planner's beliefs about the true state should be some probability distribution over 0'.
But aJ may not maximize the planner's expected payoff for any such beliefs, in which
case she prefers to "tear up" the mechanism after agent j has made his objection. In
this sense the outcome function may not be credible. The situation is even worse if
the "modulo game" in rule 3 is triggered. Rule 3 may lead to zero consumption for
everybody except the winner of the modulo game, but that may be an outcome the
planner dislikes regardless of her beliefs about the state. If the planner cannot commit
to carrying out "incredible threats" such as giving no consumption to n - 1 agents,
then the implementation problem is very difficult. Conditions under which the planner
can credibly implement the SCR given by Inequality (7) are discussed by Chakravorty,
Corch6n and Wilkie (1997) and Baliga, Corch6n and Sj6strrm (1997).
276 E. Maskin and 7TSjcstrni

On the other hand, if the planner can commit to the outcome function then explicitly
allowing her to participate as a player in the game expands the set of implementable
social choice rules. Consider a utilitarian social planner with payoff function

uo(a, 0) = ui(a, 0).

i=

The SCR F she wants to implement is the utilitarian SCR which is not even ordinal (it
is not invariant to multiplying an agent's utility function by a scalar). If the planner does
not play then this F cannot be implemented using any non-cooperative solution concept
(even virtually). However, suppose the environment is (AE,N, OE) with n > 3. If we
let the planner, who does not know the true 0, participate in the mechanism by sending
a message of her own, then the utilitarian SCR can be implemented in Bayesian Nash
equilibria for "generic" prior beliefs over 0 [Baliga and Sj6str6m (1999)]. This does
not quite contradict the fact that only ordinal social choice rules can be implemented.
Inequality (7) implies that if F(O) X F(O') then the planner's preferences over A must
differ in states 0 and 0', so all social choice rules are ordinal if the planner's own
preferences are taken into account 4 6.

5. Bayesian implementation

Now we drop the assumption that each agent knows the true state of the world and
consider the case of incomplete information.

5.1. Definitions

A generic state of the world is denoted 0 = (01, ... , 0,), where Oi is agent i's type. Let
Oi denote the finite set of possible types for agent i, and 0 _ 01 x ... x 0,. Agent i
knows his own type Oi but may be unsure about 0_i (01, ... , Oi-, i+.. 0,,).
Agent i's payoff depends only on his own type and the final outcome (private values).
Thus, if the outcome is a c A and the state of the world is 0 = (01, ... , 0,) E O, then
we will write agent i's payoff as ui(a, i) rather than ui(a, 0). There exists a common
prior distribution on O, denoted p. Conditional on knowing his own type 0i, agent i's
posterior distribution over 0-i - xji O is denoted p(. I i). It can be deduced from
p using Bayes' rule for any Oi which occurs with positive probability. If g: 0i - A

46 Hurwicz (1979b) considered the possibility of using an "auctioneer" whose payoff function agrees
with the SCR. However, he considered Nash equilibria among the n+ 1players, which implicitly requires
the auctioneer to know the true 0 (or else relies on some adjustment process as discussed in the
Introduction).
Ch. 5: Implementation Theory 277

is any function, and Oi E Oi, then the expectation of ui(g(Oi), Oi) conditional on Oi is
denoted

E{ui(g(Oi),Oi)[0i}= E P(silO,')us(g(0i),'i).
O- E Oi

A strategy profile in the mechanism F = (M,h) is denoted a = (l,..., a,),

where for each i, oi: Oi Mi is a function which specifies the messages
sent by agent i's different types. The message profile sent at state 0 is denoted
a(O) = ((), ... , n,(O,)), and the message profile sent by agents other than i in
state 0 = (O-i, Oi) is denoted

a-i ( = (a (01), ... , i- ( - 1, aI+i (i+ ), .. , a ())

Let Z denote the set of all strategy profiles. Strategy profile a E 2 is a Bayesian
Nash Equilibrium if and only if for all i E N and all Oi E Oi,

E {us (h (a (i, 0i)), i) 0i} > E{ui (h (i ( i), m), Oi) I i},

for all m i G Mi. All expectations are with respect to O-i conditional on Oi. Let
BNE r c Z denote the set of Bayesian Nash Equilibria for mechanism F.
A social choice set (SCS) is a collection F = {f ,f2, .. .} of social choice functions,
i.e., a subset of A9. We identify the SCF f: - A with the SCS F = {f}. Define
the composition h o a: O --+ A by (h o a)(0) = h(a(O)). A mechanism F = (M, h)
implements the SCS F in Bayesian Nash equilibria if and only if (i) for all f E F
there is a C BNE r such that h o a =f, and (ii) for all a e BNE r there isf E F such
that h o a =f.

5.2. Closure

A set O' C O is a common knowledge event if and only if 0' = (i, 0,') O' and
0 = (0-i, 0i) O'9 implies, for all i E N,p(Oi I 0,') = 0. If an agent is not sure about the
true state, then in order to know what message to send he must predict what messages
the other agents would send in all those states that he thinks are possible, which links
a number of states together. However, two disjoint common knowledge events O1 and
02 are not linked in this way. For this reason, a necessary condition for Bayesian Nash
implementation of an SCS F is closure [Postlewaite and Schmeidler (1986), Palfrey
and Srivastava (1989a), Jackson (1991)]: for any two common knowledge events 91
and 02 that partition , and any pairfi,f 2 E F, we havef E F wheref is defined by
f(O) =fi(0) if 0 C 01 andf(0) =12(0) if 0 E 02.
If every state is a common knowledge event, then we are in effect back to the case
of complete information, and any SCS which satisfies closure is equivalent to an SCR.
For an example of an SCS which does not satisfy closure, suppose = {0, 0'}
278 E. Maskin and T7Sjistronr

where each state is a common knowledge event. The SCS is F = ifi,f2}, where
fi(0) =t2(0') = a, fi(O') =f2(0) = b, and a b. This SCS cannot be implemented.
Indeed, to implement F we would in effect need both a and b to be Nash equilibrium
outcomes in both states, but then there would be no way to guarantee that the outcomes
in the two states are different, as required by both fi and f2. Notice that F is not
equivalent to the constant SCR F defined by F(O) = F(O') = a, b}, since F does not
incorporate the requirement that there be a different outcome in the two states.

5.3. Incentive compatibility

An SCFf is incentive compatible if and only if for all i C N and all 0i, QO
E oi,

E{ui(f (,,O , ) I0 > E{u (f (0, ,i'), ,) I O)}.

An SCS F is incentive compatible if and only if each f F is incentive

47
compatible .

Theorem 22: [Dasgupta, Hammond and Maskin (1979), Myerson (1979), Harris
and Townsend (1981)]. If the SCS F is implementable in Bayesian Nash equilibria,
then F is incentive compatible.
Proof: Suppose F = (M, h) implements F, but somef G F is not incentive compatible.
Then there is i E N and Oi, O E O, such that

E (u, (f(0), 0) I c0} < E {ui (f (0i, O) , ,) Oi}, (8)

where 0 = (Oi,Oi). Let a E BNE r be such that h o a =f. If agent i's type Oi uses the
equilibrium strategy oi(0,), his expected payoff is

E {u, (h(o(0O)), 0,) I Oi} = E {u, (f(O), Oi) I ,o}. (9)

If instead he were to send the message m' = oai(O), he would get

E Mui (h (oi (t , i (C))) 0i} = E{ u(f (0-,, ') , i) I 0.i} (10)

But Inequality (8) and Inequalities (9) and (10) contradict the definition of Bayesian
Nash equilibrium. []
A mechanism F is a revelation mechanism if each agent's message is an announcement
of his own type: Mi = i0for all i e N. Theorem 22 implies the revelation principle: if
F is implementable, then for eachf E F, truth telling is a Bayesian Nash equilibrium

47 The terminology Bayesian incentive conspatibilitv may be used to distinguished this condition from
dominant-strategy incentive compatibility (strategy-proofness).
Ch. 5: Implementation Theory 279

for the revelation mechanism (M, h) where Mi = Oi for each i E N and h =f. However,
the revelation mechanism will in general have untruthful Bayesian Nash equilibria
and will therefore not fully implement [Postlewaite and Schmeidler (1986), Repullo
(1986)].

5.4. Bayesian monotonicity

A deception for agent i is a function a: Oi -i O. A deception a = (al, ... , an)
consists of a deception a for each agent i. Let a(O) (a(0 1), ... , a,(O,)) and
a_(i) _ (aj(01), ... , al(,
aai( ia+j(0i)+l)
..., a,(0,)). The following definition
is due to Jackson (1991), and is slightly stronger than the version given by Palfrey and
Srivastava (1989a) 48.
Definition. Bayesian monotonicity: For all f E F and all deceptions a such that
f o a F, there exist i E N and a function y: O-i - A such that

E {ui (f (0-_,, 0,), ) I i} > E ui (y (i), ) I i}, (11)

for all 0i C i, and

E {ui (f (a (-, O,')) , O') O} <E {ui (y(a i (Oi)), O') O} , (12)

for some 0,'C ,.

When agents have complete information, monotonicity guarantees that a mechanism
can be built which has no undesirable Nash equilibria (compare the discussion in the
first paragraph of Section 4.6). As the proof of Theorem 23 will make clear, Bayesian
monotonicity plays exactly the same role in incomplete information environments.
A related condition called selective elimination was introduced by Mookherjee and
Reichelstein (1990).
Theorem 23: [Postlewaite and Schmeidler (1986), Palfrey and Srivastava (1989a),
Jackson (1991)]. If the SCS F is implementable in Bayesian Nash equilibria, then F
is Bayesian monotonic.
Proof: Suppose a mechanism F = (M, h) implements F in Bayesian Nash equilibria.
For eachf E F there is oa BNE r such that h o a =f. Let a be a deception such that

48 Palfrey and Srivastava (1989a) considered a different model of incomplete information. In their model,
each agent observes an event (a set of states containing the true state). A set of events are compatible
if they have non-empty intersection. Social choice functions only recommend outcomes for situations
where the agents have observed compatible events. The social planner can respond to incompatible
reports any way she wants, which (at least in economic environments) makes it easy to deter the agents
from sending incompatible reports. Thus, Palfrey and Srivastava (1989a) found it sufficient to restrict
their monotonicity condition to "compatible deceptions".
280 E. Maskinz and Sj~strni

f o a F. Now, a o a e Z is a strategy profile such that in state 0 O the agents

behave as they would under a if their types were a(0), i.e., they send message profile
(a o a)(0) = (a(O)). Since h o (a o a) = f o a F, implementation requires that
a o a ( BNE r . If a o a is not a Bayesian Nash equilibrium, then some type ,' E Oi
prefers to deviate to some message m' E Mi. That is,

E {ui (h ( (a (0-i, 0o))), Of) I O[} < E {ui (h ( i (a i (0-)),m) , O,[) O}.
(13)
Let y: O_-i A be defined by y(&Oi) = h(ao_i(O_), m'). Note that

y (a-i (i)) = h (-i (a i ( i)), m) .

Now Inequality (12) follows from Inequality (13). Moreover, Inequality (11) must hold
for each type i0e Oi by definition of Bayesian Nash equilibrium: when is played,
each type 0i E Oi prefers to send message ai(Oi) rather than deviating to m'. ]
Thus, the three conditions of closure, Bayesian monotonicity and incentive compat-
ibility are necessary for Bayesian Nash implementation. Conversely, Jackson (1991)
showed that in economic environments with n > 3, any SCS satisfying these three
conditions can be Bayesian Nash-implemented. This improved on two earlier results
for economic environments with n > 3: Postlewaite and Schmeidler (1986) proved the
sufficiency of closure and Bayesian monotonicity when information is non-exclusive49,
and Palfrey and Srivastava (1989a) proved the sufficiency of closure together with their
version of Bayesian monotonicity and a stronger incentive compatibility condition.
For general environments with n > 3, Jackson (1991) shows that closure, incentive
compatibility, and a condition called monotonicity-no-veto together are sufficient
for Bayesian Nash implementation. The monotonicity-no-veto condition combines
Bayesian monotonicity with no veto power. Dutta and Sen (1994) give an example
of a Bayesian Nash-implementable SCF which violates monotonicity-no-veto. Even
though there are only two alternatives and two possible types for each agent, any
mechanism which implements their SCF must have an infinite number of messages
for each agent.
Matsushima (1993) has shown that Bayesian monotonicity is a very weak condition
if utility functions are quasi-linear and lotteries are available. In other environments,
refinements can enlarge the set of implementable social choice functions.
Palfrey and Srivastava (1989b) showed that any incentive compatible SCF can be
implemented in undominated Bayesian Nash equilibria if n > 3, value distinction and
a full support assumption hold, and no agent is ever indifferent across all alternatives.
For virtual Bayesian implementation see Abreu and Matsushima (1990), Duggan

49Information is non-exclusive if each agent's information can be inferred by pooling the other
n 1 agents' information. Palfrey and Srivastava (1987) discuss the implementability of well-known
SCRs when information is non-exclusive.
Ch. 5: Implementation Theory 281

(1997) and Serrano and Vohra (2001). For Bayesian implementation using sequential
mechanisms see Baliga (1999), Bergin and Sen (1998) and Brusco (1995).

5.5. Non-parametric, robust andfault tolerant implementation

Most of the literature on Bayesian implementation assumes that the social planner who
designs the mechanism knows the agents' common prior p. If she does not have this
information, then the mechanism must be non-parametric in the sense that it cannot
depend directly on p. However, the planner may be able to extract information about p
by adding a stage where the agents report their beliefs. Choi and Kim (1999) construct
such a mechanism for implementation in undominated Bayesian Nash equilibrium.
They assume the agents' types are independently drawn from a distribution which is
known to the agents but not to the social planner. In equilibrium, each agent truthfully
reports his own beliefs as well as the beliefs of a "neighbor". Duggan and Roberts
(1997) assume the social planner makes a prior point estimate ofp, but implementation
is required to be robust against small errors in this estimate.
A different kind of robustness was introduced by Corch6n and Ortufio-Ortin (1995),
who assumed agents are divided into local communities, each with at least three
members. The social planner knows that information is complete within a community,
but she does not necessarily know what agents in one community know about members
of other communities. Implementation should be robust against different possible inter-
community information structures. Yamato (1994) showed that an SCR is robustly
implementable in this sense if and only if it is Nash-implementable.
Eliaz (2000) introducedfault tolerant implementation. The idea is that mechanisms
ought not to break down if there are a few "faulty" agents who do not understand the
rules of the game or make mistakes. Neither the social planner nor the (non-faulty)
agents know which agent (if any) is faulty, but all other aspects of the true state are
known to the (non-faulty) agents. A Nash equilibrium is k-fault tolerant if it is robust
against deviations by at most k faulty players. When n > 2(k + 1), any SCR that
satisfies no veto power and a condition called k-monotonicity can be implemented in
a fault tolerant way.

6. Concluding remarks

Many of the mechanisms exhibited in this survey are admittedly somewhat abstract
and complicated. Indeed, implementation theory has sometimes been criticized for how
different its mechanisms often seem from the simple allocation procedures - such as
auctions - used in everyday life.
In our view, however, these criticisms are somewhat misplaced. The fundamental
objective of this literature is to characterize which social choice rules are in principle
implementable. In other words, the idea is to define the perimeter of the implementable
set. Although considerations such as "simplicity" or "practicability" are undeniably
282 E. Maskin and T Sjstro6m

important, they will not even arise if the SCR in question is outside this set. Of course,
once theoretical implementability has been established, the search for mechanisms with
particular desirable properties can begin.
Relatedly, a major reason why many mechanisms in the implementation literature
are so "complex" is that they are deliberately devised to work very generally. That is,
they are constructed to implement a huge array of social choice rules in environments
with little restriction. For example, the mechanism devised in the proof of Theorem 2
implements any monotonic SCR satisfying no-veto-power in a completely general
social choice setting. Not surprisingly, one can ordinarily exploit the particular
structure that derives from focusing on a particular SCR in a particular environment
[a classic example is Schmeidler's (1980) simple implementation of the Walrasian rule
in an economic environment].
In fact, we anticipate that, since so much has now been accomplished toward
developing implementation theory at a general level, future efforts are likely to
concentrate more on concrete applications of the theory, e.g., to contracts [see, for
instance, Maskin and Tirole (1999)] or to externalities [see, for instance, Varian
(1994)], where special structures will loom large.
Another direction in which we expect the literature to develop is that of bounded
rationality. Most of implementation theory relies quite strongly on rationality: not
only must agents be rational, but rationality must be common knowledge. It would
be desirable to develop mechanisms that are more forgiving of at least limited
departures from full-blown homo game theoreticus. The "fault tolerant" concept (see
Section 5.5) developed by Eliaz (2000) is a step in that direction, but many other
possible allowances for irrationalities could well be considered.
Finally, it would be worthwhile to allow for the possibility that agents have
preferences over more than just the outcomes of a mechanism (i.e., that they care also
about what transpires during the play of the mechanism). An interesting step along
this line has been taken by Glazer and Rubinstein (1998).

References

Abreu, D., and H. Matsushima (1990), "Virtual implementation in iteratively undominated strategies:
incomplete information", Mimeo (Princeton University).
Abreu, D., and H. Matsushima (1992a), "Virtual implementation in iteratively undominated strategies:
complete information", Econometrica 60:993-1008.
Abreu, D., and H. Matsushima (1992b), "A response to Glazer and Rosenthal", Econometrica 60:
1439-1442.
Abreu, D., and H. Matsushima (1994), "Exact implementation", Journal of Economic Theory 64:1-19.
Abreu, D., and Arunava Sen (1990), "Subgame perfect implementation: a necessary and sufficient
condition", Journal of Economic Theory 50:285-299.
Abreu, D., and Arunava Sen (1991), "Virtual implementation in Nash equilibria", Econometrica 59:
997-1022.
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York).
Ch. 5: Implementation Theory 283

Baliga, S. (1999), "Implementation in incomplete information environments: the use of multi-stage

games", Games and Economic Behavior 27:173-183.
Baliga, S., and S. Brusco (2000), "Collusion, renegotiation and implementation", Social Choice and
Welfare 17:69-83.
Baliga, S., and T. Sjdstrdm (1999), "Interactive implementation", Games and Economic Behavior 27:
38-63.
Baliga, S., L. Corch6n and T. Sjbstrdm (1997), "The theory of implementation when the planner is a
player", Journal of Economic Theory 77:15-33.
Bergin, J., and Arunava Sen (1998), "Extensive form implementation in incomplete information
environments", Journal of Economic Theory 80:222 256.
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press).
Bowen, H. (1943), "The interpretation of voting in the allocation of economic resources", Quarterly
Journal of Economics 58:27-48.
Brusco, S. (1995), "Perfect Bayesian implementation", Economic Theory 5:429-444.
Cabrales, A. (1999), "Adaptive dynamics and the implementation problem with complete information",
Journal of Economic Theory 86:159-184.
Cabrales, A., and G. Ponti (2000), "Implementation, elimination of weakly dominated strategies and
evolutionary dynamics", Review of Economic Dynamics 3:247-282.
Chakravorty, B. (1991), "Strategy space reduction for feasible implementation of Walrasian performance",
Social Choice and Welfare 8:235-245.
Chakravorty, B., L. Corch6n and S. Wilkie (1997), "Credible implementation", Games and Economic
Behavior, forthcoming.
Choi, J., and T. Kim (1999), "A nonparametric, efficient public decision mechanism: undominated
Bayesian Nash implementation", Games and Economic Behavior 27:64-85.
Clarke, E.H. (1971), "Multipart pricing of public goods;' Public Choice 11:17-33.
Corch6n, L. (1996), The Theory of Implementation of Socially Optimal Decisions in Economics (St.
Martin's Press, New York).
Corch6n, L., and I. Ortuio-Ortin (1995), "Robust implementation under alternative information
structures", Economic Design 1:159-171.
Danilov, V (1992), "Implementation via Nash equilibria", Econometrica 60:43-56.
Dasgupta, P., PJ. Hammond and E. Maskin (1979), "The implementation of social choice rules: some
general results on incentive compatibility", Review of Economic Studies 46:185-216.
d'Aspremont, C., and L.A. G6rard-Varet (1979), "Incentives and incomplete information", Journal of
Public Economics 11:25-45.
de Trenqualye, P. (1988), "Stability of the Groves and Ledyard mechanism", Journal of Economic Theory
46:164-171.
Deb, R. (1994), "Waiver, effectivity, and rights as game forms", Economica 61:167-178.
Deb, R., P.K. Pattanaik and L. Razzolini (1997), "Game forms, rights and the efficiency of social
outcomes", Journal of Economic Theory 72:74-95.
Duggan, J. (1997), "Virtual Bayesian implementation", Econometrica 67:1175-1199.
Duggan, J., and J. Roberts (1997), "Robust implementation", Mimeo (University of Rochester, NY).
Dummett, M., and R. Farquharson (1961), "Stability in voting", Econometrica 29:33-43.
Dutta, B., and Arunava Sen (1991a), "Implementation under strong equilibria: a complete
characterization", Journal of Mathematical Economics 20:49-67.
Dutta, B., and Arunava Sen (1991b), "A necessary and sufficient condition for two-person Nash
implementation", Review of Economic Studies 58:121-128.
Dutta, B., and Arunava Sen (1994), "Bayesian implementation: the necessity of infinite mechanisms",
Journal of Economic Theory 64:130-141.
Dutta, B., Arunava Sen and R. Vohra (1995), "Nash implementation through elementary mechanisms in
economic environments", Economic Design 1:173-204.
Eliaz, K. (2000), "Fault tolerant implementation", Review of Economic Studies, forthcoming.
284 E. Maskin and Sjdstr6m

Farquharson, R. (1969), The Theory of Voting (Yale University Press).

Farrell, J. (1993), "Meaning and credibility in cheap-talk games", Games and
Economic Behavior
5:514-531.
Fudenberg, D., and D. Levine (1998), The Theory of Learning in Games (MIT
Press, Cambridge, MA).
Gaertner, W., PK. Pattanaik and K. Suzumura (1992), "Individual rights revisited",
Economica 59:
161-177.
Grdenfors, P. (1981), "Rights, games and social choice", Nous 15:341-356.
Gaspart, F. (1996), "Fair implementation in the cooperative production problem:
two properties of normal
form mechanisms", Mimeo (FUND Namur, Belgium).
Gaspart, F (1997), "A general concept of procedural fairness for one-stage
implementation", Mimeo
(FUND Namur, Belgium).
Gibbard, A. (1973), "Manipulation of voting schemes: a general result", Econometrica
41:587-601.
Glazer, J., and R. Rosenthal (1992), "A note on Abreu-Matsushima
mechanisms", Econometrica 60:
1435-1438.
Glazer, J., and A. Rubinstein (1996), "An extensive game as a guide for solving
a normal game", Journal
of Economic Theory 70:32-42.
Glazer, J., and A. Rubinstein (1998), "Motives and implementation: on the design
of mechanisms to
elicit options", Journal of Economic Theory 79:157-173.
Green, J., and J.J. Laffont (1979), Incentives in Public Decision Making
(North-Holland, Amsterdam).
Grossman, S., and M. Perry (1986), "Perfect sequential equilibrium", Journal
of Economic Theory
39:97-119.
Groves, T. (1970), "The allocation of resources under uncertainty", Ph.D.
dissertation (University of
California, Berkeley).
Groves, T., and J. Ledyard (1977), "Optimal allocation of public goods:
a solution to the 'free rider'
dilemma", Econometrica 45:783-811.
Groves, T., and J. Ledyard (1987), "Incentive compatibility since 1972", in:
T. Groves, R. Radner and
S. Reiter, eds., Information, Incentives and Economic Mechanisms (University
of Minnesota Press)
pp. 48-111.
Groves, T., and M. Loeb (1975), "Incentives and Public Inputs", Journal of Public
Economics 4:311-326.
Hammond, P.J. (1997), "Game forms versus social choice rules as models of
rights", in: K.J. Arrow,
A.K. Sen and K. Suzumura, eds., Social Choice Re-examined, Vol. 2 (Macmillan,
London) pp. 82-95.
Harris, M., and R. Townsend (1981), "Resource allocation with asymmetric
information", Econometrica
49:33-64.
Harsanyi, J.C., and R. Selten (1988), A General Theory of Equilibrium Selection
in Games (MIT Press,
Cambridge, MA).
Hayek, F. (1945), "The use of knowledge in society", American Economic
Review 35:519-530.
Herrero, M., and S. Srivastava (1992), "Implementation via backward induction",
Journal of Economic
Theory 56:70-88.
Hong, L. (1995), "Nash implementation in production economies", Economic
Theory 5:401-418.
Hurwicz, L. (1960), "Optimality and informational efficiency in resource
allocation processes", in:
K.J. Arrow, S. Karlin and P Suppes, eds., Mathematical Methods in the
Social Sciences (Stanford
University Press) pp. 27-46.
Hurwicz, L. (1972), "On informationally decentralized systems", in: R.
Radner and C.B. McGuire, eds.,
Decision and Organization (North-Holland, Amsterdam) pp. 297-336.
Hurwicz, L. (1977), "On the dimensional requirements of informationally
decentralized processes", in:
K.J. Arrow and L. Hurwicz, eds., Studies in Resource Allocation Processes
(Cambridge University
Press) pp. 413-424.
Hurwicz, L. (1979a), "Outcome functions yielding Walrasian and Lindahl allocations
at Nash equilibrium
points", Review of Economic Studies 46:217-225.
Hurwicz, L. (1979b), "On allocations attainable through Nash equilibria",
Journal of Economic Theory
21:140-165.
Ch. 5: Implementation Theory 285

Hurwicz, L., and D. Schmeidler (1978), "Construction of outcome functions guaranteeing existence and
Pareto-optimality of Nash equilibria", Econometrica 46:1447-1474.
Hurwicz, L., and M. Walker (1990), "On the generic nonoptimality of dominant-strategy allocation
mechanisms: a general theorem that includes pure exchange economies", Econometrica 58:683-704.
Hurwicz, L., E. Maskin and A. Postlewaite (1995), "Feasible Nash implementation of social choice rules
when the designer does not know endowments or production sets", in: J. Ledyard, ed., The Economics
of Informational Decentralization: Complexity, Efficiency and Stability (Kluwer Academic Publishers,
Amsterdam) pp. 367-433.
Jackson, M. (1991), "Bayesian implementation", Econometrica 59:461-477.
Jackson, M. (1992), "Implementation in undominated strategies: a look at bounded mechanisms", Review
of Economic Studies 59:757-775.
Jackson, M. (2001), "A crash course in implementation theory", Social Choice and Welfare 18:655-708.
Jackson, M., and T. Palfrey (2001), "Voluntary implementation", Journal of Economic Theory 98:1-25.
Jackson, M., T. Palfrey and S. Srivastava (1994), "Undominated Nash implementation in bounded
mechanisms", Games and Economic Behavior 6:474-501.
Jordan, J. (1986), "Instability in the implementation of Walrasian allocations", Journal of Economic
Theory 39:301-328.
Ledyard, J., and J. Roberts (1974), "On the incentive problem with public goods", Discussion Paper 116
(Center for Mathematical Studies in Economics and Management Science, Northwestern University).
Maskin, E. (1979a), "Incentive schemes immune to group manipulation", Mimeo (MIT,
Cambridge, MA).
Maskin, E. (1979b), "Implementation and strong Nash equilibrium", in: J.J. Laffont, ed., Aggregation
and Revelation of Preferences (North-Holland, Amsterdam) pp. 433-440.
Maskin, E. (1985), "The theory of implementation in Nash equilibrium: a survey", in: L. Hurwicz,
D. Schmeidler and H. Sonnenschein, eds., Social Goals and Social Organization (Cambridge University
Press) pp. 173-204.
Maskin, E. (1999), "Nash equilibrium and welfare optimality", Review of Economic Studies 66:23-38.
Maskin, E., and J. Moore (1999), "Implementation and renegotiation", Review of Economic Studies
66:39-56.
Maskin, E., and J. Tirole (1999), "Unforeseen contingencies and incomplete contracts", Review of
Economic Studies 66:83-114.
Matsushima, H. (1988), "A new approach to the implementation problem", Journal of Economic Theory
45:128-144.
Matsushima, H. (1993), "Bayesian monotonicity with side payments", Journal of Economic Theory
39:107-121.
McKelvey, R.D. (1989), "Game forms for Nash implementation of general social choice
correspondences", Social Choice and Welfare 6:139 156.
Mookherjee, D., and S. Reichelstein (1990), "Implementation via augmented revelation mechanisms",
Review of Economic Studies 57:453-476.
Moore, J. (1992), "Implementation, contracts and renegotiation in environments with complete
information", in: J.J. Laffont, ed., Advances in Economic Theory, Vol. 1 (Cambridge University
Press) pp. 182-282.
Moore, J., and R. Repullo (1988), "Subgame perfect implementation", Econometrica 56:1191-1220.
Moore, J., and R. Repullo (1990), "Nash implementation: a full characterization", Econometrica 58:
1083-1100.
Moulin, H. (1979), "Dominance solvable voting schemes", Econometrica 47:1337-1352.
Moulin, H. (1983), The Strategy of Social Choice (North-Holland, Amsterdam).
Moulin, H., and B. Peleg (1982), "Cores of effectivity functions and implementation theory", Journal
of Mathematical Economics 10:115-145.
Mount, K., and S. Reiter (1974), "The informational size of message spaces", Journal of Economic
Theory 8:161-192.
286 E. Maskin and T Sjstrmn

Muench, T., and M. Walker (1984), "Are Groves Ledyard equilibria attainable?", Review of Economic
Studies 50:393-396.
Muller, E., and M.A. Satterthwaite (1977), "The equivalence of strong positive association and strategy
proofness", Journal of Economic Theory 14:412-418.
Myerson, R. (1979), "Incentive compatibility and the bargaining problem", Econometrica 47:61-74.
Palfrey, T. (1992), "Implementation in Bayesian equilibrium: the multiple equilibrium problem in
mechanism design", in: J.J. Laffont, ed., Advances in Economic Theory, Vol. 1 (Cambridge University
Press) pp. 283-323.
Palfrey, T. (2001), "Implementation theory", in: R. Aumann and S. Hart, eds., Handbook of Game
Theory, Vol. 3 (North-Holland, Amsterdam).
Palfrey, T., and S. Srivastava (1987), "On Bayesian implementable allocations", Review of Economic
Studies 54:193-208.
Palfrey, T., and S. Srivastava (1989a), "Implementation with incomplete information in exchange
economies", Econometrica 57:115 134.
Palfrey, T., and S. Srivastava (1989b), "Mechanism design with incomplete information: a solution to
the implementation problem", Journal of Political Economy 97:668-691.
Palfrey, T., and S. Srivastava (1991), "Nash-implementation using undominated strategies", Econometrica
59:479-501.
Peleg, B. (1984), Game Theoretic Analysis of Voting in Committees (Cambridge University Press).
Peleg, B. (1998), "Effectivity functions, game forms, games and rights", Social Choice and Welfare
15:67-80.
Postlewaite, A., and D. Schmeidler (1986), "Implementation in differential information economies",
Journal of Economic Theory 39:14 33.
Postlewaite, A., and D. Wettstein (1989), "Continuous and feasible implementation", Review of Economic
Studies 56:603-611.
Reichelstein, S., and S. Reiter (1988), "Game forms with minimal message spaces", Econometrica
56:661-692.
Repullo, R. (1986), "The revelation principle under complete and incomplete information", in: K. Binmore
and P. Dasgupta, eds., Economic Organization as Games (Basil Blackwell, Oxford) pp. 179-195.
Repullo, R. (1987), "A simple proof of Maskin's theorem on Nash implementation", Social Choice and
Welfare 4:39-41.
Roberts, K.W.S. (1979), "The characterization of implementable choice rules", in: J.J. Laffont, ed.,
Aggregation and Revelation of Preferences (North-Holland, Amsterdam) pp. 321-348.
Rubinstein, A., and A. Wolinsky (1992), "Renegotiation-proof implementation and time preferences",
American Economic Review 82:600-614.
Saijo, T. (1987), "On constant Maskin monotonic social choice functions", Journal of Economic Theory
42:382-386.
Saijo, T. (1988), "Strategy space reduction in Maskin's theorem", Econometrica 56:693-700.
Saijo, T., Y. Tatamitani and T. Yamato (1996), "Toward natural implementation", International Economic
Review 37:949-980.
Samuelson, P.A. (1954), "The pure theory of public expenditure", Review of Economics and Statistics
36:387-389.
Sato, E (1981), "On the informational size of message spaces for resource allocation processes in
economies with public goods", Journal of Economic Theory 24:48-69.
Satterthwaite, M.A. (1975), "Strategy-proofness and Arrow's conditions: existence and correspondence
theorems for voting procedures and social welfare functions", Journal of Economic Theory 10:
187-217.
Satterthwaite, M.A., and H. Sonnenschein (1981), "Strategy-proof allocation mechanisms at differentiable
points", Review of Economic Studies 48:587-597.
Schmeidler, D. (1980), "Walrasian analysis via strategic outcome functions", Econometrica
48:1585-1594.
Ch. 5: Implementation Theory 287

Segal, I., and M. Whinston (2002), "The Mirrlees approach to mechanism design with renegotiation",
Econometrica 70:1-47.
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San Francisco).
Sen, Arunava (1995), "The implementation of social choice functions via social choice correspondences:
a general formulation and a limit result", Social Choice and Welfare 12:277-292.
Serrano, R., and R. Vohra (2001), "Some limitations of virtual Bayesian implementation", Econometrica
69:785-792.
Sj6str6m, T. (1991), "On the necessary and sufficient conditions for Nash implementation", Social Choice
and Welfare 8:333-340.
Sj6str6m, T. (1993), "Implementation in perfect equilibria", Social Choice and Welfare 10:97-106.
Sj6str6m, T. (1994), "Implementation in undominated Nash equilibria without using integer games",
Games and Economic Behavior 6:502-511.
Sj6strm, T. (1996a), "Implementation by demand mechanisms", Economic Design 1:343-354.
Sj6strdm, T. (1996b), "Credibility and renegotiation of outcome functions in economics", Japanese
Economic Review 47:157-169.
Sj6str6m, T. (1999), "Undominated Nash implementation with collusion and renegotiation", Games and
Economic Behavior 26:337 352.
Smith, V (1979), "Incentive compatible experimental processes for the provision of public goods",
in: VL. Smith, ed., Research in Experimental Economics, Vol. 1 (JAI Press, Greenwich, Conn.)
pp. 59 168.
Suh, S.C. (1995), "An algorithm for checking strong Nash-implementability", Journal of Mathematical
Economics 25:109-122.
Suh, S.C. (1996), "Implementation with coalition formation: a complete characterization", Journal of
Mathematical Economics 26:409-428.
Suh, S.C. (1997), "Double implementation in Nash and strong Nash equilibria", Social Choice and
Welfare 14:439-447.
Thomson, W. (1979), "Comment on Hurwicz: On allocations attainable through Nash equilibria",
in: J.J. Laffont, ed., Aggregation and Revelation of Preferences (North-Holland, Amsterdam)
pp. 420-431.
Thomson, W (1996), "Concepts of implementation", Japanese Economic Review 47:133-143.
Thomson, W (1999), "Monotonic extensions on economic domains", Review of Economic Design
4:13-33.
Tian, G. (1989), "Implementation of the Lindahl correspondence by a single-valued, feasible, and
continuous mechanism", Review of Economic Studies 56:613-621.
Varian, H.R. (1994), "A solution to the problem of externalities when agents are well-informed", American
Economic Review 84:1278 1293.
Vartiainen, H. (1999), "Subgame perfect implementation: a full characterization", Mimeo (University of
Helsinki).
Vickrey, WS. (1960), "Utility, strategy and social decision rules", Quarterly Journal of Economics
74:507-535.
Vickrey, W.S. (1961), "Counterspeculation, auctions, and competitive sealed tenders", Journal of Finance
16:8-37.
Walker, M. (1980), "On the nonexistence of a dominant strategy mechanism for making optimal public
decisions", Econometrica 48:1521-1540.
Walker, M. (1981), "A simple incentive compatible mechanism for attaining Lindahl allocations",
Econometrica 49:65-73.
Williams, S. (1986), "Realization and Nash implementation: two aspects of mechanism design",
Econometrica 54:139-151.
Yamato, T. (1992), "On Nash implementation of social choice correspondences", Games and Economic
Behavior 4:484-492.
288 E. Maskin and T Sjdstromn

Yamato, T. (1993), "Double implementation in Nash and undominated Nash equilibrium", Journal of
Economic Theory 59:311-323.
Yamato, T. (1994), "Equivalence of Nash-implementability and robust implementability with incomplete
information", Social Choice and Welfare 11:289-303.
Yamato, T. (1999), "Nash implementation and double implementation: equivalence theorems", Journal
of Mathematical Economics 31:215-238.
Yoshihara, N. (2000), "A characterization of natural and double implementation in production economies",
Social Choice and Welfare 17:571-599.
Zhou, L. (1991), "Inefficiency of strategy-proof allocation mechanisms in pure exchange economies",
Social Choice and Welfare 8:247-254.
Chapter 6

AXIOMATIC COST AND SURPLUS SHARING*

HERVE MOULIN
Rice University

Contents
Abstract 290
Keywords 290
Introduction 291
1. Rationing 294
1.1. The problem and some examples 294
1.2. The proportional method 297
1.3. Uniform Gains and Uniform Losses 298
1.4. The Contested-Garment method and Self-Duality 301
1.5. Consistent and symmetric methods 304
1.6. Equal-sacrifice methods 307
1.7. Asymmetric methods: combining the invariance axioms 309
1.8. Fixed-path methods 312
1.9. Rationing indivisible goods 313
1.10. Two variants of the rationing model 315
1.10.1. Surplus sharing 315
1.10.2. Fair division with single-peaked preferences 316
2. Sharing variable returns 316
2.1. The problem and some examples 316
2.2. Average-cost method 318
2.3. Serial cost sharing 319
2.4. Additive cost sharing 322
2.5. Variants of the model and further axioms 326
2.5.1. Distributivity 326
2.5.2. Negative cost shares and the decreasing serial method 327
2.5.3. Consistency? 328
3. Heterogeneous outputs or inputs 328
3.1. The problem 328

* Detailed comments on an earlier version by R.J. Aumann, C. Herrero, M. Kaminski, K. Suzumura,

and A. Villar, are gratefully acknowledged.
This research is supported by the NSF, under grant SES 0096230.

Handbook of Social Choice and Welfare, Volume i, Edited by KJ Arrow, A.K. Sen andK. Suzumura
© 2002 Elsevier Science B. V All rights reserved
290 H. Moulin

3.2. Binary demands: the Shapley value 330

3.3. Variable demands of indivisible goods: the dr model 335
3.4. Demand Monotonicity in the dr model 339
3.5. Variable demands of divisible goods: the rr model 341
3.6. Unit Invariance and Demand Monotonicity in the rr model 346
3.7. Non-additive cost sharing methods 349
3.7.1. The Stand-Alone core approach 349
3.7.2. Extending homogeneous-good methods 351
3.7.3. More non-additive methods and an open question 353
References 354

Abstract
The equitable division of a joint cost (or a jointly produced output) among agents
with different shares or types of output (or input) commodities, is a central theme of
the theory of cooperative games with transferable utility. Ever since Shapley's seminal
contribution in 1953, this question has generated some of the deepest axiomatic results
of modern macroeconomic theory.
More recently, the simpler problem of rationing a single commodity according to
a profile of claims (reflecting individual needs, or demands, or liabilities) has been
another fertile ground for axiomatic analysis. This rationing model is often called the
bankruptcy problem in the literature.
This chapter reviews the normative literature on these two models, and emphasizes
their deep structural link via the Additivity axiom for cost sharing: individual
cost shares depend additively upon the cost function. Loosely speaking, an additive
cost-sharing method can be written as the integral of a rationing method, and this
representation defines a linear isomorphism between additive cost-sharing methods and
rationing methods.
The simple proportionality rule in rationing thus corresponds to average cost pricing
and to the Aumann-Shapley pricing method (respectively for homogeneous or
heterogeneous output commodities). The uniform rationing rule, equalizing individual
shares subject to the claim being an upper bound, corresponds to serial cost sharing.
And random priority rationing corresponds to the Shapley-Shubik method, applying
the Shapley formula to the Stand Alone costs.
Several open problems are included. The axiomatic discussion of non-additive
methods to share joint costs appears to be a promising direction for future research.

Keywords
cost sharing, axiomatic, rationing, Shapley value, additivity

JEL classification: C71, D62, D63

Ch. 6: Axiomatic Cost and Surplus Sharing 291

Introduction

The oldest formal principle of distributive justice is, without a doubt, Aristotle's
celebrated maxim:
Equals should be treated equally, and unequals, unequally in proportion to relevant similarities
and differences
[in the modern rendition by the social psychology literature, see, e.g., Deutsch (1985)].

Inspired by the axiomatic approach to the theory of cooperative games (initiated

in 1953 by Shapley's seminal contribution), a considerable research effort explores
the logical limits of the old maxim, within a small number of simple models of fair
division. All such models stage a given production technology and a given set of users
of the technology. Individual users influence the production plan in different ways,
either by demanding different quantities of output, or contributing different quantities
of input, or both. When individual demands (or contributions) are homogeneous (total
demand is simply the sum of individual demands) and the technology has constant
returns to scale, the fair distribution of inputs and outputs among users can and
should simply follow Aristotle's proportionality principle. The logical challenge is
to deal with variable returns of the technology and heterogeneity of the individual
demands/contributions.
This survey of the theory of cooperative production is organized around three basic
models, to which most of the literature is devoted. First we discuss the rationingmodel
(Section 1), where a given amount of resources (e.g., money) must be divided among
beneficiaries with unequal claims on the resources. In this very bare model, the only
available information about the technology is a single point of the production set. Then
we look at a one-input-one-output technology (Section 2), where all users consume
(possibly different amounts of) a homogeneous output commodity, and contribute
(possibly different amounts of) a homogeneous input commodity: a typical problem
specifies a list of individual demands of output (resp. input contributions) and the entire
production function (with variable returns), and asks to divide fairly the corresponding
input cost (resp. output produced). In the third model we assume a technology with
a single homogeneous input and one heterogeneous output per agent (resp. one
homogeneous output and one heterogeneous input per agent), and we speak of the
heterogeneous goods model (Section 3). The formal definition of the cooperative
production problem is the same as in the homogeneous-good models of Section 2:
it consists of a list of heterogeneous demands and the entire production function.
The question is to divide fairly the corresponding total input cost (respectively to
divide the output produced, given the list of heterogeneous input contributions and
the technology).
One important feature in the model is whether the input or output commodities come
in indivisible units or are infinitely divisible: we speak of a discrete or real variable,
respectively. Both versions are meaningful and important in applications: the goods on
demand may be cars (discrete) or length of a runway (real); individual contributions
292 H. oulin

may come in days on the job (discrete) or cash (real) and so on. Each one of the
three models involves two kinds of variables: the exogenously given claims/demands
(rationing model, Section 1), demands of output or contributions of input (Sections 2
and 3) on the one hand, and the endogenously determined shares of the resources
(Section 1) or cost shares or output shares (Sections 2 and 3) on the other hand. Each
kind of variable can be either discrete (d) or real (r). In the rationing model (Section 1)
the main model is of the "rr" type (the exogenous and endogenous variables are both
real) but we also discuss the dd model (both kinds of variables are discrete). In the cost-
and surplus-sharing models both the rr model and the dr model (exogenous variable
discrete, endogenous variable real) play an important role; for instance the classical
model of cooperative games is of type dr, but the theory of Aumann-Shapley pricing
happens in the rr world.
Besides the issue of realism, the choice between a discrete or a real model involves
a familiar trade-off. A discrete model is mathematically much simpler, as it typically
involves no topological difficulty. For instance, in the dd model the set of possible
(rationing or cost-sharing) methods is essentially finite; in the dr model, a typical cost-
sharing method is a linear operator on a finite-dimensional space (see Section 3); in
the rr method a cost-sharing model is a linear operator on a functional space. On the
other hand, in the discrete model even the basic proportionality principle mentioned
at the outset is hard to write: it can be approximated at best.
Some comments about the type of axioms we impose are in order. A few of
them convey a simple idea of equity: of this type are Equal Treatment of Equals
and the crucial Dummy axiom in the heterogeneous-goods model, expressing some
notion of reward. There are also some incentive compatibility requirements, such
as No Advantageous Reallocation in Sections 1 and 2 or Demand Monotonicity in
Section 3. For the latter axiom, the equity and incentive-compatibility interpretations
coexist and reinforce each other.
Yet the main axiomatic tools throughout the survey are driven neither by equity nor
incentives. They are properties of structural invariance expressing the commutativity
of the allocation method with respect to certain variations in the cost- or surplus-
sharing problem under scrutiny. For instance, Consistency, the leading axiom in
Section 1, requires the method to commute with a variation in the society of agents
concerned. Additivity, by far the most important axiom in Sections 2 and 3, is
commutativity of the cost-sharing method with respect to the sum of cost functions.
Scale Invariance and Unit Invariance (Section 3) are about changing the unit in which
a particular good is measured. And so on.
The structural-invariance axioms are the powertools of the mathematical analysis,
the backbone of the most interesting characterization results. The most spectacular
example is the Additivity axiom in Sections 2 and 3. The entire set 7 of
rationing methods studied in Section 1 is shown to be linearly isomorphic to the set
of additive cost-sharing methods in the homogeneous-good model (Theorem 2.2) and
isomorphic to the extreme points of the (convex) set of additive cost-sharing methods
in the heterogeneous-goods model (Theorems 3.1 and 3.3). This double isomorphism
Ch. 6: Axiomatic Cost and Surplus Sharing 293

allows us to follow the "same" allocation method in the three different models: the
proportional rationing method (Section 1) becomes average cost sharing in Section 2
and the Aumann-Shapley method in Section 3; the uniform-gains method in Section 1
becomes serial cost sharing in Sections 2 and 3; priority rationing (Section 1) becomes
incremental cost sharing in Sections 2 and 3. New methods emerge as well: uniform
losses rationing suggests the dual serial cost-sharing method.
The main lesson to be learned from this overview is that the powerful structural
invariance axioms are double-edged swords. For instance, Additivity with respect to
cost functions implies an isomorphism between cost-sharing and rationing methods,
but it also severely limits the choices open to the mechanism designer. When a
structural-invariance axiom such as Additivity conflicts with a set of reasonable equity
and/or incentives requirements, we feel that the invariance axiom must be the first to go.
This opens up the question of finding a less restrictive version of the invariance axiom,
for which the impossibility result becomes a limited-possibility result. This leads
us to the new and yet hardly explored space of non-additive cost-sharing methods:
Section 3.7.

Relation to other chapters in the Handbook. In the current chapter, we view

the users of the technology as entirely passive: they have inelastic demands of output,
or input contributions, or claims. The axiomatic analysis is supposed to enlighten a
benevolent dictator on the possible interpretations of fairness when dividing cost or
output, or whatever resources must be split among the participants. Another approach
sees the users as rational microeconomic agents, endowed with classical preferences
and choosing independently and strategically the amount of output they want to
consume or the amount of input they choose to contribute. Any given division method
(whether or not it is fair in the sense of this chapter) yields a specific noncooperative
demand game (resp. input-contribution game) where a user's cost share depends
on the entire profile of demands (resp. his output share depends on the profile of
contributions). In this view a division method is a decentralization device: each
user knows his own preferences but may be completely unaware of other users'
preferences. The general results of Chapter 5 of this volume on mechanism design and
of Chapter 23 of Volume 2 on strategy-proofness become then relevant. In particular,
the incentives properties of the uniform-gains rationing method (and more generally of
the fixed-path methods: see Section 1.8) are reinterpreted in Chapter 23 of Volume 2
in the context of the fair-division problem with single-peaked preferences: there they
mean that the direct revelation of preferences is a strategy-proof mechanism. Similarly,
serial cost-sharing (and more generally the fixed-path generated methods in Section 3)
gives rise to a strategy-proof social choice function whenever the cost function is
supermodular.
The second important link is with Chapter 26 of Volume 2, on fair allocation,
and Chapter 20 of Volume 2, on fair compensation. A third way to look at the
cooperative production problem is as a special instance of the social-choice problem
in a particular economic environment. The social planner takes into account the
294 H. oulin

technology (production set) and the whole profile of individual preferences, then selects
a first-best (Pareto-efficient) allocation that he deems optimal. One way to do so is
by defining a full-fledged social preference over the set of feasible allocations in the
economy: Chapter 16 of Volume 2 explains why this approach will lead, in most
models, to a conceptual dead-end in classical Arrowian fashion. An alternative route
is to simply select one (efficient) allocation by means of fairness axioms: this is the
route taken in Chapter 26 of Volume 2 for a general family of allocation problems
that includes cooperative production; this is also the approach taken in Chapter 20 of
Volume 2 for a family of models very close to our homogeneous-good problems. The
main difference is that in the first-best approach the social choice function selects the
shares of output as well as the shares of input from the profile of individual preferences:
therefore the profile of cost shares, say, depends on more than the profile of demands
and the cost function, and a formula such as the Shapley-Shubik method is generally
not relevant.

1. Rationing

1.1. The problem and some examples

A rationingproblemis a triple (N, t, x) where N is a finite set of agents, the nonnegative
real number t represents the amount of resources to be divided, the vector x = (xi) C A'
specifies for each agent i a claim xi, and these numbers are such that

0 < xi for all i; 0<t< E xi.

A solution to the rationing problem is a vector y (i)i IV, specifying a share yi for
each agent i and such that

0 < y < xi for all i; ZE y = t.

ieN

The crucial inequality yi < xi may not be meaningful if claims are subjective
evaluations of needs (or responsibility): an agent may underestimate his "objective"
need (or responsibility), prompting the social planner to violate the above inequality.
Our model ignores this possibility: thus it is the most convincing when the claims xi
are "objectively" measured, as in the case of a contractual debt.
Several of the axiomatic properties of rationing methods pertain to variations in the
population (also called society) N of concerned agents. See the merging properties
in Section 1.2 and the consistency property playing the leading role from Section 1.3
onward. Therefore the formal model must specify the set A of potential agents from
which a certain subset N is selected to generate an actual problem. In general, A/'could
be finite or infinite, although the "real" society N is always finite. One exception is
Ch. 6: Axiomatic Cost and Surplus Sharing 295

the discussion of symmetric and consistent rationing methods in Section 1.4: there we
must assume that the set iV is infinite.
It is neither easy nor necessary at this stage to interpret a rationing problem directly
as a model of cooperative production. The link of the rationing model with cooperative
production will become apparent in Section 2, when we discuss the implications of
the powerful Additivity axiom in the homogeneous-goods model of cost and surplus
sharing: see Theorem 2.2.
Inheritance problems provide the oldest example on record of the rationing problem
[see O'Neill (1982) and Rabinovich (1973) borrowing examples from the Babylonian
Talmud]: here t is the liquidation value of the bankrupt firm and xi is the debt owed
to creditor i [see Aumann and Maschler (1985)].
Taxation is another important example: now t is total tax to be levied and xi is
agent i's fiscal liability [see Young (1988, 1990)]. Note that in a taxation example, the
resources to be divided are a "bad", whereas they are a "good" in a inheritance or
bankruptcy story. A microeconomic example similar to taxation is the cost sharing of
an indivisible public good: t is the cost of the good and xi is the benefit to agent i.
Rationing occurs in markets where the price of a commodity is fixed (for instance,
at zero): t is the available supply and xi is agent i's demand of good i [Benassy
(1982), Dr&ze (1975)]. Medical triage is an example: t measures the available medical
resources and xi is the quantity needed by agent i for full treatment [Winslow (1992)].
Rationing food among refugees is similar: xi measures a nutritional need, and t the
nutritional value of the available food. The supply-chain problem is a management
example: the central supplier collects orders from its retailers and can not meet all
demands at once [Cachon and Lariviere (1996)].
Often the resources to be divided come in indivisible units: organs for transplants,
seats in crowded airplanes or in popular sports events, visas to potential immigrants
[Elster (1992)] as well as cars allocated by General Motors to its car dealers. In this
case xi and t are integers, e.g., in the case of organs or visas xi can only be 0 or 1.
Important examples where xi and t are integers come from queuing and scheduling: a
server can process one job per unit of time and agent i requests xi jobs; at any time t
such that t < E xi, the service protocol solves a rationing problem.
A rationing method r associates to any rationing problem (N,t,x) a solution
y = r(N, t, x). We study rationing methods with the axiomatic methodology. In the main
model, that we call the rr model, all variables t,xi,yi vary over the nonnegative real
line, and correspond to divisible resources, claims, demands, etc. Sections 1.2 to 1.8
are devoted to the rr model. In Section 1.9 we study the dd model where t,xi,y are
all nonnegative integers.
We denote by T1 the set of rationing methods with a given potential population Af.
When required for clarity, we indicate whether claims and shares are discrete or real
variables, e.g., lZdd means that claims are integer-valued and shares are real-valued.
296 H. Moulin

All rationing methods discussed below satisfy the following property:

Resource Monotonicity (RM):

{t < t'} X {r(N, t,x) < r(N, t',x)} for all N, t, t' and x.

This is a mild and compelling requirement: when resources (whether desirable or not)
increase, no one should see his share reduced. In most of the results below, Resource
Monotonicity needs not be assumed and follows from the other axioms [e.g., Upper
Composition (1.4) implies RM]; exceptions are Theorems 1.4 and 1.6.
With the exception of Section 1.7, all methods are equitable in the sense that they do
not discriminate a priori between the agents. This corresponds to two familiar axioms:

Equal Treatment of Equals (ETE):

xi =x j = yi =yj for all N,x, t and all i,j E I

Symmetry:
y = r(N, t, x) is a symmetric function of the variables xi, i N.

Note that Symmetry implies Equal Treatment of Equals.

An important operation is the duality operator transforming gains into losses. If r
is a rationing method, its dual method r* is defined as

r*(N, t,x) = x- r(N,xN - t,x) for all N, t,x

(where we use the notation XN = Ei V Xi). Given x, the method r* allocates t units
of "gains" exactly as r allocates the corresponding losses (XN - t).
Overview of Section 1. The proportional rationing method is characterized
in Section 1.2 by the property that it treats claims as anonymous transferable
"bonds". In Section 1.3 we discuss two important methods equalizing respectively
the gains and losses on individual claims, and we introduce the Upper and Lower
Composition axioms. The celebrated Contested Garment method inspired by a
bargaining interpretation of the rationing model is the subject of Section 1.4. The next
two sections focus on the structural-invariance axiom called Consistency, leading to
the characterization of parametric methods in Section 1.5, and of the equal-sacrifice
methods in Section 1.6. Section 1.7 discusses the rich family of asymmetric methods
meeting Consistency Upper and Lower Composition. Fixed-path methods, discussed
in Section 1.8, are another family of asymmetric methods, that play a crucial role in
Section 3. The probabilistic rationing of indivisible goods is the discrete variant of
the rationing model: see Section 1.9. Finally, Section 1.10 discusses the variant of the
rationing problem where the available resources may exceed the sum of individual
claims.
Ch. 6: Axiomatic Cost and Surplus Sharing 297

1.2. The proportionalmethod

With the exception of Section 1.9, all variables t, xi and yi are real numbers: we are
in the rr model. The proportional-rationingmethod is defined as follows:
t
y = pr(N, t, x) = - ·x whenever xN > 0
XN

(whenever xN = 0, all rationing methods select y = 0).

Several related characterizations of the proportional-rationing method pertain to the
possibility of merging a subset of agents into a single agent with the combined demand,
or, conversely, to split one single agent into several smaller agents. These results are all
related to the fact that proportional rationing "discounts" each unit of claim/demand
by the same factor, irrespective of who presents this unit of claim/demand (whether the
agent has a large or small global demand is irrelevant). Hence the proportional method
is compelling when claims are transferable like anonymous bonds. Any other method
is vulnerable to manipulations by transferring claims across agents or changing their
identity by adding "artificial" agents.
For a given set N of agents and a subset S, S C N, we denote by N[s] the set with
(INI - ISI + 1) agents where all agents in S have been "merged" into a single agent
denoted S*. For instance:

N = {1,2,3,4,5}, S= {2,4,5} = N[s] = {1,S*,3}.

For any x in I N we denote xs = Eis xi, x[S]=projection of x on RS; and

x[S] E RE] is defined by x[ S] = xi if i S, xs) = xs. Now we consider four
independence properties of increasing notational complexity; yet, by Theorem 1 below,
they are logically equivalent in R.

No Advantageous Reallocation (NAR):

For all N, S, all t and all x,x' : x[S] = x'S ] = rs(N, t,x) = rs(N, t,x'). (1.1)

This says that by reallocating individual demands among the agents in S, the total
share of this coalition is unchanged, thus preventing such maneuver to be profitable.
Irrelevance of Reallocations (IR):
For all N, S, all t and all x, x':
[S
x[S] = ] = {rj(N, t, x) = rj(N, t, x') for all j c N\S}.
Reallocations of demands do not affect agents outside the scope of the reallocation.

Independence of Merging and Splitting (IMS).

For all N, S, all t and all x : r(N, t,x)[s ] = r(N s]
, t,x[S]).
The merging operation is the move from N to N[S]; splitting is the converse
transformation. By repeated applications of Independence of Merging and Splitting
298 H. oulin

we get the following property. Assume (N,)k e M is a partition of N and let x - x be

the "merging" mapping from RN into Rw given by

xk = xk for all k E M.

Then rNk(N, t,x) = rk(M, t,x*) for all k E M. The next property provides a more
precise decomposition of the rationing method by means of a partition:

Decomposition (DEC).
For any N and any partition (Nk,)k . of N, for all t, all x and all k:
r(N, t,x)[N,] = r(Nk, tk,X[NJl) where tk = rk(M, t,x*).

We compute first the shares of the members (Nk) of the partition, and then allocate
each share within the relevant coalition.
Theorem 1.1. Assume N contains three agents or more. The proportionalmethod
meets allfourpropertiesNAR, IR, IMS, and DEC. Conversely the proportionalmethod
is the only rationing method meeting any one of the four above properties.
The characterizations gathered in Theorem 1.1 are inspired from similar results by
Banker (1981), O'Neill (1982), Moulin (1987) and Chun (1988). See also Chun (1999)
and de Frutos (1999). Remarkably, the symmetry properties (such as Equal Treatment
of Equals) are not used.
In Section 1.4, another characterization of the proportional rationing method is based
on the fact that it is self-dual (r = r*), that is to say it allocates gains and losses in
exactly the same way (Proposition 1.6).

1.3. Uniform Gains and Uniform Losses

This pair of rationing methods are as important and (almost) as simple as the
proportional method. They aim at equalizing, respectively, the actual "gains" yi and
the net losses (xi - yi) across agents, under the feasibility constraints of a rationing
method:

The Uniform Gains method ug:

yi = ugi(N,t,x) = min{,xi} where A is the solution of Emin{X,xi} = t.
N
The Uniform Losses method ul:
Yi = uli(N, t, x) = (xi - ~I), where ,u is the solution of Z(xi - a)+ = t.
N

(where (z)+ = max{z, O}). In the literature, these two methods are often called
Constrained Equal Awards, and Constrained Equal Losses.
Ch. 6: Axiomatic Cost and Surplus Sharing 299

For a given rationing problem (N, t,x), let us denote by Y the set of feasible
solutions:

Y(N,t,x)= {yC RN 0 yi xi and yi =t}

One checks easily that ug(N, t, x) is the unique solution maximizing over Y(N, t, x) the
"leximin" ordering; that is, it lexicographically maximizes the smallest coordinate yi,
then the next-smallest coordinate, and so on. Similarly, ul(N,t,x) is the unique
maximizer of the "leximin" ordering applied to the vector of losses (xi -yi).
The pair ug, ul is a dual pair: ul = ug* and ug = ul*. This important fact allows a
parallel axiomatic treatment of these two methods.
Both methods ug, ul, as well as pr and all other symmetric methods discussed below,
respect the natural order of gains and losses. That is, they meet the following two
axioms:

Ranking: xi < xj Yi < Yj, (1.2)

Ranking* : xi < xj ( -Yi)
i < (Xj -yj). (1.3)

These two axioms are dual, namely a rationing method r satisfies one axiom if and
only if the dual method r* satisfies the dual axiom.
Although both methods ug,ul agree on the ranking of absolute gains and losses,
they differ sharply in the ranking of relative gains and losses. Consider the following
two dual axioms:
yj
Progressivity: O < xi < xj < Yi
Xj Xi
Regressivity: O < xi < x > Yi < Y
xi Xi

Proposition 1.1. The uniform-gains method is Progressive, but not Regressive. It is

the most progressive method among those satisfying Ranking.
The uniform-losses method is Regressive, but not Progressive. It is the most
regressive method among those satisfying Ranking*.
(The precise definition of "the most progressive" is left to the reader.)
Our next pair of dual axioms plays a very important role throughout this Section.
They are structural-invariance properties (see Introduction) allowing to decompose the
computation of shares when the available resources are estimated from above or from
below:

Upper Composition (UC):

For all N,x and t, t':
{0 < t < t' < xN} = {r(N, t,x) = r(N, t, r(N, t',x))}.
300 H. Moulin

Lower Composition (LC):

For all N,x and t, t' :
{O < t' < t < xN} => {r(N, t,x) = r(N, t',x) +r(N, t - t',x r(N, t',x))}.

If we allocate first the resources t', and later it appears that the available resources
are actually lower, namely t, Upper Composition allows to simply take the optimistic
shares r(N, t',x) as the initial demands from which to further ration until t. We
may forget about the initial demands x once we know an upper bound of the actual
resources. Note that UC implies Resource Monotonicity.
Dually, if we know a lower bound t' of the actual resources t, Lower Composition
allows to distribute the pessimistic shares r(N, t',x), subtract these shares from
the initial demands and distribute the balance (t - t') according to the reduced
claims x - r(N, t', x).
Proposition 1.2. The three methods pr, ug and ul meet the two axioms Upper
Composition and Lower Composition.
The family of methods meeting UC and LC is large: in Section 1.7 we describe a rich
set of such methods, and we show - with the help of additional requirements - that
our three basic methods pr, ug and ul play a central role within this family: Corollary
to Theorem 1.5. For the time being, we state two pairs of dual characterizations of
ug and ul. They are technically simple, but their interpretation is quite interesting. In
the following statement, we omit the variable N that plays no role.

Independence of Claim Truncation (ICT):

For all N, t,x : r(t, x) = r(t, x A t) where (x A t)i = min{xi, t}.

The part of one's claim that is not feasible has no influence on the allocation of the
resources:

Compositionfrom Minimal Rights (CMR)

For all N, t,x : r(t,x) = m(t, x) + r(t - mN(t,x), x - m(t, x)),
where mi(t, x) = (t - xN \ i)+.

Agent i's minimal claim mi(t,x) is that part of the resources that he will receive, even
in the most pessimistic case where the claims of all other agents are met in full. CMR
is the special case of LC where t' = mN(t,x).
Proposition 1.3. [Dagan (1996), Herrero and Villar (2001)]. The Uniform Gains
method is characterizedby the two propertiesLower Composition and Independence of
Claim Truncation. The Uniform Losses method is characterizedby Upper Composition
and Composition for Minimal Rights.
Ch. 6: Axiomatic Cost and Surplus Sharing 301

A different approach uses a priori bounds on individual shares, namely bounds that do
not depend on the size of other agent's claims. We denote by NI = n the cardinality
of N.
Lower Bound:
For all N, t,x, and all i yi= ri(N, t,x) >min{xi, t- },

Upper Bound:

For allN,t,x, and all i yt=ri(N,t,x)< {+(xi- )}.

It is plain that ug meets the Lower but not the Upper Bound, whereas ul meets the
Upper but not the Lower Bound. Lower Bound says that agent i is guaranteed a fair
share of the resources unless he demands no more than the fair share, in which case
his demand is met in full.
Dually, Upper Bound states that agent i's loss xi -yi is not smaller than the average
deficit XN - t, unless his claim is smaller than the average deficit, in which case he
gets no resources.
The Lower Bound has a lot of bite when t is small; if t < n mini{xi}, Lower Bound
forces equal gains: yi = t/n for all i. Similarly, if t is close enough to XN, Upper Bound
forces equal losses:
{XN - n ·min{xi} < t < xN} {Xi -Yi = Xj -yj for all i,j}.

Note that for NI = 2, Lower Bound characterizes the ug method, and (by duality)
Upper Bound characterizes the ul method. This simple fact does not extend to the
case IN > 3; however, we can still characterize the ug method if we bring Lower
Composition to the rescue. Consider the following very mild requirement:
Zero Consistency:
For all N, t,x and all i : {xi = 0} = {r(N,t,x)[N\i] = r(N\i,t,x[N\i])} (1.6)
It is hard to imagine under what circumstances the presence of a null demand agent
(who therefore receives nothing) could influence the allocation of resources among the
other, active agents.
Proposition 1.4. The Uniform Gains method is characterized by the following three
properties: Lower Bound, Lower Composition and Zero-Consistency. The Uniform
Losses method is characterized by the three properties, Upper Bound, Upper
Composition and Zero-Consistency.

1.4. The Contested-Garment method and Self-Duality

The Contested-Garment method is a rationing method for two agents only, in the
vein of the familiar "split the difference" principle for two-person bargaining. The
302 H. Foulin

interpretation of xi as the verifiable claim of agent i (as opposed to a vague demand)

is required for the application of the cg method and its n-person extensions. The
method is inspired by the following two quotes from the Babylonian Talmud [see
O'Neill (1982), Aumann and Maschler (1985)]: "R. Tahifa, the Palestinian, recited
in the presence of R. Abbahu: two [people] cling to a garment; [the decision is that]
one take as much as his grasp reaches and the other take as much as his grasp reaches
and the rest is divided equally between them." "Two hold a garment ... if one of them
says, 'It is all mine' and the other says, 'Half of it is mine', ... the former then receives
three quarters and the latter receives one quarter".
Consider a two-person rationing problem (t, xl, x2. We can interpret agent i's "grasp"
optimistically as min{xi, t} (in case his own claim takes absolute priority over the other
claim) or pessimistically as (t - xj)+ (if the other agent gets his full claim). Then we
split the remaining deficit (case of optimistic claims) or surplus (case of pessimistic
claims). Both computations yield the same method:

{ min{xl, t} + ½(t- min{xl, t} - min{x 2, t})

(t - X2)+ + 2 (t - (t - )+ - (t - 2)
(optimistic grasp),
(pessimistic grasp).

A more transparent reading of this formula in the case x l < x2 is:

if t < min{xl,x 2 }: Yl =Y2 =2t,

ifi t<x2' Yi =Xl; Y2=t-Xl, (1.7)
if max{x,,x 2} < t : Y = (t+x -2); Y2 2(t+x 2 -xl).

The Contested-Garment method is self-dual, r* = r, namely it allocates gains and

losses in exactly the same way. This property follows at once from the optimistic and
pessimistic formulas above:

rl (t , x ) =xl - (x N - - t - ( X 2 - t) + + ( x l - t ) +) = l(t +X1 + (t-Xl)w--X2-(t-2)-)

and the identity z + (z - t)y = min{z, t}.

On the other hand, the Contested-Garment method fails Upper Composition and
(by duality) Lower Composition: property (1.4) fails for xl = 10, x2 = 20, t = 15 and
t = 18.
Note that cg coincides with ug for small t, i.e., t < min{xl,x 2}, and with
ul for large t, i.e., max{xl,x 2} < t. More importantly, cg shares the two invariance
properties used above to capture ug and ul:
Proposition 1.5. [Dagan (1996)]. The Contested-Garment method is characterized
by Self-Duality and Independence of Claim Truncation; or by Self-Duality and
Composition from Minimal Rights, or by Equal Treatment of Equals, Independence
of Claim Truncation, and Composition from Minimal Rights.
Ch. 6: Axiomatic Cost and Surplus Sharing 303

Compare this result, for two-person problems, with the following compact character-
ization of the proportional method, for problems of arbitrary size:
Proposition 1.6. [Young (1988)]. The proportionalmethod is characterizedby Self-
Duality and Upper Composition; or by Self-Duality and Lower Composition.
Two natural extensions of the cg method for an arbitrary number of agents have been
proposed. The first one relies on the observation that for n = 2, the cg method is the
average of the two priority methods. The 12-priority method is the rationing method
[denoted prio(12)] that gives absolute priority to agent 1 over agent 2, hence:
if t is such that t < xl : y = (t, O);
if t is such that l < t < xl +X2: y = (xl, t -xl).
Define symmetrically the 21-priority method prio(21) and notice that Equation (1.7)
defining cg can be written as
cg = prio(12) + prio(21).
Hence the first generalization of cg as the Random-Priority method, namely the
arithmetic average of the priority methods over all orderings of N. Let a be an
ordering of N as (, o2 , ... , (o),namely ao is the highest-priority agent and so on.
We define y = prio(o)(N, t, x) as follows:
k k+l
if k is an integer such that Z xa < t < xol:
i= - i=1

y = x for j = 1, ... , k,

I ) I~~~~~~~~~(1.8)
y = 0 forj = k+2, ... , n.
Random-Prioritymethod:

y = Z prio(o)(N, t, x) where the sum bears on all orderings of N. (1.9)

The second natural extension of cg to any n uses an explicit mixture of the uniform-
gains and uniform-losses methods. This is the Talmudic rationing method due to
Aumann and Maschler (1985) (who argue convincingly that its intuition was present
already in the ancient Talmudic literature):
Talmudic method:
y = tal(N, t,x) = ug(N, mint, xN},½x) + ul(N, (t - xN)+, ix). (1.10)

The Talmudic method halves each claim and follows Uniform Gains until each half
claim is met. It then applies the Uniform Losses method to the remaining half claims.
304 . aloul

The Talmudic method coincides with cg in the case of two agents - yet another
equivalent formulation of cg.
Both the Talmudic and Random-Priority methods are self-dual. Both coincide
with Uniform Gains whenever t < mini{xi} and with Uniform Losses whenever
t > maxifxN \i}.
The next result shows a remarkable relation between the two methods, Random-
Priority and Talmudic, and the two most important value solutions for cooperative
games, namely the Shapley value and the nucleolus. These solutions are defined in
Sections 3.2 and 3.6 respectively. Fix a rationing problem (N, t,x) and define two
(dual) cooperative games, generalizing the bargaining interpretation of the contested
garment:

Forall
(S) S=Cmin{xs,
N: t} (optimistic grasp),
l w(S) = (t - XN\s)+ (pessimistic grasp).

Note that v(N)= w(N) = t.

Theorem 1.2: [O'Neill (1982), Aumann and Maschler (1985)1.
(i) The Random-Priority method allocates the resources according to the Shapley
value of the above games.
(ii) The Talmudic method allocates the resources according to the nucleolus of the
above games.

1.5. Consistent and symmetric methods

The axiom of Consistency has played a major role in the recent microeconomic
literature on distributive justice, see Chapter 26 of Volume 2. See also the surveys by
Thomson (1990) and Maschler (1990). Consistency in the rationing problem is both
very natural and extremely powerful, as demonstrated by the results of this and the
next subsections.

Consistency (CSY).
For all N, S, S C N, all t, all x:
r(N\S, t - rs(N, t, x), x[ \ s]) r(N, t, x)[N \ s]

Equivalently, Consistency can be defined by replacing coalition S with a single

agent i:

r(N\i, t - ri(N, t, x), x[N \ i]) = r(N, t, x)[N \ i]-

The axiom says that upon removing one agent (or several agents) from the society N,
and taking away the resources allocated to this agent (or agents) within N, the
allocation of shares within the reduced society remains the same. In other words,
Ch. 6: Axiomatic Cost and Surplus Sharing 305

changing the status of an agent from "active participant" to "passive expense of

resources" does not alter the overall distribution; removing one agent and his share of
resources is of no consequence to other agents. Thus Consistency is a decomposition
property with respect to changes in the set of relevant agents.
Note that Consistency is a self-dual axiom: a rationing method is consistent if and
only if its dual method is consistent as well.
In this section we discuss symmetric methods only. In this family a powerful
characterization result of (essentially) all consistent methods is available.
Our first result says that Consistency allows us to extend in at most one way a two-
person symmetric rationing method.
Proposition 1.7. Let r({ 1, 2})(t, (xl, x2 )) be a rationing method definedfor two-person
problems only. Assume that r({ 1, 2}) is symmetric and resource-monotonic. Then there
is at most one consistent rationing method r (defined for all finite societies N) that
coincides with r({1,2}) for all two-person problems. Moreover, r is symmetric and
resource-monotonic.
The discussion of parametric methods below establishes that pr, ug, ul, as well as
the Talmudic rationing method are consistent (of course this claim can be checked
directly). Therefore Proposition 1.6 has the following corollaries:
(i) The Talmudic method is the only consistent extension of the Contested-Garment
method (for two-person problems) to an arbitrary number of agents.
(ii) The Uniform-Gains method is the only consistent method satisfying Lower Bound
(yi > min{xi, t}) for two-agent problems (and a dual statement holds for Uniform
Losses by Proposition 1.4).
Proposition 1.7 begs the question: what symmetric two-person rationing methods
can be extended to a (symmetric) consistent method for an arbitrary number of agents?
A general answer is given by Dagan and Volij (1997) and Kaminski (2000): a certain
binary relation associated with the two-person method must be transitive. Theorem 1.3
below gives a much more transparent answer under one additional mild requirement:

Continuity:
r(N, t, x) is continuous in (t, x), for all N. (1.12)

We define now the family of parametric rationing methods. They are the key to
Theorem 1.3. Letf (, z) be a real-valued function of two real variables, with 0 < A,< A
and z > 0; the upper bound A may be finite or infinite. We assume:

f(0, z) = 0; f(A; z) =z;

f(A, z) is non-decreasing and continuous in 2 over [0, A].
To any such functionf we associate a unique rationing method r as follows:

For all N, t, x: ri(N, t,x) =f(, xi) where Ais a solution of f(A, xi) = t
iGN
306 H. Moulin

(this equation may have an interval of solutions but they all give the same shares
to every agent). We call r the parametric method associated with f By construction a
parametric method is symmetric; clearly, it is consistent as well.
The three basic methods pr, ug and ul are parametric, for the following functionsf:

Proportional: f(A, z) = A z, A = 1;
Uniform Gains: f (, z) = min{A,z, A = +oo;
Uniform Losses: f (, z) = (z - 1/A)., A = +o.

Among the two extensions of cg discussed in Section 1.4, the Random Priority
method is not consistent, whereas the Talmudic method is consistent. To check the
former claim, take N = {1,2, 3}, t = 10 and x = (6, 8, 10), and compute the shares
allocated under Random Priority:

y = 6+ r, 2=2,;
3 Y2
Y == 8+4= 3 1I =.' 10
Y3 146
10+t4+ 12=41
6 2=4.3

Next remove agent 3 and his share 4,3, which leaves us with the reduced problem:
N\{3},t' = 5 2, x = (6, 8). Now the shares under Random Priority are: yl =Y2 = 2.
To check the latter claim, we show that the Talmudic method is parametric. Set
A = 2 and define f(, z) as follows:

;_
, for 0 < A < z+2
for 2 < < z+2
- 2- for z+4 < <
-1fo z-2 2.
The next result establishes that parametric methods capture, essentially, all consistent
and symmetric rationing methods.
Theorem 1.3: [Young (1987)]. A parametric method is a consistent and symmetric
rationing method. Conversely, a rationing method satisfying Equal Treatment of
Equals, Consistency and Continuity can be representedas a parametric method where
f(A, z) is continuous in both variables.
Note that in the converse statement, it is enough to assume pairwise consistency,
namely the restriction of property (1.11) to subsets S containing two agents. On the
other hand, the converse statement holds only if we assume that the size of the set N
can be arbitrarily large, that is to say, the set .N of potential agents must be infinite.
This is an important limitation of Theorem 1.3, as well as of Theorem 1.4 in the next
section, that does not apply to Theorem 1.5 in Section 1.7.
The class of parametric methods is very rich. Chun, Schummer and Thomson
(1998), for instance, discuss a method of egalitarian inspiration much different from
any of the methods discussed in this survey.
Ch. 6: Axiomatic Cost and Surplus Sharing 307

1.6. Equal-sacrifice methods

The equal-sacrifice methods are an important subset of the parametric ones. They
appear early on in the discussion of equitable taxation schedules [see Mill (1859)] and
the discussion in Young (1990)].
Fix a real-valued function u(z) of the nonnegative real variable z, and suppose that
u is continuous and strictly increasing. Think of u as a reference utility function.
Loosely speaking, the equal-sacrifice rationing method associated with u is defined
by solving for all N, t,x the following system of equations:

u(xi)-u(yi) = u(xj)-u(yj) for all i,j Nand yi = t. (1.14)

iGN

Because u is strictly increasing, this system has at most one solution. Assume for a
moment that such a solution exists. Then at the allocation y, each and every agent
contributes an equal "sacrifice", namely the same net utility loss measured along the
reference utility scale u. This is especially appealing in the context of taxation. Let xi be
agent i's taxable income, let yi be his after-tax income, and let (xN - t) be the total tax
to be levied. Then the system (1.14) distributes taxes so as to equalize the net sacrifice
measured along the scale u. Concavity of u - decreasing marginal utility - means that
a dollar taken from the rich translates into a lesser sacrifice than a dollar taken from
the poor. Hence the choice of u allows the social planner to adjust the progressivity
of taxation while following the normatively transparent principle of equal sacrifice.
Here is a precise definition of the equal-sacrifice methods.
Proposition 1.8. Fix u, a continuous and strictly increasing real-valued function
defined on the nonnegative real line. For any rationingproblem (N, t, x) the following
system has a unique solution y, and y is a solution to the rationingproblem:

yi = t and for all i: {i > => u(xi) - u(yi) = max{u(x) - u(yj)}

icN
(1.15)
This rationing method satisfies Symmetry, Ranking, Consistency and Upper
Composition.
All equal-sacrifice methods are clearly consistent, but, in general an equal-sacrifice
method fails Lower Composition. The only exceptions are the Proportional and
Uniform-Losses methods. Moreover, an equal-sacrifice method meets Ranking* (1.3)
if and only if the utility function u is concave.
We turn to some examples of equal-sacrifice methods. The simplest ones involve
power utility functions:

uo(z) = log z yields the Proportional method

u (z)= z yields the Uniform-Losses method
308 H. Moulin

Interestingly, the Uniform-Gains solution is not an equal-sacrifice method, but it is

the limit of power methods. Consider the family of utility functions up:

up(z) = -- where 0 <p < +oc. (1.16)

For p close to zero the corresponding method approaches the proportional method,
whereas for p arbitrarily large it approaches the Uniform-Gains method. Let us
compute for instance the method corresponding to ul; the system (1.14) always has a
unique solution, and yields explicitly the parametric representation:
1 1 1 I xi
- -- for all ij -Ayj = for all i.
Yi Xi yj Xj 2i+ x i
Next consider the family of utilities uq:
uq(z) = zq where 0 < q < +oo. (1.17)
For q close to zero the corresponding method approaches pr, for q = 1 it is the
method ul, and for q arbitrary large it approaches the "hyperregressive" method that
gives full priority to the agents with the largest xi. In the case of two agents, this
method is defined as
prio(12)(t,x l, 2) if x2 < xl,
r(t,xl,x 2 )= prio(21)(t,xl,x 2) if xl < x2,
(t, t) if xl =x2.
Note finally that for q < 1, uq is concave and the corresponding method meets
Ranking*.
We state next a partial converse of proposition 1.7. It uses three additional axioms:
Strict Monotonicity. for all N, t, t,x : {t < t'} = {yi < y' for all i}.
Strict Ranking: for all N, t,x and all i,j: {xi < xj} => {yi < yj}.
Scale Invariance: for all N, t, x and a > 0: r(N, a -t, a x) = a ·r(N, t, x).
Strict Monotonicity and Strict Ranking are demanding properties; for instance both
ug and ul (as well as cg) fail both requirements. They are intuitively reasonable and yet
they cut a subset of rationing methods that is not topologically closed, an unpalatable
feature.
Scale Invariance, on the other hand, is an impeccable invariance axiom insisting
that the choice of the unit to measure both the demands/claims/taxable income and
the available resources, should be of no consequence whatsoever. It is satisfied by all
rationing methods discussed so far.
Theorem 1.4: Young (1988).
(i) A rationing method satisfying Consistency, Upper Composition, Strict Mono-
tonicity and Strict Ranking must be an equal-sacrifice method, defined by
system (1.14).
Ch. 6. Axiomatic Cost and Surplus Sharing 309

(ii) A rationing method satisfying Consistency, Upper Composition, Strict Monotonic-

ity, Strict Ranking and Scale Invariance, must be an equal-sacrificemethod derived
from a powerfunction up, 0 <p < oo, Equation (1.16), or must be the proportional
method.
An important open question. Replace in statement (i) Strict Monotonicity and Strict
Ranking by Monotonicity and Ranking: now all equal-sacrifice methods [given by
Equation (1.15)] with a concave utility, as well as Uniform Gains, are available. Is this
all? Similarly, if in statement (ii) we weaken the same two axioms in the same way,
all methods derived from the power functions u (1.16) as well as uq (1.17) for q < 1,
and ug, meet these requirements. Is this all?
Young (1990) offers an empirical "verification" of Theorem 1.4, by showing a
number of actual tax schedules that fit well within the family of equal-sacrifice methods
constructed from the power functions up.

1. 7. Asymmetric methods: combining the invariance axioms

When the recipients of the resources have different exogenous rights, in addition
to their possibly different demands, the symmetry axiom must be abandoned. In
bankruptcies and inheritances, creditors or heirs often have different status implying
some priorities between their claims, irrespective of their sizes. For instance the federal
government's claim on the assets of a bankrupt firm has absolute priority over the
claims of the trustees, who have priority over those of the shareholders, and so on.
The most asymmetric rationing methods are the priority methods prio(u) [Equa-
tion (1.8)]. In order to define a consistent priority method we must introduce the set Jv
from which the agents can be drawn. This set can be finite or infinite. Recall that in
the previous section about symmetric methods, A was any countably infinite set. In
the current section, by contrast, we can accommodate the case of a finite set .
We denote by a an ordering (complete, transitive, antisymmetric relation) of M and
for anyfinite subset N of M, we also write a for the induced ordering on N. Any finite
set N is ordered by a as N = (ol, F 2 , ... , on), and for any rationing problem (N, t, x),

we define the allocation prio(or)(N, t,x) exactly as in Equation (1.8). Note that the dual
of prio(a) is the priority method with the opposite ordering of KV.
The following fact is obvious: for any ordering o, the priority method prio(o) meets
Consistency, Upper and Lower Composition, and Scale Invariance.
Thus our four powerful invariance axioms are met by the three basic symmetric
methods pr, ug and ul as well as by the most asymmetric ones, the priority methods.
Theorem 1.5 below describes the relatively simple family of methods satisfying all four
axioms: they "connect" the three symmetric methods to the priority ones in interesting
ways.
We define the composition of rationing methods. Given are M and a parti-
tion Jv = UA where the parameter a varies in A. For every a we are also given
a rationing method on Ja denoted r; moreover, is a rationing method on A. The
310 H. Moulin

composition of these methods is denoted [r', a E A] = r. For any problem (N, t,x),
with a finite society N, N C A, we define N, = N n , and A is the finite subset of A
containing a if and only if Na is non-empty. The shares y = r(N, t, x) are computed
in two steps: first we split t among the subsets Na (i.e., among the "agents" of A)
according to , then the share z, allocated to N, is divided among the agents in N0
according to r:

z, = a(A, t, (xNfA)) for a C A; Yi = r(Na,zx[N,,]) for i N (1.18)

Thus, the operation of composition generates "two-tiered" rationing methods that
may apply different equity principles for the aggregate problem (on A) and for
any of the decentralized problems (on N,,). Note that the Decentralization property
(Section 1.2) says precisely that a certain method is preserved by "self-composition".
We say that the composition operation respects property Q if, whenever all methods
r, re, a C A, meet Q, so does the method [r, a E A].
Proposition 1.9.
(i) The composition of rationingmethods respects thefollowing properties. Resource
Monotonicity, Upper and Lower Composition, and Scale Invariance.
(ii) The composition operation does not respect the Consistency property, or Equal
Treatment of Equals.
(iii) If each method r, a A, is consistent, and (a is an ordering of A, the
composition prio((r)[ra, a E A] is consistent as well.
Proposition 1.9 shows that the three invariance axioms UC, LC and SI are met by
a rich family of rationing methods, obtained by composing such methods as pr, ug
and ul (as well as their asymmetric versions g" and iw , to be defined shortly) in an
arbitrary number of tiers. There are many more methods in this family, as discussed
in Moulin and Shenker (1999).
When we impose CSY as well, the set of available methods becomes much simpler,
although it still allows a great deal of flexibility. The following asymmetric versions
of ug and ul play a key role in the characterization result.
For any set of positive weights wi, one for each i e A, we define the weighted-gains
method gW as follows:
For all N, t,x : yi = g)`(N, t,x) = min{Aw i,xi}
where iAsolves Emin{Awi,xi} = t.
N

Its dual method is the weighted-losses method I"':

For all N, t, x : yi = li (N, t,x) = max{xi - uwi, 0}
where u solves E max{xi - fuwi, 0} = t.
N

The Uniform-Gains and Uniform-Losses methods are the two particular methods
corresponding to uniform weights (wi = 1 for all i). Note that when the weights of the
Ch. 6: Axiomatic Cost and Surplus Sharing 311

different agents are very unequal, the methods gW and lw become arbitrarily close to
any priority method: it will be enough to guarantee that if agent i is higher than agentj
in the priority ordering, his weight becomes infinitely bigger than agent j's weight.
Clearly, the methods gW and lW meet all four invariance axioms CSY, UC, LC and
SI. In view of Proposition 1.9, we can construct many rationing methods meeting
the four invariance axioms as follows. Partition arbitrarily the set fv in "priority
classes" and order these classes. In each priority class, use either the proportional,
or a weighted-gains, or a weighted-losses method. An example is provided by the
American bankruptcy law, which arranges the creditors in priority classes and uses
the proportional method within each class [Kaminski (2000)].
In order to state the last theorem in this section, we need two more definitions. We
say that the rationing method r gives priority to agent i over agent j if j does not
get anything unless i's demand is met in full: yj > 0 yi = xi (for all N, t,x). We
say that a rationing method is irreducible if for any pair i,j, r does not give priority
to i over j. For instance, pr, gW and lW (for any w) are all irreducible (recall that we
require positive weights wi).
Theorem 1.5: [Moulin (2000)].
(i) Let r be a rationing method meeting Consistency, Upper and Lower Composition
and Scale Invariance. Then there is a partition N = UANa, an ordering a of A,
andfor each a an irreduciblemethod r e meeting CSY, UC, LC and SI such that:

r = prio(o)[ra, a C A].

(ii) Let r be an irreducible method meeting Consistency, Upper and Lower

Composition and Scale Invariance. If A( contains at least three agents, then r is
either the proportionalmethod, or a weighted-gains method, or a weighted-losses
method.
In Moulin (2000), the somewhat involved family of irreducible methods for the case
II = 2 is described in full. Within the family uncovered in Theorem 1.5, our three
basic rationing methods are the only symmetric methods (except in the case [iV = 2).
Corollary to Theorem 1.5. Assume i contains at least three agents. Then there are
exactly three rationing methods satisfying Equal Treatment of Equals and the four
invariance axioms: they are the Proportional, Uniform Gains and Uniform Losses
methods.
A much needed next step in the theory of rationing methods is an asymmetric
version of Theorem 1.3: what is the set of methods consistent and continuous?
Naumova (2002) offers an asymmetric generalization of Theorem 1.4, where the utility
functions measuring sacrifice are personalized.
Another interesting open question [discussed by Moulin (2000)] is to generalize
Theorem 1.5 (or its Corollary) by dropping one of the four invariance axioms. For
instance, a method meeting Consistency, Scale Invariance and Upper Composition is
312 I. Moulin

priority to higher demands: given the profile of demands x, this method gives priority
to i overj if and only if xi > xj, and treats equal demands equally (thus it is symmetric
as well); it emerged in Section 1.5 as the limit of some equal sacrifice methods [see the
discussion of power methods (1.17)]. Its dual method, priority to lower demands, meets
all four axioms in Theorem 1.5 except Upper Composition. The characterization of all
rationing methods meeting Consistency, Scale Invariance and one of the composition
axioms is wide open.

1.8. Fixed-path methods

This important family of rationing methods contains asymmetric variants of the
uniform-gains method as well as the priority methods. The fixed-path methods play
an important role in Section 3 when we discuss Demand Monotonicity (Sections 3.4
and 3.6). They emerge also in the model of fair division under single-peaked
preferences (briefly discussed in Section 1.10), where they are a key example of
strategy-proof methods. In the present section we merely define these methods and
check their invariance properties.
It is necessary to place an exogenous bound on individual demands. This bound may
be finite or infinite. We call it the capacity of agent i and write X where Xi < +oc
(real or infinite). A rationing problem (N, t, x) must now satisfy 0 < xi < Xi for all i.
We always assume that xi is finite for all i.
A fixed-path method is defined from a family of monotone paths y(N), one for
each possible society N. The path y(N) is a non-decreasing mapping from [O,XN]
into [O,X[N]] such that

for all t, 0 < t < XN: E yi(N, t) = t, 0 < yi(N, t) <Xi for all i,
N
lim yi(N, t) = Xi for all i
t XN

Note that y must be continuous in t. If Xi is finite for all i, the limit property holds
true because y(N,XN) = X[N].
The fixed-path method r is now defined as follows:

r7'(N,t,x) = min{yi(N,s),xi} for all i,

where s is a solution of ymin{yi(N,s),xi} ti..9)
ti=
N

If we take x = X (x = X[N]) in this equation, we find

y(N, t) = rl'(N, t,X) (1.20)

Examples of fixed-path methods include the uniform-gains method [for the

path ug(N, t,X)], any weighted-gains method, and any priority method prio(a). Note
Ch. 6: Axiomatic Cost and Surplus Sharing 313

that a priority method can be represented as a fixed-path method only if all capacities x i
are finite (with the possible exception of the capacity of the last agent in the priority
ordering). The path t - prio(N, t,X) follows the edges of the cube [O,X] in the order
specified by a.
If Xi = X for all i,j, uniform gains is a symmetric fixed-path method. It is the only
fixed-path method meeting Equal Treatment of Equals: indeed the path r(N, t,X) must
be diagonal by ETE, so the claim follows from Equations (1.20) and (1.19).
The set of fixed-path methods is not stable by duality: for instance uniform losses
is not such a method. It contains no self-dual method.
Proposition 1.10.
(i) All fixed-path methods meet Upper Composition. They generally fail Lower
Composition.
(ii) A fixed-path method is consistent if and only if the associatedpaths N -+ y(N)
commute with the projection operator.

For all N, S, S C N: (N)[] = y(S);

namely y(N, t)[s] = y(S, ys(N, t)) for all t.

Note that all the methods obtained by a priority composition of weighted-gains

methods (see Proposition 1.9) are fixed-path methods and satisfy Lower Composition.
I conjecture that there is no other fixed-path method meeting LC.
The property (1.21) in statement (ii) is especially easy to read when the maximal
set N of potential agents is finite. The single path y(N) from 0 to X[gA] generates the
entire family of paths y(N) by simple projection on N. In this case we can really speak
of a one-path method.

1.9. Rationing indivisible goods

We modify the rationing model assuming that the commodity being distributed comes
in indivisible units. Examples include cars, appliances, seats for a concert or in a plane,
organs for transplant, etc.
The formal model is identical, except that all the variables t, x i,yi are nonnegative
integers. The definitions of a rationing problem, a solution, and a rationing method
are unchanged. The set of such methods is denoted RZdd. The duality operation is
unchanged.
It is convenient to think of a rationing method as a scheduling algorithm. Fix N and x
and restrict attention to resource-monotonicrationing methods. The path t -4 r(N, t, x)
is described as a sequence {il, ... , iK } in N, where K = XN and ii is the agent receiving
the first unit [r(N, 1,x) gives the unit to i], i2 is the agent receiving the second unit
and so on. In the sequence {il, ... , iK}, agent i appears exactly xi times, for all i.
The definitions of Consistency, Upper Composition and Lower Composition are all
unchanged. Note that Consistency has a particularly simple formulation in terms of
314 I. Moulin

the sequence {il ... i describing the path t r(N, t,x). The axiom says that by
simply dropping all occurrences of a certain agent i in this sequence, we obtain the
sequence describing the path t r(N\i, t,x[N \ ]).
Symmetry is lost when we allocate indivisible goods, as long as the allocation is
deterministic. If we now think of the division of resources as a random variable, we
can restore this basic equity property, at least in the ex ante sense. It turns out that the
probabilistic rationing of indivisible goods arises naturally in the discussion of additive
cost-sharing methods in Section 3 - an entirely deterministic model.
A probabilistic rationing method associates to every deterministic rationing model
(N, t, x) (where t and xi are integers) a random variable Y such that, with probabil-
ity one, 0 < Yi xi for all i and Y = t. The three basic methods pr, ug and ul have
a canonical probabilistic analog.
To define the proportional method, fix the profile of claims xi and throw xi balls of
color i in an urn, for each i N; drawing from the urn t times, independently and
without replacement - and with uniform probability - generates the random variable
Y = r(N, t, x) of the random proportional method. Clearly, the expected value of Y is
agent i's proportional share t ·(xi/xN).
The random proportional method meets Consistency, Upper and Lower Compo-
sition, as well as Equal Treatment of Equals (ex ante). Conversely, the method is
characterized by ETE, UC and LC [Moulin (2002)].
The probabilistic analog of uniform gains is called Fair Queuing [Shenker (1995),
Demers, Keshav and Shenker (1990)]. Given a profile of claims xi, this method gives
away one unit to each agent in round-robin fashion, selecting randomly and with
uniform probability the ordering in which they receive each unit; an agent drops out
only when his claim is not met in full. The expected value of agent i's share after
t units have been distributed is exactly his uniform-gains share in the deterministic
problem (N, t, x).
The Fair Queuing method meets Consistency and Upper Composition, but fails
Lower Composition. Moulin and Stong (2002) show that this method is characterized
by the combination of CSY, UC, and a strong form of Equal Treatment of Equals:
two agents with identical claims have equal expected shares, and their actual (ex post)
shares never differ by more than one unit.
The dual method, Fair Queuing* allocates each unit with equal probability among
the agents with the highest remaining claim, i.e., their initial claim net of the units
received in earlier rounds.
The characterization results in the probabilistic model of rationing are generally
sharper than in the classical model. Moulin and Stong (2002) provide very complete
descriptions of the set of methods meeting UC and LC, or CSY and UC (or
CSY and LC).
Ch. 6: Axiomatic Cost and Surplus Sharing 315

1.10. Two variants of the rationing model

1.10.1. Surplus sharing
In a surplus-sharingproblem (N, t, x), the resources t must be divided according to the
profile of claims x and we assume t > XN: the resources exceed the sum of individual
claims. One interpretation is that xi is the amount of investment contributed by agent i
to a joint venture, and t is the total return, allowing a profit t - xN. Alternatively,
the resources being distributed are undesirable (a tax, a workload) and agent i's claim
xi entitles him to receive no more than a share xi of the total liability. These claims
are not compatible.
A solution y to the surplus-sharing problem allocates a share yi to agent i in such
a way that 0 < xi < yi and YN = t. A surplus-sharing method d associates a
solution y = d(N, t, x) to every surplus-sharing problem (N, t, x).
The Proportional surplus-sharing method is given by the same formula as in the
rationing case. Uniform Gains is defined as follows:

Yi = ugi(N, t,x) = max{A,xi} where i is the solution of E max{,xi} = t.

N

The counterpart of the Uniform-Losses rationing method simply divides the surplus
equally, and for this reason we call it the egalitarianmethod:

Yi = egi(N, t,x) = x i + -(t

n -xN).
In the surplus-sharing model there is no duality operation, hence no analog to the
Contested-Garment method.
Consistency and Scale Invariance have the same definition but there is only one
Composition axiom:
For all N, t, t',x : XN < t' < t = d(N, t, x) = d(N, t, d(N, t', x)). (1.22)
Several axiomatic results about rationing have a direct counterpart in the surplus-
sharing model, and several new results emerge as well. For instance, the proportional
method is characterized, as in Theorem 1.1, by Independence of Merging (or Splitting),
or by Decomposition. On the other hand, many surplus-sharing methods meet
No Advantageous Reallocation, including the egalitarian method.
Theorem 1.3 about parametric methods and Theorem 1.4 about equal sacrifice
methods are readily adapted to the surplus-sharing context; see Young (1987) and
Moulin (1987).
The following result is the counterpart of Theorem 1.5 and its Corollary. The
asymmetric generalizations of the egalitarian method divide the surplus in proportion
to a set of fixed shares wi, wi > 0 for all i and WN = 1:
Yi = r(N, t,x) = xi + wi (t - XN)
The proportional method and the fixed-share method rw meet No Advantageous
Reallocation, Consistency, Composition and Scale Invariance. Conversely, these four
316 H. Moullin

axioms characterize this family of surplus-sharing methods. If we add Equal Treatment

of Equals to the list of requirements, only the Proportional and the Egalitarian methods
are left. See Moulin (1987).

1.10.2. Fair division with single-peaked preferences

Think of a context where the size of agent i's claim/demand xi is private information,
so that agent i may choose to misrepresent its actual value if this proves beneficial.
We make the following assumption on individual preferences over shares: given that
his (real) claim/demand is xi, agent i strictly prefers yi to y1 ifyj < yi < xi, but strictly
prefers y' to yi if xi < y' < yi. This is the familiar assumption of single-peakedness.
It is a realistic assumption in the rationing problem if the resources being distributed
are not freely disposable: think of food that must be eaten in one day, or of a share in
a risky venture. For examples and discussion of this assumption see Sprumont (1991)
or Barbera, Jackson and Neme (1997).
A fair-division method works as follows in this context. The mechanism elicits the
peaks of individual preferences (corresponding to the claims xi in the rationing or
surplus-sharing models) and each peak xi can be anywhere in the fixed interval [0, Xi].
For a given amount of resources t, the sum of individual claims XN may be smaller or
larger than t. Thus the allocation problem may be a rationing problem or a surplus-
sharing problem, and an allocation method is a pair of one rationing and one surplus-
sharing method.
Incentive compatibility of this mechanism is the strategy-proofness property:
reporting one's true peak is optimal for every agent, irrespective of other agents'
reports.
The key observation is that Uniform Gains (used both for the rationing and the
surplus-sharing cases) is a strategy-proof method, and so are all the fixed-paths
methods, where a different path can be used for the rationing and for the surplus-
sharing cases. Conversely, Uniform Gains is characterized by Strategy-proofness,
Efficiency and Equal Treatment of Equals: Sprumont (1991), see also Ching (1994).
Similarly, the consistent fixed-path methods are characterized by Strategy-proofness,
Efficiency, Consistency and Resource Monotonicity: Moulin (1999), see also Barbera,
Jackson and Neme (1997).
There is also a sizable literature looking at the fair-division problem with single-
peaked preferences from an equity angle, and where axioms such as No Envy
or Population Monotonicity play a big role: see Thomson (1994a,b, 1995, 1997),
Schummer and Thomson (1997) and references therein. Once again Uniform Gains
stands out as the method of choice.

2. Sharing variable returns

2.1. The problem and some examples

A (one-dimensional) cost-sharing problem is a triple (N, C, x) where N is a finite set
Ch. 6: Axiomatic Cost and Surplus Sharing 317

of agents, C is a continuous non-decreasing cost function from 1R+ into IR+ such that
C(O) = 0, and x = (xi)i CN specifies for each agent i a demand xi, xi > 0.
A solution to the cost-sharing problem (N, C, x) is a vector y = (yi)i C N specifying
a cost share for every agent and such that

yi > for all i; EYi = C(xi) (2.1)

iEN in

A surplus-sharing problem is the same mathematical object as a cost-sharing

problem but its interpretation is different: the given function is denoted F (to avoid
confusion) and is now a production function; if total input contribution is z, total
output is F(z); next xi is agent i's input contribution and yi is agent i's share of the total
output F(xN). The whole axiomatic discussion is unaffected by the choice of one or the
other context, although certain axioms are not equally natural in both contexts. With the
exception of a few examples, we use the cost-sharing interpretation and terminology
throughout Sections 2 and 3.
A cost-sharing method (resp. a surplus-sharing method) is a mapping p associating
to any cost-sharing (resp. surplus-sharing) problem a solution y = qP(N, C,x). We
denote by M the set of cost-sharing methods thus defined.
Note that variable-population axioms play no role in this section [see Subsec-
tion 2.5.2]. Therefore, omitting N in the variables of Tp, we write y = (C,x).
The question addressed in this section is the equitable division of cost (or surplus)
shares when the returns of the technology vary. In other words, our initial postulate
is that constant returns pose no equity issue whatsoever: cost (or surplus) shares must
simply be proportional to individual demands of output (resp. contributions of input).
This corresponds to the following axiom on the cost-sharing method AP:

Constant Returns:
{C(z) = A.z for all z > } = {(N, C,x) = x} (2.2)
for all N, all A > 0, all C, all x.

A simple example of a cost-sharing problem with increasing returns (decreasing

average cost) is discount pricing. The agents in N are grouping their order of wine
(there is only one quality of wine). Wine can be bought at the local store at price pi,
or at a lower price P2 from a discount retailer located far away. In the latter case a
fixed transportation cost co (independent of the shipment size) must be added. Hence
the cost function:

C(z) = min{p l . z, co +P2 z}. (2.3)

If the total demand xN justifies buying from the discount retailer [that is, if xN >
co/(pI -P2)] how should total cost be split among the buyers? With several suppliers,
318 H. loulin

the cost function C takes the form of a concave, increasing and piecewise linear
function starting at C(O) = 0.
Our second example is a cost-sharing problem with decreasing returns (increasing
average cost). Think of N as encompassing all the consumers of a certain good (N is a
monopsonist for this good) competitively supplied. Thus the demand z is met at
price S' l(z), where the supply function p - S(p) is increasing; the resulting cost
function C(z) = z S-(z) has decreasing returns.
In the surplus-sharing context, we find symmetrical examples displaying increasing
or decreasing returns technologies. For instance, the agent in N may be monopolizing
the supply of a certain good for which the demand is competitive. The market
absorbs z units of output at price D(z), where D is decreasing; hence the revenue
function F(z) = z D(z) has decreasing returns.
A simple example with increasing returns involves fixed costs [as in example (2.3)].
The agents can use a technology with constant returns r and no fixed input cost, or
they can pay a fixed input cost co and benefit from higher returns r2:

F(z) = max{rl z, r2 (z - co)}. (2.4)

A brief overview of Section 2 follows. In Section 2.2, the average cost-sharing

method is characterized in precisely the same way as proportional rationing in
Section 1.2. Serial cost sharing is introduced in Section 2.3: together with average
cost sharing, it plays the key role in the current model. In Section 2.4 the property of
Additivity (of cost shares with respect to the addition of cost functions) is defined
and the main theorem derived: the set of rationing methods is isomorphic to that
of additive cost-sharing methods; in particular, serial cost sharing corresponds to the
uniform-gains rationing method. Some variants and open questions are gathered in
Section 2.5.

2.2. Average-cost method

The simplest cost-sharing method divides total cost in proportion to individual

demands. It is denoted ac:

y = ac(C,x) = C()-- x· (2.5)

XN

(of course, if XN = 0 we must have y = 0). The average-cost method entirely ignores
the returns of the technology between 0 and the total demand XN. From all the methods
discussed in this section, it is the most informationally economical. This is convenient
from an implementation viewpoint, but has no normative appeal per se.
A first type of axiomatic justification for this method mimics those of the
proportional-rationing method in Section 1.2. The axioms of No Advantageous
Reallocations (NAR), Irrelevance of Reallocations (IR), and Independence of Merging
Ch. 6: Axiomatic Cost and Surplus Sharing 319

and Splitting (IMS), are transported word for word from that context to that of cost-
sharing methods by simply replacing the resources t in rationing by the cost function C.
Theorem 1.1 has the following counterpart:
Theorem 2.1. Assume N contains three agents or more. The average-cost method
meets the three properties NAR, IR and IMS, as well as the following property.

No Chargefor Null Demand:

{xi = 0} ~ {yi = cpi(N, C,x) = 0} for all C,x and all i. (2.6)

Conversely, the average-cost method is the only cost-sharingmethod chargingnothing

for a null demand and meeting any one of NAR, IR or IMS.
The interpretation of Equation (2.6) in the case of cost sharing is that no one should
have to pay anything for no output; in the case of output sharing, it is sometimes
referred to as "No Free Lunch": you do not receive any output if you did not participate
in the production process by contributing some money or some labor. All methods
discussed in Sections 2.2 to 2.4 satisfy property (2.6). In Section 2.5 we give some
arguments against this axiom and offer a method that violates it.
The interpretation of the three axioms NAR, IR and IMS is the same as in the
case of rationing: one does not need to monitor the "identity" of the various units of
demands (whether a certain unit comes from an agent with a large or small demand
is irrelevant). Any unit of demand is treated anonymously and therefore there is no
benefit in passing them around.

2.3. Serial cost sharing

The average-cost (average-returns) method entirely ignores the variation of the returns
between 0 and XN. When those returns vary widely and when individual demands are
of very different size as well, this result in an unpalatable distribution of costs (or
output). Consider the (decreasing-returns) cost function

C(z) = (z- 10)+ where, as usual, (a)+ = max{a, 0}. (2.7)

The first 10 units are free, and additional demands cost 1 per unit. Say N = {1,2, 3}
and consider the profile of demands x = (3, 5, 7). The average-cost method gives
y = (1, 13,24). Is it fair that agent 1 pays anything, when he could argue that his
fair share of the 10 free units is 3 and that he is not consuming that much? The
point is that agent 1 is charged the high average cost that he did not cause in the first
place: as C(3x 1) = 0, if no one else asks more than he does, no one has to pay; hence
he should not be held responsible for costs that only arise because other agents demand
more than he does.
Notice that, viewed in the light of output sharing, the argument is less convincing:
here F(z) = (z - 10)+ is a production function requiring a fixed cost of 10 before output
320 H. Moulin

can be collected [a particular example of Equation (2.4)]. Agent l's contribution of

3 units of input is useful, even if applied to pay the fixed cost; other agents should
give him some share of the output.
Next we look at the (increasing-returns) cost function [a special case of the discount
pricing example (2.3)]:

C(z) = min{z, 9 + -0z} (2.8)

with x = (3, 5, 7). Average cost yields y = (2.1,3.5, 4.9), so agent 1 ends up paying less
than his Stand-Alone cost C(xl) = 3. Note that the first 10 units cost 1 apiece, and that
the price drops to .1 for each additional unit. This time, agents 2 and 3 protest that
they were the ones responsible for reaching the low marginal cost, because 3x < 10,
so agent 1 should not get any benefit from that; his fair share of the cost is 3 because
the returns are constant up to the level 3x1.
Notice that the argument is even stronger in the output-sharing context. The
production function has a high return of 3 up to 10 units of input, after which the
return drops to L. Agent 1 is entitled to a fair share of the "good returns": as his
"demand" falls below this fair share 10/3, he should receive 9 units of output, a far
cry from what the average-returns method offers him.
The above discussion suggests the following upper and lower bounds on cost shares,
depending on the variation of marginal costs/returns. The set N is fixed and #(N) = n.

Increasing marginal costs bounds (IMC bounds):

If C is convex: C(xi) < y = (C,x) < (-- ) for all i, all x. (2.9)
n
Decreasing marginal cost bounds (DMC bounds):
If C is concave: (-- i) yi = (C,x) < C(xi) for all i, all x. (2.10)
n

We let the reader check that each one of the announced bounds is compatible with
budget balance in the corresponding domain of cost functions. For instance, a convex
cost function such that C(0) = 0 is subadditive, hence the left-hand inequality in the
IMC bound is feasible. And so on.
Consider a convex cost function. The Stand-Alone lower bound yi > C(xi) says
simply that no agent can benefit from the presence of other users of the technology.
This is compelling when marginal costs increase because the consumption of any user
creates a negative externality on that of any other user. Indeed, most cost-sharing
methods discussed in Section 2 meet the Stand-Alone lower bound when C is convex,
and the Stand-Alone upper bound when C is concave (case where any user creates a
positive externality on any other user). This is true for all additive methods: Corollary 1
to Theorem 2.1.
By contrast, the two remaining inequalities in Equations (2.9) and (2.10) fail for the
average-cost method, as shown by the numerical examples above.
Ch. 6: Axiomatic Cost and Surplus Sharing 321

Consider again a convex cost function and the Unanimity Upper Bound yi <
C(nxi)/n. This says that an agent's cost share cannot exceed her share when all agents
demand the same amount as she does (and are treated equally). Given that marginal
costs increase, this conveys the idea that agent i is entitled to a fair share of the "good"
marginal costs, namely those of the first nxi units. Think of the scheduling example:
we are saying that all agents have an equal right to the best (i.e., the earliest) slots in
the queue. If xi is much smaller than the other demands, this bound has a lot of bite.
A symmetrical interpretation holds for the Unanimity Lower Bound (yi > C(nxi)/n)
when C is concave: in the output-sharing context, it says that agent i is entitled to a
fair share of the good marginal returns; in the cost-sharing context, that she should
accept her fair share of responsibility for the "bad" marginal cost. See Moulin (1992)
for a general discussion of the notion of unanimity bounds.
The serial cost-sharing formula [Shenker (1995), Moulin and Shenker (1992)] is
directly inspired by the unanimity bounds. Fix C and a profile of demands x. We
start by relabeling the agents by increasing demands: x < x2 < ... < x,. First we
split equally the cost of the first nxl units among all agents. Now agent 1 is served
(and pays C(nxl)/n) and we split equally the cost of additional units between the
remaining agents {2, 3, ... , n}, until agent 2 is served, and so on. Formally we define
a sequence x', i = 1, ... , n, as follows:
i-1

xl = nx l; x2=xl + (n - )x2; ... ; xi=(n-i+l)xi+t j; ... ; Xn = XN.

j=l
(2.11)
Note that the sequence xi is non-decreasing. The serial cost shares are now:
C(x ) ; C(x2 ) C(2) C(x1 C(x-I))
Yi
n
; Y2 =Yl +
n-l
n-1
; Yi =Yi-i +
-i+l
(2.12)
or equivalently:
C(xl) C(x2) C(xl) c(xi) C(xj)
Yi- Y2 Yi -
n n-1 n(n - 1)' n- i 1 j -j (n
(2.13)
In the cases n = 2 and n = 3 the general formulas (2.12) and (2.13) are simple:
n = 2; xl < x2: l = C(2X 1 ); y2 = C(Xl +x 2)- QC(2xl)
n = 3; x al X2 < x3 : Yi = C(3xl); Y2 = C(xl + 2x 2)- C(3x,);
y3= C(XN)- C(XI +2X2)- C(3x1).
For instance, in the numerical examples discussed above:
{C(z) = (z - 10)+, x = (3, 5, 7)} = y = (0, 1.5, 3.5)
{C(z) = min{z, 9 + lz}, x = (3, 5, 7)} y = (3, 3.65, 3.85).
Recall from the discussion after Equation (2.7) that the serial cost share yl = 0 is
plausible in the cost-sharing interpretation, less so in a surplus-sharing story. Similarly
322 H. Moulin

in the case of the cost function (2.8), the serial cost share yl = 3 denies any cost saving
to agent 1, despite the fact that his presence increases the cost savings of the other
two agents: this is clearly an extreme interpretation of fairness in this example.
In the examples, the agent with the smallest demand prefers his serial cost share to
his average cost share in the example with increasing marginal cost, and his preferences
are reversed in the example with decreasing marginal cost. The preferences of the agent
with the largest demand are diametrically opposed. This is a general fact.
Kolpin [1998] proposes further interpretations of the serial formula in terms of linear
pricing.
We conclude Section 2.3 by generalizing the decentralized bounds (2.9) and (2.10)
for the serial cost shares to a cost function with arbitrary returns. That is, we give an
upper and a lower bound onyi = qi(C,x) that only depend on C, xi and n, the number
of users. This is important for an uninformed agent, who cannot assess the size of
other agents' demands.
Proposition 2.1. The serial cost-sharingmethod meets the IncreasingMarginal Costs
bounds (2.9) and the Decreasing Marginal Costs bounds (2.10). Moreover: for any
non-decreasing cost function C (such that C(O) = 0), it satisfies the following
Universal Bounds.

-C(xi) < Yi = p7,(C,x) < C(nxi). (2.14)

n

It is easy to check that the average-cost method fails both universal bounds. Take the
cost function (2.7) and x = (3, 5, 7): the upper bound is violated for agent 1. Take the
cost function (2.8) and x = (3, 20, 27): the lower bound is violated for agent 1.
The universal bounds are deceptively mild: they eliminate many appealing cost-
sharing methods. Among the additive methods analyzed in Section 2.4, the universal
lower bound is met by many methods besides serial cost sharing. For instance, the
Shapley-Shubik cost-sharing method (see Section 2.4) meets this bound, and so does
any convex combination of serial and Shapley-Shubik. On the other hand the universal
upper bound essentially characterizes serial cost sharing: see Theorem 2.3 below.

2.4. Additive cost sharing

In the rationing problem, the requirement that the solution y depends linearly upon the
resources t is enough to single out the proportional-rationing method: Chun (1988). By
contrast, in the cost-sharing problem with homogeneous goods, there is a rich family
of cost-sharing methods where the solution y = (C,x) depends additively on the
function C. Theorem 2.2 below establishes a linear isomorphism between this family
and the set of (resource-monotonic) rationing methods. Thus Additivity leaves a lot of
maneuvering room to the mechanism designer.
Ch. 6: Axiomatic Cost and Surplus Sharing 323

With a slight abuse of notation we denote by 1Z the set of monotonic rationing

methods (note that all rationing methods discussed in Section 1 are monotonic).
An element r of R. defines for all x E IRN a monotonic (hence continuous)
path t -- r(t,x) from 0 to x:
0 < r(t,x) <x, rN(t,x) = t for all t, 0 t <xN;
t t' = r(t,x) < r(t',x) for all t, t', 0 < t' < XN
The domain D of cost functions consists of all the functions C that can be written as
the difference of two convex functions: this domain contains all the twice continuously
differentiable functions, as well as all the piecewise linear functions. Naturally, we also
require each function C in D to be non-decreasing and such that C(O) = 0.
We denote by M the set of cost-sharing methods: an element q( of MA associates a
solution (p(C,x) to every cost-sharing problem (C,x), where C C D and x R+N . In
addition to Constant Returns (2.2), we consider the following powerful axiom:
Additivity (ADD):
((C + C 2; x)= q(C1; x)+ (C2 ; x) for all C 1,C 2 D, all x (2.15)
This property allows one to decompose the computation of cost shares whenever the
cost function itself can be additively decomposed. This commutativity brings a sharp
representation result: the additive cost-sharing methods are isomorphic to rationing
methods.
We denote by Ft the cost function F,(z) = min{z,t} (easier to interpret
as a production function: returns are one until the level t, then drop to zero).
Finally we denote by M(P, Q,...) the subset of cost-sharing methods meeting the
properties P, Q....
Theorem 2.2: [Moulin and Shenker (1994)1. Consider the following two mappings,
from R into M (CR,ADD) andfrom M (CR,ADD) into R.1Z.
r /: : qp(C,x) = foN C'(t)dr(t,x) for all C E D, all x, (2.16)
( r: r(t,x) = q((F, x) for all t,x (2.17)
These two mappings define a linear isomorphism from 7? into M (CR,ADD) and
back.

Corollary to Theorem 2.2. All cost-sharing methods in M (CR,ADD) meet the

following properties:
(i) No charge for null demand: xi = 0 = yi = 0.
(ii) Stand-Alone lower (upper) bound under increasing (decreasing) marginal costs:
C convex => ys > C(xs) for all C, x, all S C N
C concave ys < C(xs) for all C,x, all S C N.

Theorem 2.2 establishes a precise isomorphism between monotonic rationing

methods and additive cost-sharing methods. In particular, the key methods on both
324 H. Moulin

sides are matched, and many of the normative requirements in one model have a
counterpart in the other. Below is a list of rationing methods and cost-sharing methods
matched by the linear isomorphism:
(a) Proportionalrationing - average cost sharing. In X?, the proportional method
gives to every dollar of claim the same right to the resources t; similarly in M,
average cost sharing gives to every unit of demand the same responsibility in total
cost (every unit of input is entitled to the same output share).
(b) Uniform gains rationing +-+serial cost sharing. Fix N and x, and label the agent
so that xl < x2 < x,,. Agent i's share y = ugi(t,x) where t varies in [O,XN]
is easily computed with the help of the sequence x i given by Equation (2.11):

if O < t < x: Yi = t;
2
if x < t x : yi =xl + nI (t - x')

j
if x 1 <t <j : yi = x_l + I (t - j)

Xi I < t < X: yi = Xi- + n-l (t -

-
)
i
x < t: Yi = i

Therefore the cost-sharing method associated with uniform gains by Equa-

tion (2.16) is precisely given by Equation (2.12), as claimed.
In 7R the uniform-gains method gives an equal claim to all agents on the first
units of resources until their claim is met. Similarly, serial cost sharing makes
all individual demands pay an equal share of the first units produced until their
demand is met.
(c) Priority rationing +- incremental cost sharing. To an ordering of the
agents in N a mapping from {1,. .. , n} into N - we associate the following
incremental cost-sharing method:

Ya = C(x); Y2 = C(xa, +X2)-C(x,); .. Y = C(x C)-C(xj)

i=1 / =1
(2.18)
It corresponds to the priority rationing method prio(a) [see Equation (1.8)].
(d) Random priority ordering -* Shapley-Shubik cost sharing. The averaging
operation is preserved by the linear isomorphism, therefore the Random-Priority
rationing method (1.9) is associated with the arithmetic average of all incremental
cost-sharing methods. This method, originally proposed by Shubik (1962),
distributes costs according to the Shapley value of the Stand-Alone cost game:

=
0<s
z n I
s!(N-s- 1)! Z
S CN\i
[C(xsui)-C(xs)j (2.19)
Isl-
Ch. 6: Axiomatic Cost and Surplus Sharing 325

In the case of two agents, this gives the cost-sharing method corresponding to the
contested-garment rationing method:

yi {C(X +x 2 ) + C()-C(xj)} where {i,j} = {1, 2}. (2.20)

The Shapley-Shubik method plays an important role in the model with heteroge-
neous goods (Section 3); its characterization there (Corollary 1 to Theorem 3.4)
is quite convincing. Contrast this with the lack of normative arguments in favor
of Random-Priority rationing, or in favor of the Shapley-Shubik method in the
current model with one homogeneous good.
The isomorphism in Theorem 2.2 also suggests new cost-sharing methods corre-
sponding to simple rationing methods. For instance, Uniform Losses gives rise to a
"dual" serial method, where all users pay an equal share of the last units produced
(instead of the first units, in the case of serial cost sharing) until the smallest demand
is satisfied, after which the remaining users share equally the cost of the next units,
and so on. In the case of two agents with xl < x2, this gives the following cost shares:

Yl [C(x 1 + X2)- cX -l)]; Y2 = [C(xl +x2) C(x2 - x )]

The Talmudic rationing method (1.10) leads to a somewhat exotic method, except in
the case of two agents, where it coincides with the Shapley-Shubik method. For n = 3
and xl < X2 < X3, the method associated with Talmudic rationing gives the following
cost shares:
Y1 = C(x + X2 + X3 ) - Cx 2 +x3 XI) + 3X),

Y2 = C(Xl +X2 +X3) -- IC( X2 -- X)-C(X) + C(X X2) - IC(XI X3)'

y3 C(X I + X2 + X3 ) + C(x 2 +x 2X 1 )- C(3x 1 )- C(x l1 X2 )+ C(2x X3 ).

We conclude this subsection by a characterization of serial cost sharing within the set
M (CR,ADD). All methods in this set meet the Stand-Alone bounds when the cost
function is either convex or concave (Corollary to Theorem 2.2), but they typically
fail the Universal Bounds (2.14). The Shapley-Shubik method meets the lower bound
[because in the sum (2.19) the term with S = 0 has weight l/n] but fails the upper
bound (even for n = 2).
The universal upper bound is a key ingredient in the characterization of the serial
method; yet it is not sufficient to single out this method in M (CR,ADD).
Consider the counterpart of Zero-Consistency for rationing methods [property (1.6)]:

{Xi 0} {(N, C,x) = 0 and (N, C, x)[N \ i] = W(N\i, C, X[N \ i])} (2.21)
for all N, C,x, i.
Within M (CR,ADD) this property is isomorphic to the axiom (1.6). It is a very
mild requirement, met by all cost-sharing methods discussed in Section 2 [with the
exception of some methods allowing for negative cost shares: see Subsection 2.5.1].
326 H. Moulin

In order to pin down the serial method, we strengthen Zero-Consistency by allowing

the removal of a non-paying agent, provided we make sure to serve his demand (that
could be strictly positive). Given a method in M, the property is stated as follows:

{c(i(N, C,x) = O} = {pf(N, C,x)[N \ i = p(N\i, C,x[v \ i]) where C(z) = C(z +xi)}
for all N, C, x, i.
(2.22)
(Note that C will have a jump at 0 if C(xi) is positive; but the universal lower bound
guarantees C(xi) = 0).
The last ingredient is the unobjectionable equity requirement called Ranking.
xi < xj y < Yij. Note that the Ranking axiom (1.2) for rationing methods conveys
the same idea (is even written in the same way) but is not equivalent via the linear
isomorphism. If the rationing method r meets Ranking, the corresponding cost-sharing
method may not do so.
Theorem 2.3: [Moulin and Shenker (1994)]. Serial cost sharing is characterized
by the combination of the five axioms Constant Returns, Additivity, Universal
Bounds (2.14), Ranking, andproperty (2.22).

2.5. Variants of the model andfurther axioms

2.5.1. Distributivity [Moulin and Shenker (1999)]

The Distributivity axiom expresses the commutativity of the computation of cost shares
with respect to the composition of cost functions:

Distributivity (DIS):
T(C o C2 ,x) = q(C 1, (C 2,x)) for all C l, C2 G D, and all x.

The addition of cost (or production) functions corresponds to technologies operating

in parallel: a given demand of output (resp. a contribution of output) yields two types
of costs, e.g., advertising costs and production costs, (resp. enters two production
functions). Their composition corresponds to technologies running sequentially:
x - y2 = C 2(x) yt = Cl(y2), the input of C 2 is the output of C1 (and a similar
interpretation if C 1, C2 represent production functions). Both axioms, Additivity and
Distributivity, allow us to decompose the computation of cost shares if the cost or
production function itself is decomposed.
One consequence of Distributivity (with no counterpart in the case of Additivity) is
reversibility offairness.

y = p(C,x) > x = p(C- L,y) for all x,y.

Given a pair (x, y) with one profile of outputs and one profile of inputs, we can either
take the profile of demands as given and check that y is the corresponding fair profile
Ch. 6: Axiomatic Cost and Surplus Sharing 327

of costs shares for the given cost function, or we can take the vector y as given and
check that x is fair for the given production function. These two tests are equivalent
for a distributive method.
Distributive methods include average cost sharing, serial cost sharing, as well as any
incremental method. Yet the Shapley-Shubik method (2.19) is not distributive, and in
fact a proper convex combination (with fixed coefficients) of distributive methods is
never distributive!
Moulin and Shenker (1999) characterize the rich family of additive and distributive
methods (meeting Constant Returns). In this family, average cost sharing is the only
self-dual method (a result related to Proposition 1.6), and serial cost sharing is the
only method meeting the universal lower bound (or upper bound) (2.14).

2.5.2. Negative cost shares and the decreasing serial method

In some contexts it makes sense to allow negative cost shares (yi < 0) or to charge
for a null demand (xi = 0 and yi < 0). Suppose marginal costs increase (as in the
monopsonist example of Section 2.1). Then an agent who demands little or nothing
(who refrains from demanding much) is helping the agents with a large demand,
so we may want to compensate him by giving him some money (paid for by other
agents). Symmetrically, consider an output-sharing problem and suppose marginal
returns decrease. Think of the "tragedy of the commons" story: input is fishing effort
and output is the total catch in the common property lake. Then an agent who refrains
from adding more input may argue that she deserves a share of total catch (and end up
with xi = 0 and yi > 0). Note that the duals of the two stories above, where we switch
from cost sharing to output sharing or vice versa, do not ring plausible. To punish an
agent who does not work if the production function is convex (Yi < 0 for xi small), or
to charge one who does not demand any output if the cost function is concave, crosses
the line of acceptable coercion by the mechanism designer!
The decreasing serial cost-sharing method [de Frutos (1998), Suh (1997)] follows
exactly the formulas (2.11), (2.12) and (2.13), except that individual demands are
arranged in decreasing order: xl > x2 > · > x, (so that the sequence x i is decreasing,
too). With two agents and xl > x2:

l 2C(2xl); Y2 = C(xl +x 2 )- C(2x).

If C is strictly concave, yi is positive whenever xi is zero and xj is positive; on the other

hand, no agent receives a negative cost share [this is clear in the case n = 2 and can be
checked in general on Equations (2.1 1) and (2.12)]. If C is strictly convex, yi is negative
whenever xi is zero and xj is positive. The decreasing serial cost-sharing method fails
both universal bounds (2.14) and has not received an axiomatic characterization at the
time of this writing.
Hougaard and Thorlund-Petersen (2001) propose an interesting mixture of the
increasing and decreasing serial methods, arguing that we should keep the former if
328 H. oulin

C is convex and the latter if C is concave. Their method is not additive with respect
to cost functions.

2.5.3. Consistency?

The Consistency axiom played a key role for the analysis of rationing methods, but it
is absent from that of cost-sharing methods. Using the linear isomorphism between
rationing and cost-sharing methods, one would like to characterize the subset of
M (CR,ADD) associated with consistent rationing methods. This may even suggest an
appropriate definition of Consistency for general cost-sharing methods. A definition of
Consistency is offered by Tijs and Koster (1998): it suffers from the same drawback as
the definition discussed in Remark 3.1 below, namely it does not work in a domain of
non-decreasing cost functions. A related and equally natural question is to characterize
the subset of M (CR,ADD) associated with the (symmetric) parametric methods
(Section 1.5). Both questions are wide open.

3. Heterogeneous outputs or inputs

3.1. The problem

In the cost-sharing version of the more general model now under scrutiny, each agent i
demands a different good, and the technology specifies the total cost C(xl,x2,
,..., x,).
In the output-sharing version, each agent i contributes the amount xi of an "input i",
and total output F(xl, ... , x,) must be shared among the participants. Thus we identify
"good i" and "agent i".
Examples of such cost-sharing problems include sharing the cost of a network
connecting geographically dispersed users (so the heterogeneity of demand comes from
the heterogeneity of space, as in road networks), or of a telecommunication network in
which the users need different service (e.g., different bandwidth, or different degrees
of reliability in service, or they use the network at different times of the day). Another
example is the cost sharing of a large project (dam, space station) between various
beneficiaries [e.g., power company, farmers, tourism industry, in the dam example:
see Straffin and Heaney (1981)].
Examples of both cost sharing and output sharing are commonplace in the
accounting literature [see Thomas (1977)]. The various divisions of the firm contribute
heterogeneous inputs to a common project, say the launching of a new product: how
should the revenue of the project be distributed among them? The cost-sharing issue
arises when the divisions share a common service, such as the central administration
unit.
The main simplifying assumption of the current model is that each agent demands
exactly one output good (or contributes exactly one input). On the other hand, the
Ch. 6: Axiomatic Cost and Surplus Sharing 329

domain of cost (or production) functions is very general: C(0) = 0 and C non-
decreasing in each xi, are the only restriction we impose when the variables are discrete
(Sections 3.2, 3.3); when xi is a real number, we add some regularity conditions.
The mathematical complexity of the models raises significantly above that in
Sections 1 and 2. We look first at the case of binary demands (each xi is 0 or 1) in
Section 3.2, i.e., the classical theory of values for cooperative games with transferable
utility. We consider variable demands of indivisible goods (each xi is an integer) in
Sections 3.3 and 3.4, and finally variable demands of divisible goods in Sections
3.5 and 3.6.
In Sections 3.2 through 3.6, we look at additive methods only, as we did in most
of Section 2. We extend the isomorphism between rationing methods and additive
cost-sharing methods (Theorem 2.2): in the case of heterogeneous goods, the set of
rationing methods is identified with the extreme points of the set of additive methods
meeting the Dummy axiom (Theorems 3.1 and 3.3).
The Shapley-Shubik cost-sharing method, and its asymmetric counterparts, the
random-order values, emerge forcefully from the axiomatic discussion. Shapley's
characterization result in the context of binary demands (Proposition 3.1) now has
company in the variable-demand model, whether demands are integer valued or real
valued (see Corollary 2 to Theorem 3.1 and Corollaries 1 and 3 to Theorem 3.4).
The two other prominent methods are the Aumann-Shapley pricing method,
extending average cost sharing to the context of heterogeneous goods, and the additive
extension of serial cost sharing: they are discussed in Sections 3.3 through 3.6 and
characterized in Section 3.6 (Corollaries 2 and 3 to Theorem 3.4).
Up to 1995, the literature on cost sharing with variable demands was unanimously
arguing for the Aumann-Shapley method. The initial axiomatic characterization by
Billera and Heath (1982) and Mirman and Tauman (1982) [see also Billera, Heath
and Raanan (1978)] was refined in several ways [Tauman (1988) is a good survey].
One version of this result is in Corollary 2 to Theorem 3.4. Moulin (1995a) spells
out a critique of the Aumann-Shapley method based on the properties of Demand
Monotonicity and Ranking. The former says that the cost share of an agent should not
decrease when his demand of output increases, ceteris paribus. The latter says that,
when all goods enter symmetrically in the cost function, the ranking of individual
cost shares is the same as that of individual demands.
Both properties DM and RKG are compelling when each good is identified with a
different agent. They are less compelling if the demand of good i aggregates many
small individual demands, which is the standard interpretation in the literature on
the Aumann-Shapley method. In this survey we stick to the first interpretation and
emphasize the critique of the AS method. In turn this pushes the Shapley-Shubik and
serial methods to the forefront.
Additivity of cost shares with respect to the cost function, the main assumption
throughout Sections 2 and 3, is a powerful mathematical tool, yet not a compelling
normative requirement. Additivity narrows down the set of cost-sharing methods
drastically, thus bringing a number of impossibility statements when we require other
330 H. Moulin

properties with more normative appeal: an example is the combination of Demand

Monotonicity and Average Cost for Homogeneous Goods (see Proposition 3.3 and
Corollary 2 to Theorem 3.4). When the impossibility hurts, the first axiom to go should
be Additivity. The literature on non-additive methods is reviewed in Section 3.7: it
contains very few papers but its potential for growth is huge.

3.2. Binary demands: the Shapley value

This is the model of the classical cooperative games with transferable utility where
the only restriction is our assumption that the cost function is non-decreasing.
A binary cost sharing problem is a triple (N, C,x) where N is a finite set of
agents, C is a non-decreasing function from {0, 1}N into R1+such that C(0) = 0, and
x = (xi)i EN is a profile of demands, where each xi is 0 or 1.
For convenience, we denote the vector of demands x as a, possibly empty, subset S
of N: xi = 1 iff i E S. Thus the cost function C associates to each coalition S, S c N,
a number C(S), interpreted as the cost of serving all agents in S and only them. Our
assumptions on C are:

C(0)= 0; S C T C(S) < C(T) for all S, T C N.

A solution to the binary cost-sharing problem (N, C, S) is a profile of cost shares

y = (yi)i E N, where each yi is a real number and:

yi> 0 for all i, E yi C(S).

iEN

A binary cost-sharing method is a mapping qp associating to any problem (N, C, S) a

solution y = (N, C, S).
The idea of sharing costs in proportion to demands reduces in this model to dividing
equally C(S) among all agents in S (and charging nothing to those outside S). However
this method violates the basic principle of reward, namely that cost shares should
reflect responsibilities in generating the costs. A minimal requirement to that effect
is that an agent who "obviously" is not generating any cost should pay nothing. The
Dummy axiom conveys just that idea.
We use the notation OiC(S) = C(S) - C(S\i) for the marginal cost (saving) of
subtracting agent i from coalition S. Of course, OiC(S) = 0 if i S.

Dummy (DUM):
{aiC(T) = 0 for all T C N} = {Yi = q)f(N, C, S) = 0} for all N, S, i, C. (3.1)

An agent is called a dummy for the cost function C if it costs nothing to serve
her, irrespective of the number of other users being served. The egalitarian method
Ch. 6: Axiomatic Cost and Surplus Sharing 331

(yi = C(S)/#(S) if i S, i = 0 otherwise) charges a dummy agent as any other,

therefore it violates Dummy.

Additiuity (ADD):
q(N, C1 + C 2,S)= p(N, C l , S) + (N, C2, S) for all N, Ck, S.

Note that Dummy and Additivity together imply a generalization of the Constant
Returns property (2.2). If C is linear, C(x) = cixi, the method simply "separates"
costs:

qi(N,C,S)=ci.Xi; where xi= I if i E S,

0 if i S.

We denote C (DUM,ADD) the family of cost-sharing methods meeting Dummy

and Additivity. These two axioms place no restriction on the method across different
populations N and N': therefore Proposition 3.1 describes this family in the fixed-
population context, where N is fixed and S varies (note that most of the literature
only looks at the case S = N). Next we introduce a mild consistency requirement
linking the solutions across variable populations; in turn, the corresponding methods
take a natural structure: Theorem 3.1 below.
Given N, an incremental cost-sharing method specifies for each non-empty subset
S of N (including N itself) an ordering o(S) = (l, ... , as) where s = #(S). The
cost shares y = q'(N, C, S) are computed as follows:

Yi= 0 if i 8 S;

Y"o(s) = C({ ()}),

Yo2(s) = 9a 2(s)C({Ol(S), Q2(S)}) = C({o 1 (S), 2 (S)}) - C({( 1(S)}), (3.2)

yoi(s) = d,(s)C({s(S), ... , k(S)})

for all k = 1, ... , s.

A random-order value is a convex combination of incremental methods where the

weights of the combination are independent of C. Denoting by S(S) the set of
permutations of S, a random-order value is written as

y = (N,C,S) = E /(s)c(S)(N,C,S) for all S. (3.3)

a(s) C S(s)

Note that we can choose an arbitrary set of convex coefficients il(S) for each
coalition S. For instance, in S = {1,2, 3} we may choose the incremental method
with ordering 2,1,3, and in S' = {1,2, 4} we may choose that with ordering 1,2,4.
332 H. Moulin

Finally we need an equity property to state Shapley's original characterization. If

two agents affect the cost function symmetrically, we require that they receive the
same share:
Equal Treatment of Equals (ETE).
{C(T U i) = C(T U j) for all T such that i,j ~ T}
{Cp(i(N,C,S) = yqj(N,C,S) for all S, S C N}
for all C, i,j.

Proposition 3.1. Fixed population, Weber (1988). The set of random-order values
coincides with the set C (DUM,ADD) of the cost-sharingmethods meeting the Dummy
and Additivity axioms.

Corollary to Proposition 3.1 [Shapley (1953)1. The three axioms Dummy, Additivity
and Equal Treatment of Equals characterize a single method, namely the Shapley
value; that is, the set C (DUM,ADD,ETE) contains a single method.

p(N, C,S) = t!(s - t ! aC(T U i) for all i C S,

tT: Trs\i (3.4)
#(T) - t

(p-(N,C,S) - 0 if jS.

Incremental methods [Equation (3.2)] and random-order values [Equation (3.3)]

defined in the fixed-population context may allocate priorities (or weigh the various
priority orderings) inconsistently when S changes. In order to avoid this unpalatable
feature, we must switch to the variable-population context and impose a mild
consistency requirement. We denote by i/ the maximal set from which agents can be
drawn (a finite or infinite set) and by a a priority ordering of V. On each finite set S,
this ordering induces an ordering denoted a(S), and the corresponding formula (3.2)
defines the -incremental cost-sharing method. Similarly, a consistent random-order
value is a convex combination of the a-incremental methods, where a varies over all
orderings of Ki and the coefficients of the convex combination are independent of N,
C and S:
p(N,C,S)= ),,p(s)(NC, S)
AE for allN, C,S. (3.5)
oE S(n)
The following axiom corresponds to the Zero-Consistency property for rationing
methods (1.6): a dummy agent can be removed without affecting the distribution of
costs among the rest of the agents:
Dummy-Consistency (DCY):
{9iC(T) = 0 for all T C N} {(N, C, S)[N \ i] = (N\i, C, S\i) for all S}
for all N, i, C.
Ch. 6: Axiomatic Cost and Surplus Sharing 333

(where the restriction of C to N\i is denoted C as well).

Proposition 3.2. Variable population. The set of consistent random-order values
coincides with the set C (DUM,DCYADD) of the cost-sharing methods meeting
Dummy, Dummy-Consistency and Additivity.
Several alternative characterizations of the Shapley value and the random-order values
have been proposed in the literature. They replace the Additivity axiom by another
powerful requirement; the two most striking results rely on the property of marginalism
and the notion of potential. We describe these two results in the fixed-population
context.
In a random-order value, the cost share of an agent depends only on his marginal
costs diC(T) for the various coalitions containing i. This property, called Marginalism,
is defined as
{&iC'(T) = iC2 (T) for all T C S} = {pi(N, C',S)= (pi(N, C 2, S)} (3.6)
for all N, Ck, S, i.
Loehman and Whinston (1974) and Young (1985a) show that the Shapley value
is characterized by Marginalism and Equal Treatment of Equals. Khmelnitskaya
(1999) shows that the combination of Marginalism, Dummy, and an axiom called
Monotonicity characterizes the random-order values when N contains three agents or
more. The monotonicity requirement is as follows:
{Cl(T) = C 2(T) for all T C N, T S and Cl(S) < C2 (S)}
X {qi(N, C, N) < ,li(N,C 2 , N) for all i E S}
for all N, Ck, S
If we add Dummy-Consistency to this list of requirements, we characterize the family
of consistent random-order values.
The second characterization result concerns the Shapley value alone. Consider the
following potentialfunction:

P(N, C) = (s 1) (n s)! C(S) where n = #(N), s = #(S). (3.7)

SCN

The Shapley value (3.4) can be equivalently written as

qoP(N, C, S) = OiP(S, C) = P(S, C) - P(S\i, C). (3.8)
Thus agent i's share is simply the ith derivative of the potential function. As
second derivatives commute, this implies for all i,j in N:
j (pi(N, C, S) = oDzi j(N, C, S)
<:* qic(N, C, S) - qoi(N, C, S\j) = qj(N, C, S) - qpj(N, C, S\i).
The effect on i's share of removing j is the same as that onj's share of removing i.
334 H. Moulin

Hart and Mas-Colell (1989) show that the Shapley value is fully characterized by
the existence of some potential function P of which derivatives deliver the individual
cost shares as in Equation (3.8) and such that P(O, C) = 0. Equivalently, property (3.9)
alone is enough to characterize the Shapley value.
Remark 3.1. A third characterization [also due to Hart and Mas-Colell (1989)] of the
Shapley value follows from a Consistency axiom. Consider a cost-sharing method 0p.
For any cost function C and agent i, we define a new function as follows:

Ci''(T) = C(T U i) - p(N, C, T U i). (3.10)

Note that this function depends on N as well, but under no charge for null demands
this plays no role.
The idea is that agent i wants "service" but is leaving the jurisdiction of the division
problem; he will pay whatever cost share is recommended by the method, whatever is
the set of other agents who want service as well. The Consistency axiom says that this
move should not affect the allocation of cost shares among the remaining agents:

Consistency.
cq(N, C, S)[N \ i] = (N\i, Ci' , S\i) for all N, C, S, i.

The result is that if we choose the standard value (i.e., the Shapley value) for two-
person problems, the Consistency axiom forces the Shapley value for every set N.
There are two problems with the above axiom. First of all, the definition of the
reduced cost function Ci 4 does not preserve the monotonicity of costs: we may have
S c T and C ' (S) > C'~"(T). In fact this inequality does occur even if p is the
Shapley value itself. As an example, take N = {1,2, 3} and the cost function

C(N)= C(13) = C(23) = 1; C(12) = C(i) = 0 for i = 1,2, 3.

Compute

C3"(1) = C(13) - q3(C, {13})= 1-=

C 3'¢(12) = C(123) - p3(C, {123}) = 1 - _ = .
Thus the only way to make sense of the axiom is to allow for decreasing cost
functions. But in this enlarged domain, the very foundations of our model must be
revised: allowing negative cost shares is compelling (think of an agent whose presence
eliminates all costs); the axioms of Demand Monotonicity, and the Upper and Lower
Bounds (properties (3.30) and (3.15), respectively) must be abandoned, and so on.
The second difficulty is the interpretation of the axiom. It does not represent a clear-
cut reduction of the allocation problem to the subset N\i of agents, because agent i
must still be ready to pay a different share when the set of other agents who want
Ch. 6: Axiomatic Cost and Surplus Sharing 335

service changes. This is in sharp contrast with the Consistency property in the rationing
(or surplus-sharing) problem where agent i puts his money on the table (or takes his
share of the output) and departs without leaving an address: the remaining division
problem can be conducted entirely without him.
The above difficulty applies to all forms of Consistency for the binary model (the
classical cooperative-game framework) such as the concept due to Davis and Maschler
(1965) used to characterize the nucleolus by Sobolev (1975). See Section 3.7. It also
applies to the Consistency for general cost functions proposed by Friedman (1999).

3.3. Variable demands of indivisible goods: the dr model

Each agent i demands xi units of the idiosyncratic good i, xi E {0, 1,2, ... , Xi}, where
Xi is the capacity of this good. For the main representation theorem below, we must
assume that X is finite, although the model makes sense for Xi = +oo as well.
A cost function C is now a mapping from [O,X[N]] (the Cartesian product of the
integer intervals [O,X]) into R+ such that

C(O) = 0 and x < x' = C(x) < C(x').

As in Section 2, the vector x is called the demand profile, x C [O,X[N]]. The definition
of a cost-sharing problem (N, C, x) is now complete. A solution is a vector y in RN
such that

y > 0, EYi = C(x). (3.11)

N

As usual a cost-sharing method (q associates a solution to each problem. This model

generalizes the binary model of the previous subsection where we had Xi = 1 for all i.
Our first task is to generalize Propositions 3.1 and 3.2 to the variable-demand
context. As in Section 3.2, we give first the fixed-population version of the result, based
on the two axioms Dummy and Additivity: Theorem 3.1. Next we give the variable-
population result, with the help of the additional axiom Dummy-Consistency: corollary
to Theorem 3.1.

Dummy (DUM):
{OiC(x) = 0 for all x E [0,X[N]]}
X {p(N, C,x) = 0 for all x E [O,X[N]]}
for all N, C, and all i C N,

where the notation OiC(x) = C(x) - C(xlixi - 1) and aiC(x) = 0 if xi = 0 stands for the
marginal cost when i raises her demand from xi - 1 to xi. The interpretation is as in
336 H. Moulin

Section 3.2: the demand of a dummy agent, no matter how large, does not cause any
extra cost, hence this agent should never be charged.

Additivity (ADD).
(p(N, Cl1,x) + (N, C2 ,x) = (N, C + C 2,x) for all N, Ck,x.

The incremental cost-sharing methods of the binary model (Section 3.2) as well as
of the model with homogeneous goods [Section 2.4, see Equation (2.18)] generalize.
We fix an ordering a of N (recall that N is fixed for the time being) and define
the a-incremental method (or method with priority ordering a) y = OP(N, C,x) as
follows:

Yo = C(x[,J, 0),

Y 2 = C(x[,a 2], 0) - C(x[,], O),

(3.12)
Yo, = C(x[, .. a], 0)- C(x[, .... ], 0) for all i
Y = C(x) - C(X[N \ ),

where (X[T],O) denotes the vector with the same projection as x on T, and zero
on N\T.
The incremental methods obviously meet DUM and ADD, and so do their convex
combinations. Yet there are many more methods in C (DUM,ADD). We construct a
family of such methods, called the path-generatedmethods: these methods are the key
to the representation results below.
Pick a monotonic rationing method r for indivisible goods: their set is denoted aRdd
and they are discussed in Section 1.9. The society N is fixed for the time being so we
write r(t,x) instead of r(N, t,x), where x E [O,X[N]] and 0 < t < XN.
Recall that the path t -- r(t,x) is equivalently described by a sequence s(x)=
{il, i2, .. , i} where agent i appears exactly xi times. To each rationing method r
in Rldd, or equivalently to each family of sequences s(x) (one for each x in [O,X[N]])
we associate the following cost-sharing method y = "pr(N,C,x):
-TN

Yi = Z iC(r(t,x))dri(t,x) for all N,C,x,i, (3.13)

=t-

where dri(t,x) = 1 if i = it is the tth element of the sequence s(x), and dri(t,x) = 0
otherwise.
The cost-sharing method (3.13) is called path-generated because for each x the
cost shares are computed along the path t -- r(t,x), i.e., along the sequence s(N,x) as
follows: C(r(l,x)) is charged to agent il, C(r(2,x)) - C(r(l,x)) is charged to agent i2,
and so on. Definition (3.13) makes it very clear that this method satisfies the axioms
ADD and DUM. As convex combinations respect these two properties, we find that
Ch. 6: Axiomatic Cost and Surplus Sharing 337

any convex combination of path-generated methods do, too. In fact there are no other
methods.
Theorem 3.1: Fixed population [Wang (1999)]. Every cost-sharing method meeting
Dummy and Additivity is a convex combination ofpath-generatedmethods (where the
coefficients may depend on N andx but not on C). No other cost-sharingmethod meets
these two properties. Identifying a rationing method r E add with the path-generated
cost-sharing method (3.13) we write this result as follows:

C (DUM, ADD) = CO[Rdd] X dd = C (DUM, ADD)

(where 2 denotes the set of extreme points of Z, and CO the convex hull).
We turn now to the variable-population framework of the result. As in the binary
model (Section 3.2) we require that dropping a dummy agent from the society be of
no consequence:

Dummy Consistency (DCY):

{,iC(x) = 0 for all x E [0,X[N]]}
{ {(N, C,x)[N\i] = (N\i, C, X[N \i]) for all x E [O,X[N]]}
for all N, C, i.

The following fact is easy to prove: a path-generated method is dummy consistent

if and only if the corresponding rationing method is consistent. It is easy to
check that Consistency of r amounts to the following property of the generating
sequences s(N,x): the sequence s(N\i,x[N \i]) obtains from s(N,x) by removing all
occurrences of i. Thus we call a method generatedby consistent paths if it is derived
from a consistent rationing method via Equation (3.13).
Corollary 1 to Theorem 3.1. Variable population. Every cost-sharing method in
C (DUM,DCYADD) is a convex combination of methods generated by consistent
paths. There are no other methods in C (DUM,DCYADD):

C (DUM, DCY, ADD) = CO[Rdd(CSY)] Rdd(CSY) = C (DUM, DCY, ADD).

Note that a convex combination of paths can be interpreted as one of the probabilistic
rationing methods discussed in Section 1.9. This interpretation is especially useful for
Examples 3.2 and 3.3 below.
We illustrate the results by three crucial symmetric cost-sharing methods.
338 Hi. oulin

Example 3.1. The Shapley Shubik method: The arithmetic average of all incremental
methods is also called the Shapley-Shubik method [Shubik (1962)], namely

(piS(N CX)=
s0-
s!(n-s- 1)!
n
z
S: SC N\i
{C(x[sui],O)-C(x[s],O)}. (3.14)
#(S)=s

This method is not path-generated; it is a proper convex combination of path-generated

methods, namely the incremental methods. Contrast this with the Shapley-Shubik
method for homogeneous goods (2.19), that is path-generated (like all methods in
M (CR,ADD): see the discussion in Section 2.4). ·
Remarkably, the Shapley-Shubik method can be characterized by one single
additional axiom within C (DUM,ADD), namely a lower bound on individual
cost shares that depends only on N, C and xi:
Lower Bound:
cpi(N, C, x) > Oc) C(xi, O[N \ i]) for all N, C, x, i,
where O(C) is the number of non-dummy goods in C.
This bound generalizes to the heterogeneous-goods context the universal lower bound
(2.14) of Section 2. In the homogeneous-good context, the Lower Bound is met by
many costs-sharing methods, such as serial cost sharing, Shapley-Shubik and more.
Its impact in the current model is much more dramatic.
Corollary 2 to Theorem 3.1. The Shapley-Shubik method is the only method in
C (DUM,ADD) meeting the Lower Bound axiom.

Example 3.2. The Aumann-Shapley method (dr model) [Moulin (1995a), de

Nouweland and Tijs (1995)]: This is the discrete version of the Aumann-Shapley
cost-sharing method for divisible goods. It plays a major role in the next subpart.
For a given problem (N, C,x), the method is the uniform average of all path-
generated methods, in other words it gives an equal weight to each path between
Oand x. Hence the Aumann-Shapley method corresponds, in the representation (3.13),
to the random proportional rationing method described in Section 1.9.
Straightforward computations give the formulas for the AS cost shares. For any
vector t in NN, we use the notation
(t)=)!
f ti !'
N
for the number of monotonic paths from 0 to t in [0, t]. Then we have

eAs(iN, C, (t
(1) ) (t t) C(t), where ti =xi- ti
C [Ox] t t
(3.16)
(with the convention X(O) = 0). ·
Ch. 6: Axiomatic Cost and Surplus Sharing 339

An important feature of the AS method is that it coincides with average cost sharing
when the goods are homogeneous, that is to say when C takes the form C(x) = C*(xN).
Sprumont and Wang (1998) argue that the AS method is the most natural extension
of proportional cost sharing to the context of heterogeneous goods. Calvo and Santos
(2000) propose an alternative characterization.
Example 3.3. Serial cost sharing (dr model) [Moulin (1995a)]: In the homogeneous-
good model (Section 2), serial cost sharing is associated with the Uniform-Gains
rationing method via Equation (2.16). Similarly, in the case of heterogeneous
goods serial cost sharing is associated, via Equation (3.13), with the Fair Queuing
(probabilistic) method described in Section 1.9. The corresponding cost shares are as
follows in the case of two agents:
if x l < X2:
l =1 (C,x) = C(xl,xl) + - [C(t, t - 1)- C(t - 1, t)],
t=

Y2 = C(x)-Yl = C(x)- C(xl, xl)+ E [C(t- , t)- C(t, t- 1)].

t=l

3.4. Demand Monotonicity in the dr model

Demand Monotonicity (DM):

Yi = (ti(N,C,x) is non-decreasing in xi, for all N, C,x, i.
This Demand Monotonicity can be viewed as a mild incentive compatibility require-
ment. Absent DM, the cost-sharing method is vulnerable to "sabotage" when the output
goods are freely disposable: an agent can artificially raise her demand, throw away the
excess good and receive a smaller bill!
Thefixed-path cost-sharing methods are derived from the fixed-path rationing methods
(Section 1.8) via Equation (3.13). Recall that for a given society N and a capacity X,
for each agent i, a fixed-path rationing method is defined by a single path y(N) from
0 to X[N]:
For all i
t 7yi(N, t) is non-decreasing from [O,XN] onto [0, Xi]
and Z yi(N, t) = t for all t c [O,XN].
N

The corresponding cost-sharing method reads as follows, where the variable N is

omitted from y for simplicity:

yi = qp1(N, C,x) = E aC((t) A x)

t: yi(t I) < i(t) <xi
where z A x stands for (z A x)i = min{zi,xi}.
340 H. Moulin

This formula makes it clear that a fixed-path cost-sharing method is demand-

monotonic: increasing xi to x' enlarges the set of indices t at which the marginal
cost OiC(y(t) A x) is charged to agent i, moreover y(t) A x = y(t) A (x[,\ i],x') for
all t in the initial set.
Because DM is preserved by convex combinations, every convex combination of
fixed-path cost-sharing methods (where the coefficients are independent of C and x)
meets DM as well. Examples include the Shapley-Shubik and serial methods.
Theorem 3.3 below states the converse statement. Before writing this important result,
we note that the Aumann-Shapley method (Example 3.2) is not demand-monotonic.
Proposition 3.3. (Moulin (1995a)]. Let be a cost-sharing method in C (DUM,ADD)
that coincides with average cost sharing when the goods are homogeneous, i.e., when
the cost function takes the form C(x) = C*(XN). Then cp is not demand-monotonic.

Theorem 3.2: Fixed population.

(i) A path-generatedcost-sharing method is demand-monotonic if and only if it is a
fixed-path method.
(ii) A cost-sharing method meets Dummy, Additivity and Demand Monotonicity if and
only if it is a convex combination of fixed-path methods (where the coefficients
depend on N but not on C or x).

C (DUM, ADD, DM) = CO[Rdd (fixed paths)].

In the variable-population formulation of theorem 3.2, the extreme points of the

relevant set of methods are the consistentfixed-path cost-sharing methods, for which
the rationing method defined by the fixed paths N -i y(N) is consistent. As noticed in
Section 1 [see Equation (1.21)], this means that y commutes with the projection over
subsets, namely

For all N, S: S C N = y(N)[s] = y(S). (3.17)

Corollary to Theorem 3.2. Variable population. Every cost-sharing method in

C (DUM,DCYADD,DM) is a convex combination of consistent fixed-path methods.
No other method meets these four axioms.

C (DUM, DCY, ADD, DM) = CO[dd (consistent fixed path)].

Remark 3.2: Two characterizations of the random-order values. The random-

order values are the convex combinations of incremental cost-sharing methods (3.12).
They are an important subset of C (DUM,ADD,DM) because of their simplicity. Two
characterizations of this subset within C (DUM,ADD) have been proposed in the
Ch. 6: Axiomatic Cost and Surplus Sharing 341

dr model. The first [Wang (2000)] uses one additional axiom, a dr counterpart of the
Unit-Invariance axiom of the rr model to be described in Section 3.6. The second
[Sprumont (2000)] relies on Demand Monotonicity and the property Informational
Coherence.
Unit Invariance is defined below by Equation (3.26), with the operator r given by
Equations (3.24) and (3.25). Because i o = T, the axiom is written equivalently
as

qp(N, TfA(C),x) = Tp(N, C, IA(x)).

In the dr model, the domain of C is NN, so if A is an integer, the function ri/(C) is

well defined over NN [see Equation (3.25)]. The dr version of Unit Invariance (which
Sprumont calls Measurement Invariance) requires the above equality to hold for every
N, C,x, i and every integer A.
In order to define Informational Coherence, consider the following reformulation of
Unit Invariance in the rr model. For every vector kt, x in IIRN, we write I ® x their
coordinatewise product: (u ® x)i = puixi. Define:

C(t X x) = C(x) for all x (3.18)

(where ui is positive for all i). Unit Invariance (3.26) is equivalent to the following
property:

(N, C, x) = c(N, C, /y x) for all t and all N, C,x. (3.19)

In the dr model, the rescaling vector yuis integer-valued, hence for a given C and a
given u, property (3.18) does not define a unique cost function CA: it only specifies
its value on a subset of NN. The axiom Informational Coherence requires that for all
problems (N, C,x) and all (integer-valued) rescaling vectors , there exists at least
one cost function C ; meeting properties (3.18) and (3.19).
Ironically, the dr version of the Aumann-Shapley method fails both Informational
Coherence and Unit Invariance (dr version); by contrast, in the rr model Unit Invariance
is one of the keys to the characterization of the Aumann-Shapley method (Corollary 2
to Theorem 3.4). See Sprumont (2000) for more comments and interpretations.

3.5. Variable demands of divisible goods: the rr model

The only change from the model in the previous subsection is that the output goods
are perfectly divisible and the demand profile x is in IIRN. The cost function C remains
non-increasing, and C(0) = 0. However we must now assume that C is sufficiently
regular to apply the general results about linear operators in functional spaces.
We assume throughout that the cost function C is twice continuously differentiable.
This is not the only conceivable regularity assumption on C.
342 H. Moulin

All the dr axioms discussed in Sections 3.3 and 3.4 (with the exception of
Remark 3.2) are defined in exactly the same way in the rr model. For instance, agent i
is a dummy in C iff the partial derivative OiC is identically zero, which amounts to
saying that C is independent of the variable xi, as in the dr model.
Much of the theory in the rr model parallels that in the dr model: this is true for the
representation of the sets C (DUM,ADD) and C (DUM,DCYADD) (Theorem 3.3) and
for the impact of DM in these two sets (Theorem 3.4). The only new feature is the Unit-
Invariance axiom that has no counterpart in the dr model (see, however, Remark 3.2).
In this subpart we state Theorem 3.3 and present the main examples. The discussion
of Demand Monotonicity and Unit Invariance are the subject of Section 3.5.
As in the dr model, the key to the representation of C (DUM,ADD) is the path-
generated methods. We denote by R,rr the set of monotonic rationing methods (with
divisible claims and shares) studied in Sections 1.2 through 1.8. Recall that a method r
in this set is a mapping y = r(N, t,x) where t,xi,yi vary in l-i, where yN = t and ri is
non-decreasing in t. These assumptions imply that ri is continuous in t as well. The
path-generatedcost-sharing method qp associated with r G R7r is defined as follows:
X'N

i = (pi(N,C,x) = iC(r(N,t,x))dri(N, tx) for all N,C,x,i, (3.20)

where the integral is the Stieljes integral of a continuous function with respect to a
monotonic function, and dri represents the derivative of ri(t,x) with respect to t. This
definition corresponds to Equation (3.13) in the dr model.
Theorem 3.3: Fixed population [Friedman (1999), Haimanko (1998)]. A cost-
sharing method satisfies Dummy and Additivity if and only if it is an (infinite) convex
combination of path-generatedmethods.

C (DUM, ADD) CO[,,r] ,,rr= C(DUM, ADD).

In the above statement, a convex combination may be infinite; in other words it is a

positive probability measure over the (illfinite-dimensional) set of monotonic rationing
methods [see Friedman (1999) for details]. By contrast, in the dr model the set Rdd
is finite. This difference notwithstanding, Theorem 3.3 is the exact counterpart of
Theorem 3.1.
In the variable-population context, the extreme points of the relevant set of methods
are the methods generated by consistent paths: they are derived via formula (3.20)
from a consistent rationing method; equivalently, the paths y(N,x) commute with the
projection operator

For all N,S, S C N: y(N,x)[s] = y(S,x[s]).

Hence the counterpart of Corollary 1 to Theorem 3.1.
Ch. 6: Axiomatic Cost and Surplus Sharing 343

Corollary 1 to Theorem 3.3. Variable population.

(i) A path-generatedcost-sharing method is Dummy-Consistent if and only if it is
generated by a consistent path.
(ii) A cost-sharing method meets Dummy, Dummy-Consistency and Additivity if and
only if it is an (infinite) convex combination of methods generated by consistent
paths.
We can now attach a cost-sharing method to any rationing method discussed
in Section 1. For instance, the contested-garment method (Section 1.4) yields the
following (path-generated) cost-sharing method among two agents:
xl <X2 a
x1/2 x1/2
yl = fS d C(t, t) dt + f 1 C(xl - t, x 2 - t) dt,
o 0
x1 /2 x1/2
Y2 = f 2 C(t,t)dt+ [C(x,x 2 - I X)-C(xl,IxI)] + f 02 C(x -t,x 2 -t)dt
o o

(and a symmetrical formula if x 2 < xl).

From the rich family of monotonic rationing methods, only three generate a cost-
sharing method that has received some independent attention in the literature. These
are proportional rationing, uniform-gains rationing and priority rationing (and convex
combinations of the latter).
We start by the family of random-order values, and their symmetric member
the Shapley-Shubik method. They are defined exactly as before: formulas (3.12)
and (3.14) are unchanged. So is the characterization of the Shapley-Shubik method by
the Lower-Bound axiom, Dummy and Additivity: Corollary 2 to Theorem 3.1 remains
true, word for word.
Next we turn to the method generated by proportional rationing.
Example 3.4. The Aumann-Shapley method (rr model) [Aumann and Shapley
(1974)]: This method has been the subject of the most voluminous research, and its
characterization by Billera and Heath (1982) and Mirman and Tauman (1982) initiated
the literature on cost sharing with variable demands.
The AS method is path-generated: its path is the simplest path between 0 and x,
namely the straight line; equivalently, the AS method is derived from the proportional-
rationing method via formula (3.20). Thus it generalizes average cost sharing for
homogeneous goods (Section 2) to the heterogeneous goods model. The method is
written as follows:
X x xXi I

Yi/ oic y/t

=
x C i dt= OiC(-
XX x) dt=xi iC(tx)dt. (3.21)
(3.21)
0 0 0

It can be shown that this AS method is the limit of the AS methods in the dr model
(Example 3.2). Given N and x, we approximate each interval [O,xi] by a sequence of
344 H. Moulin

increasingly fine discrete grids, and define a sequence of dr cost-sharing problems

by simply restricting the initial cost function to the profiles of the grid. Then
the AS cost shares (3.18) of the discrete approximating problems converge to the
cost shares (3.21). ·

Example 3.5. Serial cost sharing (rr model) [Friedman and Moulin (1999)]: As in
the homogeneous-goods model, serial cost sharing is generated by the uniform-gains
rationing method. Thus it is path-generated. Using the notations (a A b)i = min{ai, bi}
and e = (1, 1, ... , 1), the uniform-gains method writes
r(t, x) = )Le A x where (e AX)N = t.

Change the variable t in Equation (3.13) to X: the latter varies from 0 to maxi xi when
t varies from 0 to XN; moreover OiC(Xe A x) = 0 whenever 3`> xi. Hence the integral
formula for serial cost sharing:
Xi

Yi =f OiC(te Ax)dt. (3.22)

o
For instance, take N = {1,2} and x such that xl < X2:
X X1I

Y = a c(t,
tt)d;
t)d; Y= C(xl,x 2 )-y1 = 02 C(t, t)dt+C(x,x 2)-C(xl,x).
o o
The conscientious reader will check that (3.22) yields the cost-sharing formula (2.13)
if the cost function C takes the form C(x) = C*(xN). ·
We look now at some examples of the cost function C for which our three methods
are both easy to compute and interestingly different.
Consider first a Cobb-Douglas cost function. Fix some positive numbers ai and
define:
C(X i... Xn) = Xl X 2 ... Xal

The Shapley-Shubik method splits total cost equally between all users; similarly, the
Aumann-Shapley method assigns the fraction ai/aN to user i. Thus both methods
disregard the relative size of individual demands, which creates a strong "tragedy of
the commons" effect. The serial method yields a more appealing set of cost shares,
notwithstanding the fact that it relies on the interpersonal comparison of individual
demands. For instance,

xl < xi for all i * yl = I al .

aN
Consider next the cost function of an excludable public good: C(x) = C*(maxixi).
Note that this function is not continuously differentiable, so we need to approximate
Ch. 6: Axiomatic Cost and Surplus Sharing 345

the given function by regular cost functions and take the limit. This is straightforward.
Examples include the cost sharing of a capacity, as in the celebrated airport landing
game [Littlechild and Owen (1973)] where each user requests a certain length of
runway suitable for his own airplanes.
The Shapley-Shubik and serial cost-sharing methods coincide in this example.
Assume xl < x 2 < · · < xn, then the cost shares are

i n-I
C*(xl ),
Y2 = Y + [C*(x 2 )-C*(XI)l = lC*(X2) C*(XI)

i-
Yi = Yi- 1+ n-il [C*(xi) - C(xi- )]= ni C(xi) i- . C*(x)
(3.23)
These cost shares make good sense: they separate total cost C*(xn) into n components
C*(xi) - C*(xi_ l) and split the cost of each component between the agents who "use"
this component.
By contrast, the Aumann-Shapley method recommends an unpalatable division of
costs:

(xi < x, for i 1, ... , n -l} {nY = C*(xn), yi = 0 for i = 1, .. , n- 1}.

The agent with the largest demand bears the full cost: this follows from Equation (3.21)
because on the straight line [0,x] the only non-zero partial derivative is OC. These
cost shares change discontinuously whenever the identity of the largest demander
changes; they make it extremely costly to raise one's demand by a very small amount.
The next example of the function C is easier to interpret as a production function,
i.e., in the surplus-sharing model, so we write it as F(x) = F*(minxi). The inputs
of the various agents are perfect complements in the production of the single output
[as in several coordination games of a macroeconomic inspiration; see, e.g., Bryant
(1983)]. The Shapley-Shubik and serial cost shares again coincide: they simply split
total output equally, yi = F*(minxj)/n, for all i. This time the Aumann-Shapley method
gives the full surplus to the agents with the smallest input contribution!
{xl < xj for allj = 2, ... , n}• y = F*(xl), yi = 0 for i = 2, ... , n.
Thus agent 1's share drops to zero when he raises his input contribution above the next
smallest contribution, in sharp contradiction of Demand Monotonicity.
The above two examples make a powerful point against the Aumann-Shapley
cost sharing: it makes little sense when the cost or production function exhibits strong
complementarities. In axiomatic form, this critique of the AS method hinges on two
axioms, Demand Monotonicity and Ranking:
Ranking:
{C symmetrical in all variables and xi < x} = { pi(N, C,x) < pj(N, C,x)}
for all N, C, x, i,j.
346 H. Moulin

Ranking is a strengthening of Equal Treatment of Equals. When all goods affect the
cost function symmetrically, their quantities are structurally comparable and Ranking
is a compelling equity requirement.
Serial cost sharing and the Shapley-Shubik methods both meet Ranking and
Demand Monotonicity. In the case of the function F(minixi), the Aumann-Shapley
method (strongly) violates both axioms. Note that the dr version of the Aumann-
Shapley method (Example 3.2) also violates Ranking.

3.6. Unit Invariance and Demand ionotonicity in the rr model

We explore the impact in the set C (DUM,ADD) [and in C (DUM,DCY,ADD)]

of two important requirements: Demand Monotonicity and the new axiom Unit
Invariance. Theorem 3.4 gives a complete answer to both questions, and leads to three
characterization results of, respectively, the Shapley-Shubik, Aumann-Shapley and
serial methods.
Given an agent i and a positive number A, we denote by the L-rescaling of the
ith coordinate:

For all x c R.¥ Tr(x) = (x[N\i], Xi) (3.24)

Given a cost function C, the A-rescaling of xi transforms C into Tr(C):

Tr(C)(x) = C(TI/)(x)) for all x. (3.25)

Unit Invariance (UI).

cp(N, C,x) = cp(N, Tr(C), Tr(x)) for all N, C,x, i and A > 0. (3.26)

Unit Invariance says that changing the unit in which a particular good is measured
would not affect cost shares: whether I measure a quantity of corn in kilos, pounds
or bushels should not matter. This is compelling when the different goods entering
the cost function are not only different but also noncomparable (e.g., corn and fruits).
Less so if the goods are of the same nature, but do not enter symmetrically in the cost
function. Even less so if the goods affect C symmetrically.
Unit Invariance is violated by serial cost sharing, as its generating path (namely the
uniform-gains path) along the diagonal is not invariant by resealing.
As in the dr model, it is sufficient to analyze the impact of the two axioms DM and
UI on the extreme points of the (convex) sets C (DUM,ADD) and C (DUM,ADD,DM),
namely the path-generated methods and the methods generated by fixed paths:
Theorem 3.4 says that there are no other extreme points in the relevant set of methods.
A path-generated method is demand-monotonic if and only if it is a fixed-path
method, namely a method derived via formula (3.20) from a fixed-path rationing
method (Section 1.8). Recall that such a rationing method is constructed, via
Ch. 6: Axiomatic Cost and Surplus Sharing 347

formula (1.19), from a family of paths y(N) joining 0 to X[N], where Xi is agent i's
capacity, Xi < +oo. Hence the corresponding cost-sharing method is written as
follows:
Si

Yi = (iq(N, C, x) = f iC(y(N, s) A x) dyi(N, s) (3.27)

0
where si is such that yi(N, si) = xi.

Naturally, a fixed-path cost-sharing method is Dummy consistent if and only if the

corresponding fixed-path rationing method is consistent, or equivalently if the fixed
paths N y(N) commute with the projection operator: property (3.17).
A path-generated method is unit-invariant if and only if the corresponding family
of paths (N,x) - y(N,x) are such that:

y(N, rT(x))= 'r(y(N,x)) for all N,x,i, i2,

or equivalently, the corresponding rationing method is such that

{For all t there exists t' such that : r,(r(N, t,x)) = r(N, t', rT(x)}
for all N, x, i, X.

In this case we speak of a unit-invariant path-generated method.

Theorem 3.4: [Friedman (1999)].
(i) A cost-sharing method satisfies Dummy, Additivity and Demand Monotonicity if
and only if it is an (infinite) convex combination offixed-path methods:

C (DUM, ADD, DM) = CO[TZrr (fixed path)].

(ii) A cost-sharing method satisfies Dummy, Additivity and Unit Invariance if and
only if it is an (infinite) convex combination of unit-invariantpath methods:

C (DUM, ADD, SI) = CO[LZrr (SI)].

(iii) C (DUM, DCY, ADD, DM) = CO[7Zrr (consistent fixed path)].

(iv) C (DUM, DCY, ADD, SI) = CO[R,,rr (CSY, SI)].
As in the dr model, the random-order values are the convex combinations of the
incremental cost-sharing methods (3.12) (where the coefficients are independent of
C and x). They have a very simple characterization.
Corollary 1 to Theorem 3.4. [Friedman and Moulin (1999)]. A cost-sharing method
satisfies Dummy, Additivity, Demand Monotonicity and Unit Invariance if and only if
it is a random order value.
348 H. Moulin

If we add Dummy-Consistency to the list of requirements, we characterize the set of

consistent random-order values: these methods are described as in the binary model
[see Equation (3.5)]. Similarly, if we add Equal Treatment of Equals to the list of
axioms in Corollary 1, we characterize the Shapley-Shubik method.
Remark 3.2 in Section 3.4 describes two characterizations of the random-order
values in the dr model; they are the counterpart of Corollary 1 in the case of indivisible
units of demand.
Remark 3.3. Wang (2000) offers a characterization of the random-order values and
the Shapley-Shubik method where the two axioms DM and UI are replaced by a single
property called ordinality. The latter strengthens UI by requiring the invariance of
cost shares when we change the measurement of any one good in a nonlinear way.
Let A denote an increasing (and differentiable) one-to-one mapping of IR+into itself.
We generalize the definition of A,-rescaling [Equations (3.24), (3.25)] as follows:

r(x) = (X[N\ i], (X,)); T1(C)(X) = C(T, (X)) (3.28)

The ordinality axiom is defined exactly as Unit Invariance, with A now varying over
all increasing bijections of Ri+.
The set of random-order values is characterized by the combination of Dummy,
Additivity and Ordinality. Adding Equal Treatment of Equals singles out the Shapley-
Shubik method.
We now turn to the celebrated characterization of the Aumann-Shapley method.
Corollary 2 to Theorem 3.4. [Billera and Heath (1982), Mirman and Tauman
(1982)].
(i) The Aumann-Shapley cost-sharing method is characterizedby the four properties
Dummy, Additivity, Unit Invariance and the following.

Average Cost fobr Homogeneous Good (ACH)

C(x) = C*(xN) for all x} m {q(N, C,x) = ) x for all x} (3.29)
XN (3.29)
for all N, C.

(ii) There is no cost-sharing method in C (DUM,ADD) meeting Average Cost for

Homogeneous Goods and either one of Demand Monotonicity or Ranking.
Note the tension between the axioms UI and ACH, the former bearing on the case
of heterogeneous, noncomparable goods, the latter on the case of identical goods.
Statement (ii) in Corollary 2 includes the rr version of Proposition 3.3. In fact, the
incompatibility of ACH and RKG [in C (DUM,ADD)] holds true in the dr model as
well. The proof uses exactly the same numerical example as that of Proposition 3.3.
Remark 3.4. Young (1985b) offers an alternative characterization of the Aumann-
Shapley method using neither Dummy nor Additivity. Instead, he proposes the
Ch. 6: Axiomatic Cost and Surplus Sharing 349

Strong Aggregation Invariance (SAI) axiom that considerably strengthens ACH, and
a Symmetric Monotonicity (SM) axiom combining the intuition of Marginalism (3.6)
with the interpersonal comparison of marginal costs.

Corollary 3 to Theorem 3.4.

(i) The serial cost-sharing method is characterized by the four properties Dummy,
Additivity, Demand Monotonicity and

Upper Bound (UB):

{C is symmetric in all xi} = {q(i(N, C,x) < C(xie) for all x, i} (3.30)
for all N, C.
(ii) The serial cost-sharing method is characterized by the four properties Dummy,
Additivity, Demand Monotonicity and

Serial cost sharesfor Homogeneous Good (SCH)

C(x) = C*(XN) for all x} E {qF(N, C, ) is serial cost sharing} (3.31)
for all N, C.
(iii) The Shapley-Shubik method is the only method in C (DUM,ADD) meeting Lower
Bound (3.15).
Note that statement (i) still holds if we weaken UB by restricting attention to
homogeneous cost functions [i.e., we only require the inequality when C takes the
form C*(XN)]. Statements (i) and (iii) have direct counterparts in the dr model (see,
respectively, the discussion after the corollary to Theorem 3.2 and Corollary 2 to
Theorem 3.1). Statement (ii) can be adapted to that model as well, by adding Equal
Treatment of Equals to the requirements.

3.7. Non-additive cost sharing methods

3.7.1. The Stand-Alone core approach

In the theory of classical cooperative games (namely the model in Section 3.2) surplus-
sharing methods that are not additive in the production (or cost) function have been
proposed for three decades [Schmeidler (1969)]. For simplicity we fix the population N
throughout this section.
Consider the binary problem (N, C, N), denoted (N, C) for simplicity, and suppose
that the cost function C is subadditive, namely:

For allS, TCN: SnT=0 X C(SUT) <C(S)+C(T).

This property is plausible in every problem where "serving coalition S" is orthogonal
to "serving coalition T": the costs incurred by T ( S) are the same whether or not
S ( T) is served.
350 i. Moulin

Under subadditivity of costs, a natural equity requirement on the solution y to the

problem (N, C) is the Stand-Alone core property:

Eyi < C(S) for all S C N (3.32)

iCS

If this property fails, the coalition of agents S has an objection to y: if S were standing
alone it would benefit from a reduced cost share, whereas the cost share of N\S would
increase.
In view of budget balance, YN = C(N), the above system of inequalities is equivalent
to

y, > C(N) - C(N\S).

iCS

This is the No-Subsidy principle: S should pay at least the incremental cost of service
(if this inequality fails, coalition N\S has a Stand-Alone objection).
The following fact is well known: subadditivity of the cost function C is not
sufficient to guarantee the existence of a solution y in the Stand-Alone core [i.e.,
meeting (3.32)]. The characterization of those cost functions generating a non-empty
core is one central theme of the theory of cooperative games, and is well understood:
see Owen (1982) or Moulin (1988, 1995b) for textbook presentations.
The Stand-Alone core property for a cost-sharing method Cqtakes the following
form (for a given population N and demand N):

For all subadditive cost functions C,

the solution y = qp(N, C) is in the Stand-Alone core (3.32) (3.33)
whenever the latter is non-empty.

An interesting result connects the Stand-Alone core to the incremental cost shares
q;O(N, C) (3.2): the Stand-Alone core is contained in the convex hull CO[p(N, C)]
of all incremental cost shares when a varies over all permutations of N [Weber
(1988)]. In other words, any allocation y in the Stand-Alone core is achieved by at
least one random-order value. However, the choice of a particular random-order value
(the choice of the convex coefficients over the methods (pa) depends on the particular
cost function C. If we pick the same random value cp for all cost functions C, it must be
the case that for some choices of the subadditive cost function C, the Stand-Alone core
is non-empty and does not contain the solution qi'(N, C) [this is an easy consequence
of Theorem 1 in Young (1985a)]. In view of Theorem 3.1, there is no additive cost-
sharing method meeting the Stand-Alone core property (3.33).
Thus if we are committed to the Stand-Alone core property, a non-additive method
must be used. The abundant literature on classical cooperative games has proposed a
number of such methods; the most prominent among these is the nucleolus [Schmeidler
(1969)], selecting a central point in the Stand-Alone core by equalizing as much as
Ch. 6: Axiomatic Cost and Surplus Sharing 351

possible the cost savings C(S) - ys across all coalitions of N. See Owen (1982) or
Moulin (1988) for a textbook presentation.
The Stand Alone core approach is easily adapted to the model with variable
demands. Given a profile of demands x, the Stand-Alone cost of coalition S is
C(x[s], 0); inequality (3.32) becomes Ys < C(x[s], 0), and the subadditivity of the
cost function C writes C(x + x') < C(x) + C(x').
An important exception to the incompatibility of the Stand-Alone core and
Additivity is when the cost function C is submodular, a considerably more demanding
requirement than subadditivity:

Submodularity:
Binary model: C(S) + C(T) > C(S n T) + C(S U T) for all S, T C N.
Variable-demands model: C(x) + C(y) > C(x A y) + C(x V y) for all x,y,

(where (x A Y)i = min{xi,yi} and (x V y)i = max{xi,yi}). If C is submodular, the

inclusion of the Stand-Alone core in the range of the random-order values becomes
an equality [Ichiishi (1981)]:

SA Core(C) = CO[qp(N, C)]; SA Core(C,x) = CO[q(p(N, C,x)]

(in the binary and variable-demands models, respectively). In particular, any random-
order value meets the Stand-Alone core property, and the latter property is thus always
true for additive cost-sharing methods in the binary model. The same holds true in the
variable-demands model. The Stand-Alone requirement comes for free in the world of
additive methods. An instance of this general fact is statement (ii) in the Corollary to
Theorem 2.2: with a homogeneous good and C(x) = C*(XN), submodularity of C is
equivalent to the concavity of C*.

3.7.2. Extending homogeneous-good methods

One of the natural requirements in the heterogeneous-goods model is that the cost-
sharing method coincides with a given method whenever the goods are actually
homogeneous. In other words, we wish to impose the solutions p(N, C, x) whenever
C takes the form C(x) = C*(xN).
The ACH axiom (3.29) and the SCH axiom (3.31) are the key to characterize
respectively the Aumann-Shapley and serial methods (Corollaries 2 and 3 to
Theorem 3.4). Note that in the case of the AS method, we do not know of any
characterization result that would dispense with ACH.
Among additive methods, the two properties ACH and SCH also lead to severe
impossibility results. Within the set C (DUM,ADD) of additive methods, each one of
the following pairs of requirements are incompatible:
(i) ACH and Demand Monotonicity;
(ii) ACH and Ranking;
352 H. Moulin

(iii) SCH and Unit Invariance;

(iv) ACH and Serial for excludable public good, namely
{C(x) = C*(maxixi) for all x} = {y given by (3.23)}
(v) ACH and Ordinality (Remark 3.3);
(vi) SCH and Ordinality.
The first two incompatibilities are statement (ii) in Corollary 2 to Theorem 3.4. The
next two are also easy to derive, from Theorem 3.4, and the last two follow Sprumont's
result in Remark 3.3.
I regard each one of these impossibility results as a strong argument against the
Additivity requirement. Each pair of axioms is normatively meaningful, whereas
Additivity is only a structural-invariance property.
Sprumont (1998) proposes a handful of non-additive methods for which the axioms
listed above are compatible. For instance, in the two-agent case he constructs a method
satisfying ACH, DM and Ordinality as follows. We are given a problem (N, C, x), and
assume that all partial derivatives of C are strictly positive and bounded away from
zero. We say that two problems (C, x) and (C, x) are ordinally equivalent if there are
rescaling functions .i, one for each i E N, such that:

C(z) = C(A 1(zl), .. , n,(z,)) for all z; x = (l(5l), . , n(n)).

Given the problem (C,x) one shows that there exists a unique problem (C, ) such
that

(C,x)and (C, ) are ordinally equivalent, and

,iC(t ) = 1 for i = 1,2 and all t > 0

Then the ordinallyproportionalrule is defined as

(C, x) = X.
x1 +X2

This definition has been generalized to an arbitrary number of agents: Wang and
Zhu (2000) and Wang (2001). It is also limited by restrictive regularity conditions.
Nonetheless the combined properties of ACH, DM and Ordinality are a remarkable
achievement.
At least two ordinal extensions of serial cost sharing have been proposed. We
describe one of them, called the Moulin-Shenker rule by Sprumont (1998) who was the
first to analyze it formally. See also Koster (1998a,b) and Koster, Tijs and Borm (1998).
We are given a problem (C,x) where we assume, again, that all partial derivatives are
positive and bounded away from zero. The number of agents is arbitrary. The following
ordinary differential equation has a unique solution t -* r(t,x), namely a monotonic
Ch. 6: Axiomatic Cost and Surplus Sharing 353

rationing path from 0 to x. This path depends on the cost function C itself, but for
simplicity we omit this from the notation:

For all i, all t, 0 < t < xN :

dri Ir
a(t,x).OiC(r(t,x))
if r(t,x)< xi,K (.x )
dt ,x) 0 if ri(t,x) xi, (3.34
where a(t,x) = #{i G N I ri(t,x) < xi}

By definition of the counting operation a(t, x) we have

- 9i1 C(r(t,x)) . d(t,x) = 1 = rN(t,x)= t,

hence t --+r(t,x) is indeed a rationing path. Then we define the Moulin-Shenker cost-
sharing method:

J
XN

(C,X) = eiC(r(C, t, x)) dr(C, t, x)) for all N, C,x, i.

The intuition behind this method is simple: as long as his demand is not met, the
"active" agent i is served at a speed that equalizes the marginal cost of service among
all active agents:

dri drj
tiC(r(t,x)). dt(t,x) = d C(r(t,x)) d-(t,x) for all active i,j.

Sprumont (1998) proposes another ordinal extension of serial cost sharing in the vein
of the ordinal extension of average cost sharing described above. He gives axiomatic
characterizations for both extensions. Koster (1998a) offers a related characterization
of the Moulin-Shenker rule.

3.7.3. More non-additive methods and an open question

The study of non-additive cost-sharing methods has just begun and it shows great
potential. In the homogeneous-good model, Hougaard and Thorlund-Petersen (2001)
propose a non-additive method mixing the increasing and decreasing versions of
serial cost sharing: see Subsection 2.5.2. In the same homogeneous-good model, Tijs
and Koster (1998) propose a very natural non-additive generalization of incremental
cost sharing. Fix an ordering of N, say a = 1,2, . .. , n. Denote by L the Lebesgue
354 h. Moulin

measure of a measurable set in R+. The method in question charges to agent 1 the
cheapest marginal costs in [O,xN]:

yl =min{/C'(t)dt A1 C [O,xN] andL(A)=xl} (3.35)

A1

f JC'(t)dt A 2 C [,xN LA2) =x2 and A2 nA 0

Y2 = min f
where A* is a solution of (3.35).

Finally, a largely unexplored model is the cost-sharing problem where several

outputs are jointly produced and each agent demands some amount of every good.
Kolpin (1996) extends to that context the incompatibility of Additivity, SCH and
Unit Invariance; see also Tjedo and Truchon (1999). McLean and Sharkey (1996,
1998) adapt the Aumann-Shapley method to that context and extend the classical
characterization result (Corollary 2 to Theorem 3.4).

References

Aumann, R.J., and M. Maschler (1985), "Game theoretic analysis of a bankruptcy problem from the
Talmud", Journal of Economic Theory 36:195-213.
Aumann, R.J., and L.S. Shapley (1974), Values of Non Atomic Games (Princeton University Press,
Princeton, NJ).
Banker, R. (1981), "Equity considerations in traditional full cost allocation practices: an axiomatic
perspective", in: S. Moriarty, ed., Joint Cost Allocations (University of Oklahoma Press, Oklahoma
City) pp. 110-130.
Barbera, S., M.O. Jackson and A. Neme (1997), "Strategy-proof allotment rules", Games and Economic
Behavior 18:1-21.
Benassy, J.-P. (1982), The Economics of Market Disequilibrium (Academic Press, New York).
Billera, L., and D. Heath (1982), "Allocation of shared costs: a set of axioms yielding a unique procedure",
Mathematics of Operations Research 7:32-39.
Billera, L., D. Heath and J. Raanan (1978), "Internal telephone billing rates: a novel application of non
atomic game theory", Operations Research 26:956-965.
Bryant, J. (1983), "A simple rational expectations Keynes-type model", Quarterly Journal of Economics
98:525-528.
Cachon, G., and M. Lariviere (1996), "Capacity choice and allocation: Strategic behavior and supply
chain performance", Mimeo (Duke University, Fuqua School of Business, Durham, NC).
Calvo, E., and J.C. Santos (2000), "A value for multichoice games", Mathematical Social Sciences
40(3):341-354.
Ching, S. (1994), "An alternative characterization of the uniform rule", Social Choice and Welfare
11(2):131-136.
Chun, Y. (1988), "The proportional solution for rights problems", Mathematical Social Sciences 15:
231-246.
Chun, Y. (1999), "Equivalence of axioms for bankruptcy problems", International Journal of Game
Theory 28:511-520.
Ch. 6: Axiomatic Cost and Surplus Sharing 355

Chun, Y., J. Schummer and W. Thomson (1998), "Constrained egalitarianism: a new solution for claims
problems", Mimeo (University of Rochester, NY).
Dagan, N. (1996), "New characterizations of old bankruptcy rules", Social Choice and Welfare 13:
51-59.
Dagan, N., and O. Volij (1997), "Bilateral comparisons and consistent fair division rules in the context
of bankruptcy problems", International Journal of Game Theory 26:11-25.
Davis, M., and M. Maschler (1965), "The kernel of a cooperative game", Naval Research Logistic
Quarterly 12:223-259.
de Frutos, M.A. (1998), "Decreasing serial cost sharing under economies of scale", Journal of Economic
Theory 79:245-275.
de Frutos, M.A. (1999), "Coalitional manipulation in a bankruptcy problem", Review of Economic
Design 4:255-272.
de Nouweland, A., and S. Tijs (1995), "Cores and related solution concepts for multi-choice games",
Mathematical Methods of Operations Research 41:289-311.
Demers, A., S. Keshav and S. Shenker (1990), "Analysis and simulation of a fair queuing algorithm",
Internetworking: Research and Experience 1:3-26.
Deutsch, M. (1985), Distributive Justice: A Social-Psychological Perspective (Yale University Press,
New Haven, CT).
Drhze, J. (1975), "Existence of an equilibrium under price rigidity and quantity rationing", International
Economic Review 16:301 320.
Elster, J. (1992), Local Justice (Russell Sage Foundation, New York).
Friedman, E.J. (1999), "Paths and consistency in additive cost sharing", Mimeo (Rutgers University).
Friedman, E.J., and H. Moulin (1999), "Three methods to share joint costs and surplus", Journal of
Economic Theory 87(2):275-312.
Haimanko, 0. (1998), "Partially symmetric values", Mimeo (Hebrew University, Jerusalem).
Hart, S., and A. Mas-Colell (1989), "Potential, value and consistency", Econometrica 57:589-614.
Herrero, C., and A. Villar (2001), "The three musketeers: four classical solutions to bankruptcy problems",
Mathematical Social Sciences 42(3):307-328.
Hougaard, J.L., and L. Thorlund-Petersen (2001), "Mixed serial cost sharing", Mathematical Social
Sciences 41(1):1-18.
Ichiishi, T. (1981), "Supermodularity: applications to convex games and to the greedy algorithm for
LP", Journal of Economic Theory 25:283-286.
Kaminski, M. (2000), "Hydraulic rationing", Mathematical Social Sciences 40(2):131-156.
Khmelnitskaya, A. (1999), "Marginalist and efficient values for TU games", Mathematical Social Sciences
38(1):45-54.
Kolpin, V (1996), "Multi-product serial cost sharing: an incompatibility with the additivity axiom",
Journal of Economic Theory 69(1):227-233.
Kolpin, V (1998), "Equitable nonlinear price regulation: an alternative approach to serial cost sharing",
Games and Economic Behavior 22(1):61-83.
Koster, M. (1998a), "Serial cost sharing for multi-commodity situations: a characterization of the
Moulin-Shenker rule", Mimeo (Tilburg University, Netherlands).
Koster, M. (1998b), "Multi-service serial cost sharing: an incompatibility with smoothness", Mimeo
(Tilburg University, Netherlands).
Koster, M., S. Tijs and P. Borm (1998), "Serial cost sharing methods for multi-commodity situations",
Mathematical Social Sciences 36:229-242.
Littlechild, S., and G. Owen (1973), "A simple expression for the Shapley value in a special case",
Management Science 20:370-372.
Loehman, E., and A. Whinston (1974), "An axiomatic approach to cost allocation for public investment",
Public Finance Quarterly 2:236-251.
Maschler, M. (1990), "Consistency", in: T Ichiishi, A. Neyman and Y. Tauman, eds., Game Theory and
Applications (Academic Press, New York) pp. 187-215.
356 I. Mollin

McLean, R., and W Sharkey (1996), "Probabilistic value pricing", Mathematics of Operations Research
43:73-95.
McLean, R., and W Sharkey (1998), "Weighted Aumann-Shapley pricing", International Journal of
Game Theory 27(4):511-524.
Mill, J.S. (1859), On Liberty (Parker, London). Reprinted, 1977, in: J.M. Robson, ed., The Collected
Works of John Stuart Mill, Vol. XVIII (The University of Toronto Press, Toronto).
Mirman, L., and Y. Tauman (1982), "Demand compatible equitable cost sharing prices", Mathematics
of Operations Research 7:40-56.
Moulin, H. (1987), "Equal or proportional division of a surplus and other methods", International Journal
of Game Theory 16(3):161-186.
Moulin, H. (1988), Axioms of Cooperative Decision Making (Cambridge University Press, Cambridge).
Moulin, H. (1992), "Welfare bounds in the cooperative production problem", Games and Economic
Behavior 4(3):373-401.
Moulin, H. (1995a), "On additive methods to share joint costs", Japanese Economic Review 46(4):
303-332.
Moulin, H. (1995b), Cooperative Microeconomics: A Game Theoretic Introduction (Princeton University
Press, Princeton, NJ).
Moulin, H. (1999), "Rationing a commodity along fixed paths", Journal of Economic Theory 84:41-72.
Moulin, H. (2000), "Priority rules and other asymmetric rationing methods", Econometrica 68(3):
643-684.
Moulin, H. (2002), "The proportional random allocation of indivisible units", Social Choice and Welfare,
forthcoming.
Moulin, H., and S. Shenker (1992), "Serial cost sharing", Econometrica, 50(5):1009-1039.
Moulin, H., and S. Shenker (1994), "Average cost pricing versus serial cost sharing: an axiomatic
comparison", Journal of Economic Theory 64(1):178-201.
Moulin, H., and S. Shenker (1999), "Distributive and additive cost sharing of an homogeneous good",
Games and Economic Behavior 27:299-330.
Moulin, H., and R. Stong (2002), "Fair queuing and other probabilistic allocation methods", Mathematics
of Operations Research, forthcoming.
Naumova, N.I. (2002), "Nonsymmetric equal sacrifice solutions for claim problems", Mathematical
Social Sciences 43(1):1-18.
O'Neill, B. (1982), "A problem of rights arbitration from the Talmud", Mathematical Social Sciences
2:345-371.
Owen, G. (1982), Game Theory, 2nd Edition (Academic Press, New York).
Rabinovich, N. (1973), Probability and Statistical Inference in Medieval Jewish Literature (University
of Toronto Press, Toronto).
Schmeidler, D. (1969), "The nucleolus of a characteristic function game", SiAM Journal on Applied
Mathematics 17:1163-1170.
Schummer, J., and W. Thomson (1997), "Two derivations of the uniform rule and an application to
bankruptcy", Economics Letters 55:333-337.
Shapley, L.S. (1953), "A value for n-person games", in: H.W. Kuhn and A.W. Tucker, eds., Contributions
to the Theory of Games II, Annals of Mathematics Studies 28:307-317.
Shenker, S. (1995), "Making greed work in networks: a game-theoretic analysis of gateway service
disciplines", IEEE/ACM Transactions on Networking 3(6):819-831 (circulated 1989).
Shubik, M. (1962), "Incentives, decentralized control, the assignment of joint costs and internal pricing",
Management Science 8(3):325-343.
Sobolev, A.I. (1975), "Characterization of the principle of optimality for cooperative games through
functional equations", in: N.N. Voroby'ev, ed., Mathematical Methods in the Social Sciences (Vipusk
6, Vilnius, USSR) pp. 92-151 (in Russian).
Sprumont, Y. (1991), "The division problem with single-peaked preferences: a characterization of the
uniform allocation rule", Econometrica 59(2):509-519.
Ch. 6: Axiomatic Cost and Surplus Sharing 357

Sprumont, Y. (1998), "Ordinal cost sharing", Journal of Economic Theory 81:126-162.

Sprumont, Y. (2000), "Coherent cost-sharing", Games and Economic Behavior 33:126-144.
Sprumont, Y., and Y.T. Wang (1998), "A characterization of the Aumann-Shapley method in the discrete
cost sharing model", Mimeo (Universit6 de Montreal).
Straffin Jr, P.D., and J.P Heaney (1981), "Game theory and the Tennessee Valley Authority", International
Journal of Game Theory 10(1):35-43.
Suh, S.C. (1997), "Two serial mechanisms in a surplus sharing problem", Mimeo (University of Windsor,
Ont., Canada).
Tauman, Y. (1988), "The Aumann-Shapley prices: a survey", in: A. Roth, ed., The Shapley Value
(Cambridge University Press, Cambridge) pp. 279-304.
Tej6do, C., and M. Truchon (1999), "Multi-product serial cost sharing: an incompatibility result", Mimeo
(Universit6 de Laval, Quebec).
Thomas, A. (1977), A Behavioral Analysis of Joint Cost Allocation and Transfer Pricing (Arthur
Anderson and Company Lectures series, Stipes Publishing Company).
Thomson, W. (1990), "The consistency principle", in: T Ichiishi, A. Neyman and Y. Tauman, eds., Game
Theory and Applications (Academic Press, New York) pp. 187-215.
Thomson, W (1994a), "Consistent solutions to the problem of fair division when preferences are
single-peaked", Journal of Economic Theory 63:219-245.
Thomson, W (1994b), "Resource monotonic solutions to the problem of fair division when preferences
are single-peaked", Social Choice and Welfare 11:205-223.
Thomson, W (1995), "Population monotonic solutions to the problem of fair division when preferences
are single-peaked", Journal of Economic Theory 5:229-246.
Thomson, W. (1997), "The replacement principle in economies with single peaked preferences", Journal
of Economic Theory 76(1): 145-168.
Tijs, S., and M. Koster (1998), "General aggregation of demand and cost sharing methods", Annals of
Operations Research 84:137-164.
Wang, Y.T. (1999), "The additivity and dummy axioms in the discrete cost sharing model", Economic
Letters 64:187-192.
Wang, Y.T. (2000), "Simple random order methods to share costs", Mimeo (Sabanci University, Istanbul).
Wang, YT. (2001), "Proportionally adjusted marginal pricing to share costs", Mimeo (Sabanci University,
Istanbul).
Wang, YT., and D. Zhu (2000), "Ordinal proportional cost sharing rule", Mimeo (Sabanci University,
Istanbul).
Weber, R. (1988), "Probabilistic values for games", in: A. Roth, ed., The Shapley Value (Cambridge
University Press, Cambridge) pp. 101-119.
Winslow, G.R. (1992), Triage and Justice (University of California Press, Berkeley).
Young, H.P. (1985a), "Monotonicity in cooperative games", International Journal of Game Theory
13:65-72.
Young, H.P. (1985b), "Producer incentives in cost allocation", Econometrica 53:757-765.
Young, H.P (1987), "On dividing an amount according to individual claims or liabilities", Mathematics
of Operations Research 12:398-414.
Young, H.P (1988), "Distributive justice in taxation", Journal of Economic Theory 48:321-335.
Young, H.P (1990), "Progressive taxation and equal sacrifice", American Economic Review 80(1):
253-266.
Chapter 7

POSITIONAL RULES OF COLLECTIVE DECISION-MAKING*

PRASANTA K. PATTANAIK
Department of Economics, University of CaliforniaRiverside

Contents
Abstract 362
Keywords 362
1. Introduction 363
2. The basic notation and definitions 364
3. Some notions of positionalist social ranking rules and social decision rules 367
3.1. The definition of positionality 367
3.2. Some positionalist rules 371
4. Arrow's conditions, the principle of simple majority, and positionalist SDRs 375
4.1. Arrow's conditions and positionalist rules 375
4.2. The simple majority rule and positionalist SDRs 378
5. The structure of the Borda rule 380
5.1. Axiomatic characterizations of the Borda rule 380
5.2. A Bayesian justification for the BDR (I) 384
5.3. A Paretian argument for the BDR (I) 385
5.4. A general comparison of the Borda procedure and other positionalist voting rules 386
6. Score-based rules, dominance-based scoring rules, and runoff procedures 386
6.1. Score-based ranking and decision rules 386
6.2. Score-based ranking rules and the principle of dominance 388
6.3. Runoff systems 390
7. Concluding remarks 392
References 393

* My greatest debt is to Amartya Sen and Kotaro Suzumura for numerous incisive suggestions. For
helpful discussions, I am grateful to Nick Baigent, Rajat Deb, Wulf Gaertner, Santosh Panda, and
Maurice Salles. I am also grateful for the generous hospitality of the School of Economics, Deakin
University, Burwood, Australia, where I did part of the work on this chapter.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K Sen and K. Suzumura
© 2002 Elsevier Science B. V All rights reserved
362 PK. Pattanaik

Abstract

In this chapter, we seek to review some of the central concepts and results in the
literature on positionalist voting rules, which goes back to Borda (1781). After the
introduction of the basic notation and definitions in earlier parts of the chapter, we
explore in Section 3 the distinction between positionalist and non-positionalist social
ranking rules and social decision rules. We define alternative notions of positionality
and give several examples. Section 4 discusses the positionalist decision procedures in
the context of the conditions figuring in Arrow's celebrated impossibility theorem and
also in the context of the simple majority rule. In Section 5, we review the literature on
the structure of the Borda ranking rule and the Borda decision rule. Some well-known
axiomatic characterizations of the Borda ranking rule and the Borda decision rule
are discussed here. In Section 6, we discuss some results on the structure of score-
based ranking rules and score-based decision rules, which include, as special cases, the
Borda ranking rule and the Borda decision rule, respectively. This section also contains
a discussion of dominance-based scoring rules and social decision rules known as
runoff systems. Section 7 concludes the chapter.

Keywords

social ranking rules, social decision rules, positionalist social ranking rules,
positionalist social decision rules, Borda ranking rule, Borda decision rule, simple
majority rule, score-based ranking rules, score-based decision rules, runoff systems

JEL classification: D71

Ch. 7: PositionalRules of Collective Decision-Making 363

1. Introduction

Consider a committee of seven members with the task of choosing one of three options,
a, b, and c. The first 4 voters strictly prefer a to b, b to c, and a to c; the other three
voters strictly prefer b to c, c to a, and b to a. Which option should the committee
choose? One possible answer is that the committee should choose a since a majority of
individuals prefer a to b, and also a majority of individuals prefer a to c. This answer
relies on two underlying principles:
(i) The principle of comparing every pair of alternatives, say, a and b, on the basis of
the voters' preferences over a and b only, without referring to the positions (first,
second or third) of a or b in the orderings of the voters over {a, b, c}.
(ii) The principle of determining the committee's choice from {a, b, c} on the basis of
the results of all the pairwise comparisons of alternatives belonging to {a, b, c}.
The prescription that the majority winner a should be chosen by the committee
seems to combine (i) and (ii) with the use of the simple majority rule for making the
pairwise comparisons referred to in (i). However, one can question the prescription
of the choice of a, arrived at in this fashion. Is a really the 'obvious' choice for the
committee? After all, though a occupies the first position in four individuals' orderings
over {a, b, c}, a also occupies the very last position in as many as three individuals'
orderings. Considering this and considering that b occupies the first position in three
individuals' orderings and no lower than the second position in any voter's ordering,
one could argue that b is a better choice for the committee than a. Note that this
intuitive argument for the choice of b is based crucially on the positions that each
alternative occupies in the seven individuals' overall orderings over {a, b, c}. Thus,
we have two distinct ways of approaching the problem of the committee's choice. One
approach goes by (i) and (ii) and uses the simple majority rule to determine the results
of the pairwise comparisons required by (i), thus ignoring much information regarding
the alternatives' positions in the overall orderings of the members of the committee.
The other approach takes into account the position of each alternative in the overall
ordering of each member.
Each of the two approaches outlined above seems to have much intuitive plausibility.
It is not, therefore, surprising that debates regarding 'non-positionalist' and 'position-
alist' voting rules and the relative strengths and weaknesses of specific rules belonging
to these two categories are at least as old as the formal theory of voting itself. Borda
(1781) and Condorcet (1785) are two of the founders of the formal theory of voting
and elections, and it was Borda who introduced what seems to be the best known
positionalist voting procedure, namely, the Borda rule, while Condorcet, a strong
critic of the Borda rule, was a champion of what is perhaps the most analysed non-
positionalist voting principle, namely, the principle that, if a candidate defeats every
other candidate on the basis of the simple majority rule, then that candidate should
be the winner in the election. The discussion of these principles has often originated
in more than just abstract intellectual interest: ever since these voting rules were first
formulated, they have been put to practical use. Indeed, following Borda's presentation
364 PK. Pattanaik

of a paper on voting rules in a meeting of the Academy of Sciences of France in 1784,

the Academy decided to use Borda's rule for the election of its members, and continued
to use it for the next 16 years or so .
The two distinct strands of thought that had their origins in the works of Borda
and Condorcet have continued to be explored in the recent theory of social choice.
In particular, the positionalist voting rules have been often seen as an attractive way
of escaping the negative impact of the celebrated impossibility theorem of Arrow
(1951, 1963), which was proved in a non-positionalist framework 2. The purpose of this
chapter is to review some of the basic conceptual developments and results relating
to positionalist voting rules. The literature on positionalist voting rules is vast, and
I shall not try to give an exhaustive account of it. Instead, I shall focus on some central
concepts and results. The plan of the chapter is as follows. In Section 2, I introduce
some basic notation and definitions. In Section 3, I introduce the distinction between
positionalist and non-positionalist social ranking rules and social decision rules and
give several examples. In Section 4, I consider the positionalist decision procedures in
the context of the conditions figuring in Arrow's (1951, 1963) impossibility theorem
and also in the context of the simple majority principle. Section 5 reviews the literature
on the structure of the Borda ranking rule and the Borda decision rule. In Section 6,
I discuss some results on the structure of score-based ranking rules and score-based
decision rules, which, respectively, include as special cases the Borda ranking rule
and the Borda decision rule. This section also considers what I call the dominance-
based scoring rule and social decision rules known as runoff systems. I conclude in
Section 7.

2. The basic notation and definitions

Let N be the set of all nonnegative integers. Let N be the class of all non-empty and
finite subsets of N. The elements of N will be called societies and will be denoted by
S, 5', etc. The elements of a society will be called individuals in that society.
Let X be the universal set of alternatives. I assume that X is a finite set and #X > 3
(for every set Y, #Y denotes the number of elements of Y). Let X denote the set of all
non-empty subsets of X. The elements of X will be denoted by A, A', etc., and will
be called issues. An issue is to be interpreted as a feasible set of alternatives.

l For a historical account of the contributions of Borda and Condorcet, see the classic book by Black
(1958), pp. 156-185.
2 See, for example, Fine and Fine's (1974b, pp. 474-475) interpretative comments on their results on
positionalist voting rules: "The novelty here lies in the rejection of the Arrowian heritage of social
choice theory. This heritage, founded on the impossibility theorem, includes a premature pessimism
about finding and justifying reasonable mechanisms ... ". See also Goodman and Markowitz (1952)
who, shortly after Arrow's (1951) seminal contribution, used the framework of a positionalist rule to
derive a social welfare function based on a specific conception of cardinal and interpersonally comparable
individual utilities.
Ch. 7: PositionalRules of Collective Decision-Making 365

For every A EX, V(A), W(A), R(A) and L(A) denote, respectively, the set of all
reflexive binary relations over A, the set of all reflexive and connected binary relations
over A, the set of all orderings over A, and the set of all linear orderings over A 3. For
all A l Xal ll R(A) E V(A), and all x,y E A, (xP(A)y iff [xR(A)y and not yR(A)x]) and
(xI(A)y iff [xR(A)y and yR(A)xj). Every element R(A) of V(A) will be interpreted as
a weak preference relation over A. Accordingly, P(A) and I(A) will be interpreted,
respectively, as the strict preference relation and the indifference relation corresponding
to R(A). When A =X, we write V, W, R, L, R, P and I, instead of writing V(A), W(A),
R(A), L(A), R(A), P(A) and I(A). For every A CX, every non-empty subset A' of A,
and every R(A) E V(A), R(A)IA' denotes the restriction of R(A) to A' 4 . A non-empty
subset V' of V is said to be balanced iff, whenever a binary relation R belongs to
V', V' contains every binary relation R' such that, for some permutation a on X,
[for all x,y E X, o(x)R'(y) iff xRy].
For every S E N and every non-empty subset W' of W, let WIS denote the #S-fold
Cartesian product of W'. The elements of Ws will be denoted by Rs, R, etc., and
will be called preference profiles for S. Each such element will be interpreted as a
conceivable #S-tuple of individual binary weak preference relations, with exactly one
binary weak preference relation for each individual in S. Depending on the assumptions
made about individual preferences either explicitly or implicitly in a model, some
conceivable preference profiles may not be admissible: for example, if we assume that
individual preferences are orderings over X, then a preference profile where some of
the individual preference relations are not transitive will not be admissible. The weak
preference relation of individual i (i C S) figuring in the #S-tuples Rs, Rs etc. will be
denoted, respectively, by Ri, Ri , etc. For every A EX and every preference profile Rs
for a society S, RsIA denotes (Ri/A)i s.
Note that, while N is the set of all conceivable societies, a model may assume the
society to be fixed so that variations in the society are not permitted in the model.
Thus, in many areas of the theory of social choice, the society is assumed to be
fixed [see, for example, the classic contributions of Arrow (1963) and Sen (1970)].
However, in the literature on positionalist voting rules, the variability of the society
or the electorate often plays an important role [see, for instance, Smith (1973) and
Young (1974)]. So, throughout this chapter, I shall use a framework that allows the
society to vary. Again, while some writers [e.g., Smith (1973) and Young (1974)]

3 Let T be a binary relation defined over an issue A. T is:

(i) reflexive over A iff, for all x E A, xTx;
(ii) connected over A iff, for all distinct x,y E A, xTy or yTx;
(iii) transitive over A iff, for all x,y,z C A, [xTy and yTz] implies xTz;
(iv) antisymmetric over A iff, for all distinct x,y E A, not [xTy and yTx];
(v) an ordering over A iff it is relexive, connected and transitive over A; and
(vi) a linear ordering over A iff it is an antisymmetric ordering over A.
4 Let T be a binary relation defined over an issue A and let A' be a given subset of A. Then the
restriction of T to A' is a binary relation T' over A' such that, for all x,y E A', [xT'y iff xTy].
366 PK. Pattanaik

work with a fixed set of alternatives, so that X is the only admissible issue, other
writers [e.g., Fishburn (1973a,b)] work with X as the class of admissible issues. In
this chapter, I assume that X is the class of all admissible issues. However, I do not
necessarily assume W or V to be the class of admissible individual weak preference
relations. Much of the time I shall assume L to be the set of admissible individual
weak preference relations, though in some contexts, I shall specify other subsets of W
as the set of admissible individual preferences. For every non-empty subset W' of W,
let D(W') denote the set of all (S, A, Rs) such that S e N, A CX and Rs E W 's.
As is the case with many other areas of the theory of social choice, the discussion
of positionalist social decision procedures has been conducted sometimes in terms
of social rankings of the alternatives and sometimes in terms of social choice from
a feasible set of alternatives. It will be convenient to have the basic concepts for
each of these two formulations, since the translation of the axioms and results in one
framework into axioms and results in the other framework is not always unique and
obvious.
Definition 2.1.
(2.1.1) A social ranking rule (SRR) is a function G which, for every (S,A,Rs)
D(W'), specifies exactly one R(A) V(A), where W' is a given non-empty,
balanced subset of W. We write R(A) = G(S,A,Rs). A social ranking rule G
will be called a social ordering rule (SOR) iff, for every (S,A,Rs) in its
domain, G(S,A, Rs) is an ordering over A.
(2.1.2) A social decision rule (SDR) is a function H which, for every (S, A, Rs) c
D(W'), specifies exactly one non-empty subset A of A, where W' is a given
non-empty, balanced subset of W. We write A = H(S,A, Rs) 5 .
G(S, A, Rs) in Definition 2.1.1 is to be interpreted as the social weak preference relation
over A, given the society S, the issue A and the preference profile Rs. H(S, A,Rs)
in Definition 2.1.2 is to be interpreted as the set of alternatives 'chosen' by the
society S, when A is the issue and Rs is the preference profile for S. W' figuring in the
domain D(W') of an SRR or SDR in a model reflects the assumptions made about what
type of individual preferences are admissible in the model. Unless specified otherwise,
it is to be understood that the SDR or SRR under consideration has D(L) for its domain.
When I explicitly mention that an SRR or SDR has D(W') as its domain, W' is to be
understood to be an arbitrarily given balanced, non-empty subset of W. There is one
point about the definition of an SRR which may deserve some clarification. As in the
standard Arrowian framework, the definition of an SRR here is formulated in terms
of a social weak preference relation, but, under our definition, given a society and a
preference profile, the SRR specifies, for each issue A, a social weak preference relation

5 1 have built into the definitions of an SRR and an SDR the requirement that W' is balanced. One
can omit this requirement, but, in that case, the definitions of some of the properties later will become
a little more involved.
Ch. 7: PositionalRules of Collective Decision-Making 367

over A: there is no stipulation that all these different weak preference relations over
the different issues are ultimately restrictions, to the respective issues, of one single
weak preference relation defined over the universal set X. This provides flexibility in
discussing different types of positionalist procedures 6.

3. Some notions of positionalist social ranking rules and social decision rules

3.1. The definition of positionality

It may be useful to start with definitions of positionalist and non-positionalist SRRs
and SDRs, especially since these notions are not often defined in the literature 7 . As
we shall see, it is not always obvious how to define these concepts, even though, in
some cases, we do have some strong intuition about the matter.
Consider first the notion of positionalist SRRs. It seems clear that one would like to
exclude from the scope of this notion all those SRRs under which, given an issue,
the social ranking of any two alternatives in the issue depends exclusively on the
individual rankings over those two alternatives only. Such SRRs hardly have any
feature that can justify the epithet 'positional'. Beyond this, one can think of various
alternative possibilities. First, one can think of a rather weak concept of positionality
where an SRR is considered positionalist if, for some society and for some issue, the
social ranking of some pair of alternatives in the issue does not depend exclusively
on the individual rankings over those two alternatives. Alternatively, one can think
of a stronger notion of positionality, where, for every society, every issue and every
preference profile for the society, the social ranking of any two alternatives depends
on only the 'positions' of the two alternatives in the individual orderings (the notion
of a 'position' needs further clarification), but does not depend exclusively on the
individual rankings over those two alternatives. Definition 3.1 below seeks to capture
some of these alternative possibilities.
For every linear ordering R* over X, every issue A, and every x E A, the rank or the
position of x in A in terms of R* is defined to be r(x, A, R*) _ #A - #{a C A : xP*a}.
For example, if X = a, b, c, d}, A = {a, b, d} and the linear ordering R* is such
that aP*cP*bP*d, then the rank of b in A in terms of R* is r(b,A,R*) = 2. For
every society S, every issue A, every Rs &Ls, and every x c A, let r(x, A, Rs) denote

6 For example, this flexibility is useful in distinguishing between the Borda ranking rule (I) and the
Borda ranking rule (II) introduced in Section 3.2.
7 Fine and Fine (1974b) p. 460 are an exception. They provide definitions of what they call 'weak
positionality', 'local positionality' and 'global positionality' for decision procedures that specify social
rankings of alternatives. See also Murakami's (1968, p. 61) distinction between social ranking procedures
based on 'pairwise comparisons' and social ranking procedures based on 'overall comparisons'. In my
definition of Type I strong positionality of an SRR (see Definition 3.1.4), I have basically used the notion
of 'weak positionality' of Fine and Fine (1974b), though with some modification so as to rule out SRRs
based on pairwise comparisons.
368 P3K. Pattanaik

(r(x,A, Ri))iE s. r(x,A, Rs) is then the vector of the ranks of x in A in terms of the
linear orderings of the individuals in S, given the preference profile Rs. When R*(A),
the restriction of R* to A, is not an ordering over A, or when R*(A) is an ordering
over A but not a linear ordering over A [so that R*(A) involves indifference between
distinct alternatives belonging to A], it is not obvious what would be an intuitively
compelling notion of the position or rank of an alternative in A in terms of R*. I shall
avoid this difficulty, at least in the context of the distinction between positionalist and
non-positionalist SRRs, by often considering only the case where the set of admissible
individual weak preference relations is L, the set of all linear orderings over X (see,
for instance, Definition 3.1.4).
Definition 3.1.
(3.1.1) Let G be an SRR with domain D(W'). Then G is based on pairwise
comparisons iff, for every society S, every issue A, and all Rs,R' C W's ,
[for all x, y e A, if Rs/{x,y} = R}/{x,y}, then R(A)/{x,y} = R'(A)/{x,y},
where R(A) = G(S, A, Rs) and R'(A) = G(S,A, R)].
(3.1.2) Let G be an SRR with domain D(W'). G satisfies independence of irrelevant
alternatives iff, for every society S, every issue A, and all Rs, R E W's,
[Rs/A = R/A] implies [G(S, A, Rs) = G(S, A, R})].
(3.1.3) Let G be an SRR with domain D(W'). G is positionalist in the Type I weak
sense (equivalently, weakly positionalist (I)) iff G satisfies independence of
irrelevant alternatives, but is not based on pairwise comparisons. G is posi-
tionalist in the Type II weak sense (equivalently, weakly positionalist (II))
iff G violates independence of irrelevant alternatives.
(3.1.4) An SRR with domain D(L) is positionalist in the Type I strong sense
(equivalently, strongly positionalist (I)) iff

it is not based on pairwise comparisons and for every society S,

every issue A, all Rs, R in Ls, and all x,y E A, such that (3.1)
r(x,A,Rs) = r(y,A,R;), R(A)/{x,y} = R'(A)/{x,y}.

An SRR G with domain D(L) is positionalist in the Type II strong sense

(equivalently, strongly positionalist (II)) iff it is not based on pairwise
comparisons and

for every society S, every issue A, all Rs,R' in Ls, and all x,y E A,
such that r(x,X,Rs) = r(y,X,R'), R(A)/{x,y} = R'(A)/{x,y}.
(3.2)
Intuitively, if an SRR G is based on pairwise comparisons, then, for every (S, A, Rs) in
its domain, the ranking of every pair of alternatives in A depends just on the individual
rankings over that pair of alternatives and does not depend on any other feature of the
individual orderings. If an SRR is based on pairwise comparisons, then clearly one
would not like to call it positionalist in any sense. Therefore, it is excluded in every
Ch. 7: PositionalRules of Collective Decision-Making 369

definition of positionality given above, including the definition of weak positional-

ity (II), where such exclusion does not figure explicitly in the definition but is implied
by it. Weak positionality (I) implies that, in ranking two alternatives, the SRR does not
make use of any information not contained in the restrictions of the individual weak
preference relations to the issue under consideration. At the same time, the definition
also implies that sometimes, in ranking a pair of alternatives belonging to an issue,
the SRR makes use of some information about individual preferences over the issue,
other than just the individual rankings over the pair of alternatives under consideration.
Weak positionality (II) tells us that sometimes, in determining the social ranking of
two alternatives in an issue, the social ranking rule makes use of information beyond
the individual weak preference relations over the alternatives in the issue; of course,
when this happens, the SRR cannot be based on pairwise comparisons. An SRR G
with domain D(L) is strongly positionalist (I) when: (i) the social ranking of any two
alternatives x andy in an issue A remains the same when the rank in A assigned by each
individual's (linear) ordering to x, as well as to y, remains the same, even though the
individual orderings may undergo changes otherwise; and (ii) G is not based on pair-
wise comparisons. The interpretation of strong positionality (II) of an SRR is similar
except that now we consider the ranks of x and y in X rather than in the given issue A.
To see the restrictions imposed by the different types of positionality considered
above, consider the following example:
Example 3.2. Let S = {1,2, 3},X = {al, a 2 ,... ,as}, and A = {al, a2 , a3, a4 }. Assume
that D(L) is the domain of the SRR. Let Rs, R, R' G D(L) be as follows:

Rs RI R"
R1 R R3 RI R; R3 R' R1 R]
a5 a5 as a5 a2 a4 a5 a5 a3
al a3 a4 a a5 a5 aI a3 a4
a2 al al a3 al al a2 al al
a3 a4 a3 a2 a4 a2 a3 a4 as
a4 a2 a2 a4 a3 a3 a4 a2 a2

Note that, for every individual, the position of al, as well as the position of a4,
in the issue A remains the same as between the two preference profiles Rs and R:
r(al,A,Rs) = r(al,A,R') = (1,2,2) and r(a4,A,Rs) = r(a4,A,R') = (4,3, 1). Given
this, if the SRR is strongly positionalist (I), then the social ranking of al and a4 must
not change as between R(A), which corresponds to Rs, and R'(A), which corresponds
to RS. However, if the SRR is either weakly positionalist (I) or weakly positionalist (II),
then R(A) and R'(A) may not coincide over {al, a4}, since some individuals' orderings
over A have changed. Also, if the SRR is strongly positionalist (II), the social ranking
of x and y may change as we go from R(A) to R'(A), since the position of a4 in X has
changed for some individuals: r(a4 ,X, Rs) = (5, 4, 2) and r(a4 ,X, R) = (5, 4, 1). Now
370 PK. Pattanaik

consider Rs and R". Since r(al,X, Rs)= r(al,X,R') and r(a4 ,X, Rs)= r(a4,X, R"),
R(A) and R"(A) must coincide over {al, a4} under a strongly positionalist (II) SRR.
However, since r(a4, A, Rs) = (4, 3, 1) and r(a4, A, R") = (4, 3, 2), under a strongly posi-
tionalist (I) SRR, the ranking of al and a 4 may change as we go from R(A) to R"(A). U
One can easily formulate for SDRs the counterparts of the concepts of weak posi-
tionality that we introduced for SRRs. However, the definition of strong positionality
for SDRs is more difficult, since, in the case of SDRs we are concerned with social
choice from an issue rather than the social ranking of two alternatives, given an issue.
In this case, when we seek to formulate the definition of strong positionality in terms
of positions of the alternatives, it is certainly plausible to say that, given an issue A,
if the position, in A, of every alternative belonging to A remains the same for every
individual, then social choice from A should remain the same. However, if the position
in A of every alternative belonging to A remains the same, then the restrictions of
individual preferences to the issue also remain the same. Thus, the proposed restriction
in terms of the positions of all the alternatives in the issue comes very close in spirit to
the notion of weak positionality (I) for SRRs. It is possible to think of more restrictive
conditions involving positions of the alternatives, which can be made the criteria for
strong positionality of SDRs. But I shall not discuss them here. Instead, I shall just
define the notions of weak positionality for SDRs.
Definition 3.3.
(3.3.1) Let H be an SDR with domain D(W'). We say that H is based on pairwise
comparisons iff:
(i) for all S C N, all a, b C X, and all Rs, Rs E W' s, if [Rs/{a, b} = R./{a, b}],
then [H(S, {a, b}, Rs) = H(S, {a, b}, RS)];
(ii) for all S N, all A N, and all Rs,R s E W 's , if [H(S,{x,y}, Rs)=
H(S, {x,y}, R) for all x,y E A], then [H(S,A,Rs) = H(S,A,R)] 8 .
(3.3.2) Let H be an SDR with domain D(W'). G satisfies independence of irrelevant
alternatives iff, for every society S, every issue A, and all Rs, Rs W',
[Rs/A = RI/A] implies [H(S, A, Rs) = H(S, A, R)].
(3.3.3) An SDR H with domain D(L) is positionalist in the Type I weak
sense (equivalently, weakly positionalist (I)) iff it satisfies independence of
irrelevant alternatives, but is not based on pairwise comparisons.
(3.3.4) An SDR H with domain D(L) is positionalist in the Type II weak
sense (equivalently, weakly positionalist (II)) iff it violates independence of
irrelevant alternatives.
If an SDF is based on pairwise comparisons, then the social choice from any two-
element issue depends exclusively on the individual rankings of the two alternatives
in the issue; and, for every issue A, the social choice from A does not change so

8 Cf. Mur-akami (1968) p. 61.

Ch. 7: PositionalRules of Collective Decision-Making 371

long as the social choices from each two-element subset of A remains the same. We
say that an SDR is weakly positionalist (I) if and only if: (i) the choice of a society
from every issue depends only on the restrictions of the individual preferences to that
issue; and (ii) the SDR is not based on pairwise comparisons. An SDR is weakly
positionalist (II) iff it violates independence of irrelevant alternatives. In defining
weak positionality (II) for an SDR, we do not have to specify that the SDR is not
based on pairwise comparisons, since an SDR that violates independence of irrelevant
alternatives cannot be based on pairwise comparisons.

3.2. Some positionalistrules

I now give several examples of positionalist SRRs and SDRs. In all these examples,
I assume that L is the set of permissible individual weak preference relations. Note
that, only in the first example involving the well-known Borda procedure, I give
the SRR version as well as the SDR version. In all other cases I confine myself to
the SDR versions; the corresponding SRR versions can be formulated in a routine
fashion.
Definition 3.4. The Borda ranking rule (Type I) [BRR (I)] with domain D(L) is an
SRR with domain D(L), such that

For every (S,A,Rs) E D(L) and all x,y E A,

xR(A)y iff E [#A - r(x,A,Ri)] > E [#A - r(y,A,Ri)]. (3.3)

iS its

The Borda decision rule (Type I) [BDR (I)] with domain D(L) is an SDR H with
domain D(L), such that

For every (S,A, Rs) D(L),

H(S,A,Rs) = {x E A: E [#A - r(x,A,Ri)] > E [#A - r(y,A,Ri)], for all y E A}.

iCS iCS
(3.4)
Since, for every a E A, [#A - r(a,A,Ri)] is the number of alternatives in issue A
which are below a in the ordering Ri of individual i, the BRR (I) specifies the social
ranking of the alternatives in A as follows. For every individual i, an alternative x
gets the score 0 if x is Ri-least in A; it gets the score 1 if it is the second-lowest
alternative in A in terms of Ri, and so on (recall that individual preferences are assumed
to be linear orderings). The alternatives in A are then socially ranked on the basis of
the total scores received by each alternative from all individuals in the society. The
BDR (I) follows a similar procedure, but, instead of ranking the alternatives in the
issue, it declares the alternatives with the highest scores in the issue to be socially
chosen. It is obvious that, given the issue A and a linear ordering Ri for individual i,
instead of assigning to each x in A the score [#A - r(x, A, Ri)], one could assign to each
372 PK. Pattanaik

x in A the score a + l[#A - r(x,A,Ri)], where il is positive and a and r7 are invariant
with respect to x and i, without making any difference to the results yielded by the
BRR (I) or BDR (I). That the BDR (I) satisfies independence of irrelevant alternatives
and the BRR (I) satisfies (3.1) is obvious. To see that the BDR (I) is not based on
pairwise comparisons, consider the following example.
Example 3.5. Let S = { 1,2, 3, 4, 5} and X = {x,y, z}. Let D(L) be the domain of the
SRR and let Rs, Rs E D(L) be such that

Rs Rs
R1 R2 R3 R4 R RI RI2 RI RI R'
x x y z x y x y z x
y y z x y x y z x y
z z x y z z z x y z

It can be checked that, given the BDR (I), the society's choices from every issue
with no more than two elements remain the same as between the two preference
profiles. However, the BDR (I) yields {x} as the set of chosen elements corresponding
to X when the preference profile is Rs, and {x,y} as the set of chosen elements
corresponding to X when the preference profile is R'. Thus, in addition to satisfying
independence of irrelevant alternatives, the BDR (I) is also not based on pairwise
comparisons. Therefore, the BDR (I) is a weakly positionalist (I) SDR. One can also
construct an example to show that, not only does the BRR (I) satisfy (3.1), but it is
also not based on pailwise comparisons, and, hence, is strongly positionalist (I). U
If one replaces r(x,A,Ri) and r(y,A,Ri), respectively, by r(x,X,R) and r(y,X, R i )
everywhere in Definition 3.3, one would have the definitions of another SRR and
another SDR, which I shall call the Borda ranking rule (II) (BRR (II)) and the Borda
decision rule (II) (BDR (II)).
So far, in considering the Borda ranking rule or the Borda decision rule of either
type, I have assumed that L is the set of permissible preferences. How would one define
the scores, when individuals have preference orderings which may not be linear? If
individual preference orderings are not linear, then the notion of the rank or the position
of an alternative in an individual's preference ordering becomes ambiguous. In that
case, there are several alternative ways in which one can specify the score received by
an alternative x from an individual i. In the context of the Type I Borda ranking rule or
Borda decision rule, given an issue A, one possibility is to say that the score received
by x from i is the number of indifference classes (defined in terms of i's ordering Ri)
in A which are below x. Another possibility is to say that the score received by x
from i is the number of alternatives in the issue which are less preferred than x by i.
Yet another possibility is to take a 'linearization' of the original preference ordering
Ri of i, calculate the average score that x and all the other alternatives indifferent to
Ch. 7: PositionalRules of Collective Decision-Making 373

x under Ri get for this linearized version of Ri, and assign this average score as the
score that x gets for Ri.
Definition 3.6. The plurality decision rule (PDR) with domain D(L) is an SDR H
with domain D(L) such that, for all (S, A, Rs) E D(L), H(S, A, Rs) is the set of all x E A
such that, for all y e A, the number of individuals who consider x to be best in A is
at least as large as the number of individuals who consider y to be best in A. 9
The PDR is a widely used method of voting. It is also a weakly positionalist (I) SDR.
That it satisfies independence of irrelevant alternatives is obvious. To see that it is not
based on pairwise comparisons, consider Rs and R' as specified in Example 3.5. It can
be checked that, under the PDR also, the socially chosen alternatives for every issue
with no more than two elements remain the same as between Rs and R', and, yet, the
set of socially chosen alternatives for X is {x} when the preference profile is Rs and
{x,y} when the preference profile is Rs.
The BDR (I) and the PDR are two members of a large class of SDRs which I shall
call score-based decision rules and which I define below.
Definition 3.7. A score-based ranking rule (SCRR) with domain D(L) is an SRR G
with domain D(L), for which the following holds:
For every positive integer m, there exists an m-tuple of numbers
(n[l,m],n[2,m],.. . ,n[m,m]) such that, for all (S,A,Rs) D(L), and all x,y C A,
xR(A)y iff E n[r(x,A,Ri),#A] > E n[r(y,A,Ri),#A].
CES iCS
(3.5)
A score-based decision rule (SCDR) with domain D(L) is an SDR H with
domain D(L), for which the following holds:
For every positive integer m, there exists an m-tuple of numbers
(n[l,m],n[2,m],... ,n[m,m]) such that, for all (S,A,Rs) D(L),

H(S,A,Rs) = {x E A: E n[r(x,A,Ri), #A] > n[r(y,A,Ri), #A], for all y c A}. 10

its icS
(3.6)

9 Under my definition of the plurality decision rule, one considers, for each alternative x in the issue
under consideration, the number of voters who consider x to be the best alternative in that issue. One
can think of an alternative definition under which one would consider, for each alternative x in the
issue under consideration, the number of voters who consider x to be the best in X. Thus, one could
have introduced a distinction between Type I and Type II plurality decision rules. I have not introduced
this distinction here; instead, I have introduced only what could have been called the Type I plurality
decision rule and have called it simply the plurality decision rule.
10 As in the case of the plurality decision rule (see note 7 above), in defining SCRRs and SCDRs I have
ignored the distinction between 'Type I' and 'Type II'. What I call SCRRs and SCDRs are 'Type I'
374 PK. Pattanaik

The m-tuple (n[I, m], n[2, m],...,n[m, m]) figuring in Equation (3.5) [resp. (3.6)]
will be called the scoring vector for mn-alternative issues under the score-based rank-
ing rule [resp. score-based decision rule]. The scoring vector, (n[l, m],... , n[m, m]), is
said to be monotonic (resp. strictly monotonic) iff n[l, m] > n[2, m] > ... > n[m, mi]
(resp. n[l,m] > n[2, m] >... > n[m, m]).
Under an SCRR, for every positive integer m, we have an m-vector, (n[l, m], n[2, m],
... , n[m, m]), where, intuitively, n[j, m] represents the score that an alternative x in any
issue A, #A = m, will receive, if a voter gives x the j-th rank in A. Given a society S,
an issue A, and a profile Rs for S, the SCRR ranks the alternatives in A on the basis
of the total score received by each alternative from all the individuals in the society,
an alternative with a higher total score being ranked higher than an alternative with a
lower total score. A score-based decision rule can be interpreted similarly. It is clear
that every SCRR is an SOR.
The BDR (1) is an SCRR such that, for every positive integer m, and every j,
m >j > 1, n[j, m] = m -j; and the PDR is an SCDR such that, for every positive
integer m, n[1,m] = 1, and, for everyj, m >j > 1, n[j,m] = 0.
My next example of a weakly positionalist (I) SDR is what I shall call the Black
decision rule. It represents a mixture of the majority-voting principle, which is based
on the principle of pairwise comparisons, and the BDR (I). This hybrid procedure due
to Black (1958) is motivated by the desire to overcome, with the help of the Borda
procedure, the difficulty that arises when there is no majority winner.
Given a society S, an issue A and a preference profile Rs E Ws, let M(S,A, Rs)
and M(S,A,Rs) denote, respectively, {x e A: #{i C S: xPiy} > #{i G S: yP,x}}
and {x E A: #i C S: xP iy} > #i E S: yPix}}. Clearly, M(S,A,Rs) contains at
most one alternative - the alternative in A, which strictly wins over every other
alternative in A when the simple majority principle is used to make the pairwise
comparisons. Similarly, M(S, A, Rs) is the set of all alternatives x in A, which weakly
win over every other alternative y in A in terms of the simple majority rule.
Definition 3.8. The Black decision rule with domain D(L) is an SDR H with
domain D(L), such that, for all (S, A, Rs) E L s , [if M(S, A, Rs) • 0, then H(S, A, Rs)
= M(S, A, Rs)] and [if M(S, A, Rs) = 0, then H(S, A, Rs) = B(S, A, Rs), where B is the
BDR (I)].
It can be checked that Black's decision rule is a weakly positionalist (I) SDR.
I conclude this section with an example of a weakly positionalist (I) SDR with
domain D(L), which has a very different structure from the ones that I have mentioned

procedures. However, one can easily formulate the Type II counterparts of these. The notions of an
SCRR and an SCDR are, respectively, similar to Smith's (1973) notion of a point system and Young's
(1975) notion of a simple scoring function; the only difference is that both Smith (1973) and Young
(1975) use formal frameworks where the issue is fixed. Fishburn's (1973b, pp. 217-218) concept of a
summation social choice functions is, in many ways, more general than the concept of a score-based
decision rule.
Ch. 7: PositionalRules of Collective Decision-Making 375

above, and which, as we will see later, is a member of a large class of weakly
positionalist (I) SDRs with a similar general structure. I call this SDR the runoff
system based on the BRR(I). Given (S,A,Rs) E D(L), the runoff system based on
the BRR (I) specifies the winners as follows. First, the BRR (I) is applied to (S, A, Rs).
If we have a social ordering R(A) such that all the alternatives in A are indifferent in
terms of R(A), then all of them are declared winners. However, if R(A) defines more
than one indifference class in A, then the alternatives in the lowest indifference class
are eliminated from A. Let the (non-empty) set of remaining alternatives be called A .
Then one applies BRR (I) to (S, A 1,Rs). Now, if the resultant social ordering R(A1)
has only one indifference class in Al, then one declares Al to be the set of winners.
Otherwise, one throws out from Al the lowest indifference class in Al, defined by the
BRR (I), and calls the reduced set A 2. One proceeds in this fashion until one reaches
a stage where there is only one indifference class in the reduced set corresponding to
that stage, defined by the ordering which the application of the BRR (I) yields at that
stage. An illustration may be helpful here.
Example 3.9. Let B denote the BRR (I); let S = {1,2,3,4}, A = {a, b, c, d}, and let
Rs be a profile of linear orderings of the individuals in S, such that [aPsbPlcPld;
bP2 dP 2cP2 a; aP3 bP3 cP3 d; cP4dP 4aP4 b]. Then R°(A)=B(S,A,Rs) defines three
indifference classes in A, with {d} the lowest indifference class (d secures a total
score of 4 as against 6 for c and 7 for both a and b). Eliminating d from A,
we have Al = {a,b,c}. Now consider R'(AI) = B(S,A 1,Rs). The lowest indifference
class defined for Al by R'(A1) is {c}. Eliminating c from Al, we have A 2 = {a,b}.
The lowest indifference class in A2 defined in terms of R2(A 2) = B(S,A 2, Rs) is {b}.
Eliminating b from A 2, we have A 3 = {a}. R3(A 3) = B(S,A 3,Rs) obviously defines
exactly one indifference class for A3 = {a}. Therefore, a emerges as the winner under
the runoff system based on the BRR (I). ·

4. Arrow's conditions, the principle of simple majority, and positionalist SDRs

4.1. Arrow s conditions and positionalistrules

It may be helpful to view positionalist rules in the light of the well-known conditions in
Arrow's (1951, 1963) impossibility theorem, especially, the requirement of collective
rationality, independence of irrelevant alternatives, and the Weak Pareto criterion.
As in Arrow (1951, 1963), consider a fixed society S, but to make the exposition
easier, assume that L is the set of permissible individual preferences . Then Arrow's
Condition 1' requires his social welfare function to specify a social ordering over X for

11 Arrow (1963), of course, assumed R to be the set of permissible individual preferences, but nothing
essential in his framework changes, if, instead of R, L is assumed to be the set of permissible individual
preferences.
376 PK. Pattanaik

every preference profile Rs in LS, while his Condition 3 (independence of irrelevant

alternatives) requires that, for all Rs, R' E Ls, and for every issue A, if Rs/A = R/A,
then the set of R-greatest elements in A must coincide with the set of R'-greatest
elements in A, where R and R' are the social orderings over X, yielded by the
social welfare function. Note that Arrow's independence of irrelevant alternatives is
equivalent to the following condition:

For all Rs, R E Ls, and for every issue A, if Rs/A = R/A, then R/A = R/A,
(4.1)
where R and R' are the social orderings over X, yielded by the social welfare function
for Rs and R', respectively.
First, compare the structure of an Arrowian social welfare function satisfying
Arrow's Conditions 1' and 3 (independence) with the structure of weakly position-
alist (I) social ordering rules, which are special cases of weakly positionalist (I)
social ranking rules. Given the fixed society S, a social ordering rule, say the
BRR (I), specifies exactly one ordering over A for every preference profile Rs.
Further, every weakly positionalist (I) SOR satisfies the condition of independence
of irrelevant alternatives (see Definition 3.1.2) that resembles (4.1): for every
issue A, and all Rs, R' L s , if Rs/A = R/A, then the ordering over A specified
by the weakly positionalist (I) SOR for Rs must be the same as the ordering
over A specified by it for R'. Also, note that, like a social ordering rule, given a
preference profile Rs, Arrow's social welfare function does implicitly specify, for
every issue A, an ordering over A, this implicitly specified ordering being simply the
restriction to A of the ordering over X specified by the social welfare function for Rs.
There is, however, a major difference between an Arrowian social welfare function
satisfying Arrow's condition of independence of irrelevant alternatives and a weakly
positionalist (I) SOR: given a preference profile Rs, the orderings over the different
issues which are implicitly specified by Arrow's social welfare function for those
issues are all embedded in one ordering over the universal set. But, under a weakly
positionalist (I) SOR, this would not be always true: for some society S and some
preference profile Rs, the orderings over the different proper subsets of X will not be
embedded in the ordering over X. For example, under the BRR (I), which is a weakly
positionalist (I) SOR, given a preference profile Rs, the different orderings specified
for the different proper subsets ofX are not typically embedded in the ordering over X.
Thus, given S, X and Rs, as specified in Example 3.5, the BRR (I) yields xP'({x,y})y
and xI'(X)y, and it is, therefore, impossible that R'({x,y}) can be embedded in R'(X).
The situation is different for weakly positionalist (II) SORs, such as the BRR (II). By
definition, weakly positionalist (II) SORs do not satisfy the property of independence
of irrelevant alternatives for SRRs (see Definition 3.3). For example, under the
BRR (II), even if the restriction to an issue A of every individual's ordering remains
the same as between Rs and R', the ordering over A specified by the BRR (II) can
change when Rs is replaced by R'. On the other hand, under the BRR (II), given a
profile Rs, the orderings specified for the different issues are all embedded in one
Ch. 7: PositionalRules of Collective Decision-Making 377

ordering over X [this, however, need not necessarily be true under every weakly
positionalist (II) SRR].
Now consider weakly positionalist (I) SDRs. By definition, they all satisfy a
condition similar in spirit to (4.1): if, for an issue A, Rs/A = R/A, then, under every
weakly positionalist (I) SDR, the set of alternatives socially chosen from A must
remain the same as between Rs and R'. However, for some preference profile, the
choices yielded by the weakly positionalist (I) SDR for the different issues will not be
'rationalizable' in terms of any ordering overX (or, indeed, in terms of any binary weak
preference relation over X). Consider again S, X and R', as specified in Example 3.5.
It is clear that the choice of x alone from {x,y} and of both x and y from X, yielded
by the BDR (I), cannot be rationalized in terms of any binary weak preference relation
defined over X. In contrast, the BDR (II), which is a weakly positionalist (II) SDR,
does not satisfy the property of independence of irrelevant alternatives for SDRs, but
it is easy to see that, for every preference profile, the social choices yielded by the
BDR (II) for the different issues are rationalizable in terms of an ordering over X.
So far I have commented on the Type I and Type II weakly positionalist SRRs
and SDRs. Very similar remarks hold for strongly positionalist (I) and strongly
positionalist (II) SRRs. I have considered the connection between the alternative
notions of positionality, as I have defined them, and the condition of independence
of irrelevant alternatives as formulated by Arrow and certain counterparts of this
condition. However, one can think of many versions of Arrow's independence of
irrelevant alternatives, and one can investigate the relation between these and the notion
of positionality. For a detailed and illuminating discussion of several such versions of
Arrow's condition of independence of irrelevant alternatives, the reader may refer to
Hansson (1973).
It is easy to see that, if a score-based decision rule with domain D(L) is such
that, for every positive integer m, the scoring vector for m-alternative issues is strictly
monotonic, then it will satisfy the principle of Pareto optimality (note that when the
permissible individual preferences are constrained to be linear orderings, there is no
distinction between weak and strict Pareto optimality). Consider a society S, an issue A
and a profile Rs of linear orderings. Suppose the score-based decision rule is such
that, for every positive integer m, the scoring vector for m-alternative issues is strictly
monotonic. Then, if a voter i strictly prefers an alternative x in an issue A to another
alternative y in A, the score received by x from i will be higher than the score received
by y from i. Therefore, if every individual strictly prefers x to y, then, given any
issue A containing both x and y, the total score received by x will be higher than that
received by y, and, hence, y will not be socially chosen from A. Similarly, a score-based
ranking rule with domain D(L) will satisfy the Pareto criterion.
The Pareto criterion can be violated by a score-based ranking rule G, if, for some
positive integer m, the scoring vector for m-alternative issues is not strictly monotonic
under G. Consider, for example, the plurality ranking rule under which the alternatives
in an issue are ranked on the basis of the number of first positions in the issue that each
alternative occupies in the preference orderings of the individuals. Given an issue A,
378 PK. Pattanaik

and a preference profile Rs for a society S, if x and y belonging to A never occupy

the top position in A in any individual's ordering, then we would have xl(A)y even
though x may be strictly preferred to y by every individual.

4.2. The simple majority rule and positionalist SDRs

Since the principle of simple majority has always had a powerful attraction, it is
of interest to consider to what extent positionalist SDRs are compatible with this
principle. In the context of an SDR, the principle of simple majority can be articulated
in terms of what have been called the Condorcet conditions.
Definition 4.1. An SDR H satisfies the weak Condorcet condition (WCC) [resp.
the strong Condorcet condition (SCC)] iff, for all (S, A, Rs) in the domain of H,
if M(S,A,Rs) X 0 [resp. M(S,A,Rs) # 0], then H(S,A,Rs) = M(S,A, Rs) [resp.
H(S,A, Rs) = M(S,A, Rs)].
Condorcet himself was aware of the possible conflicts between the BDR (I) and
the WCC, and, since he was a staunch advocate of the simple majority principle, he
considered the tension between the BDR (I) and the WCC to be a weakness of the
BDR (I). To see how the BDR (I) can conflict with WCC, consider the example given
at the beginning of Section 1. In that example, b is the sole socially chosen alternative
under the BDR (I), while WCC dictates the choice of a.
Nor is this conflict confined to the BDR (I) in the class of weakly positional-
ist (I) SDRs. Indeed, as a number of important results of Gardenfors (1973) show,
every score-based decision rule conflicts with WCC. The following proposition follows
from Theorem 4.1 of Girdenfors (1973) 12
Proposition 4.2. Every score-based decision rule with domain D(L) violates WCC
and hence SCC.
In view of Proposition 4.2, it is of interest to note that the runoff procedure based on the
BRR (I) (see Section 3.2) does satisfy WCC. [Therefore, by Proposition 4.2, it follows
that the runoff procedure based on the BRR (I) is not a score-based decision rule]. The
following proposition is due to Girdenfors (1973) [for a related result, see Smith's
(1973) Theorem 3].
Proposition 4.3. Assume that L is the set of admissible individualpreferences. Then the
runoff system based on the BRR (I) satisfies WCC. Furtherfor all (S, A, Rs) in D(L),
M(S, A, Rs) is a subset of the set of alternatives chosen under the runoffsystem based
on the BRR (I).

12 Grdenfors (1973) proved his result for a class of decision procedures which, insome ways, is more
general than the class of SCDRs: the decision procedures considered in Giirdenfors's Theorem 4.1 do
not necessarily satisfy the property of neutrality (i.e., symmetric treatment of the different alternatives),
while SCDRs do satisfy this property.
Ch. 7: Positional Rules of Collective Decision-Making 379

How worrying is the violation of WCC by all SCDRs? One response to this problem
can be to accept WCC as a compelling condition, and to see 'how frequently' specific
SCDRs (more generally, specific SDRs) violate WCC. An important paper of Merrill
(1984) approached the problem through computer simulation. A more recent paper by
Vandercruyssen (1999) pursues the approach of Merrill (1984) for a larger number
of voters and a larger list of voting procedures [see also Gehrlein and Lepelley
(1999)]. Essentially, the approach is to use the computer to simulate elections with
some given number of voters and given number of alternatives. Given a randomly
chosen preference profile for the voters, one checks whether there exists a strict
Condorcet winner, i.e. an alternative which has a strict majority over every other
alternative. Suppose given the number of alternatives and the number of voters,
g' out of g randomly chosen profiles yield a strict Condorcet winner. Then, for each
of these profiles, one checks whether a given SDR yields the strict Condorcet winner
as the sole outcome. Now suppose the SDR yields the strict Condorcet winner as
the sole outcome for g" out of the g' randomly chosen profiles where the strict
Condorcet winner exists. Then one can define g"/g' to be a measure of how well the
SDR performs when judged in terms of WCC. One of the findings of Vandercruyssen is
that, with 25 voters, this index of "Condorcet efficiency" for the BDR (I) is 0.8785 for
3 alternatives and 0.8357 for 20 alternatives. For the plurality decision rule the figures
are 0.7363 for 3 alternatives, and 0.1933 for 20 alternatives. [Of course, the index is 1
for the Black decision rule (I), since, by definition, whenever a strict Condorcet winner
exists, the Black decision rule will choose it]. While such simulation results have much
interest, insofar as they give us insight into the relationship between WCC and the SDR
under consideration, one would not necessarily view the high Condorcet efficiency of
an SDR as a justification for the SDR or the low Condorcet efficiency of an SDR
as a weakness of the SDR, unless one accepts WCC as a basic intuitive requirement.
For example, if one's intuition was more in favour of the Borda rule, then one could
just as well compute the index of the Borda efficiency of different voting procedures,
following an analogous method [indeed, Vandercruyssen (1999) computes the Borda
efficiency indices, as well as the Condorcet efficiency indices, for different voting
procedures].
Though the WCC has its own intuitive attraction, it is by no means a compelling
condition, and, therefore, the violation of WCC by many SDRs, including all score-
based ranking rules, need not necessarily be considered a damning shortcoming
of these social decision rules. Even before undertaking a detailed examination
of the structural properties of these SDRs, it is easy to see that some of the
positionalist (I) SDRs, such as the BDR (I), have direct intuitive appeal, and that there
can be situations where, from a purely intuitive point of view, the choice under the
BDR (I) appears more attractive than the Condorcet choice. Black (1976) visualizes
one such situation when he writes, "A type of example in which, to common sense, the
Borda count would seem to give the more appropriate result, would be that in which
ah, say, chosen by the Condorcet criterion, gets a bare majority against each of the
other motions, whereas ak, chosen by the Borda count, although just failing to get a
380 PK. Pattanaik

majority against ah, yet is able to get a substantial size of majority against each of
the other motions". The type of situation that Black (1976) seems to have in mind is
illustrated by the example given at the beginning of Section 1, where out of 7 voters
four have the linear ordering aPi bPi c and three have the linear ordering bPi cPi a. As
we noted, one can argue for the choice of b [the alternative prescribed by the BDR (I)]
rather than a, on the ground that b is either the best or the second-best option for
every voter and never the worst option for any voter, while a is the worst option for
3 voters. It is also possible to justify the choice of b from a different intuitive angle.
Note that, in the comparison between a and b, b gets 3 votes, and in the comparison
between b and c, b gets 7 votes. Thus b gets altogether 10 votes out of the total of
14 votes in the two pairwise comparisons in which b is involved. The corresponding
figures for a and c are 8 and 3 respectively. Thus, the chosen alternative under the
BDR (I) maximizes the total number of votes received by an alternative in all the
pairwise comparisons involving that alternative. A little thought will confirm that this
is not peculiar to the particular example that we have chosen, and that it is a general
characteristic of the BDR (I). To see this, given an issue A, an alternative x in A, and a
linear ordering Ri of the individual i, consider the score r(x, A, Ri) that x receives from i
under the BDR (I). This is simply the number of alternatives in A which are below x
in the linear ordering Ri. Therefore, in all the pairwise comparisons that involve x
and some other alternative belonging to the issue A, x will get r(x, A, Ri) votes from
individual i. It follows that, under the BDR (I), the total score received by x for all
the individuals is simply the total number of votes that x will receive from all the
individuals in all the pairwise contests involving x and any other alternative a in A.
Thus, ifx wins in A under the BDR (I), then x must maximize the total number of votes
received by an alternative in A in all the pairwise contests involving that alternative
and any other alternative in A. Interpreted in this fashion, the BDR (I) does seem to
have considerable prima facie appeal. In view of such intuitive appeal of the BDR (I),
it is not surprising that the structural properties of the Borda rule have been the subject
of several detailed investigations. I consider some of these in the next section.

5. The structure of the Borda rule

5.1. Axiomatic characterizationsof the Borda rule

A priori, what properties would one like a positionalist SRR or SDR to satisfy?
Further, what plausible properties do certain familiar SRRs and SDRs satisfy? Can one
characterize different positionalist SDRs in terms of sets of plausible properties? An
important part of the literature has been devoted to these issues. In particular, the Borda
decision rule (I) and the Borda ranking rule (I) are some of the positionalist social
decision procedures to have been investigated most thoroughly through axiomatic
analysis [see, among others, Gardenfors (1973), Young (1974, 1975), Hansson and
Sahlquist (1976), Fishburn and Gehrlein (1976), Fine and Fine (1974b), Nitzan and
Ch. 7: PositionalRules of Collective Decision-Making 381

Rubinstein (1981) and Saari (1990)]. The axioms, as well as the proof techniques, used
by different writers in this context often differ considerably. I shall not examine the
detailed differences between the alternative sets of axioms which have been used to
provide the different characterizations of the Borda rule; instead, I shall discuss only
one characterization each for the BDR (I) and the BRR (I).
One of the most well-known and elegant characterizations of the BDR (I) is due
to Young (1974). An important feature of Young's (1974) characterization is that his
formal framework allows the society to vary 13.
Definition 5.1 below introduces the properties of an SDR used by Young 14, together
with the additional properties of anonymity and monotonicity for SDRs which I shall
refer to later.
For all disjoint sets S and S' and all profiles Rs and R, for S and S',
respectively, let (Rs +R,) denote a preference profile R ,/s for S US', such that,
for all i E S, R' = Ri, and, for all i E S', Ri = R.
Definition 5.1. Let H be an SDR with domain D(W'). H satisfies:
(5.1.1) neutrality iff, for every (S,A,Rs) E D(W') and every permutation on X,
H(S,c (A),R) = a(H(S,A,Rs)), where, for all i S, and all x,y EX,
[o(x)Rao(y) iff xRiy];
(5.1.2) anonymity iff, for all societies S and S' such that #S = #S', every issue A, and
all Rs, R, E W's, [if there exists a one-to-one function y from S to S' such
that, for all i E S, Ri = Rt(,)], then H(S,A,Rs) = H(S,A, R,);
(5.1.3) monotonicity iff, for every society S, every issue A, all Rs,R E W's,
and all x E A, if [x G H(S,A,Rs)] and [for all y,z C X - {x}, Rs/{y,z}=
R/{y,z}] and [for all z EX - {x} and all i C S, (xliz implies xRlz) and
(xPi z implies xPlz)], then x E H(S,A, R);
(5.1.4) consistency iff, for all (S, A, Rs), (S', A', Rs) E D(W'), if [S and S' are disjoint,
A =A', and H(S,A, Rs) n H (S',A', R) 0], then [H(S U S',A,Rs + R)=
H(S, A, Rs) n H(S', A', R})];
(5.1.5) cancellation iff, for all (S,A,Rs) E D(W'), [for all distinct x,y E X,
#{i G S: xPiy} = #{i E S: yPix}] implies H(S,A,Rs) = A;
(5.1.6) faithfulness iff, for every (S,A, Rs) E D(W'), and every i E N, [S = {i}] im-
plies that H(S,A, Rs) is the set of Ri-greatest elements in A.
Neutrality requires that the SDR should treat the alternatives in a symmetric
fashion. Anonymity requires that the SDR should treat the individuals symmetrically.
Monotonicity requires that, other things remaining the same, if an initially chosen

13 This is also a feature of the contributions of Smith (1973) and Nitzan and Rubinstein (1981). While
the variability of the society is not a very common trait in most areas of social choice theory, it has
proved very useful in the axiomatic analysis of positionalist group decision procedures.
14 Young (1974) defined neutrality, consistency, cancellation and faithfulness in a framework where the
issue was assumed to be fixed and there was no distinction between the universal set and an issue. Here
the definitions have been adapted for the case of variable issues.
382 PK. Pattanaik

alternative moves up in some individuals' preferences without moving down in

anyone's preference, then that alternative continues to be chosen by the society 15
The property of consistency is concerned with situations where we merge two disjoint
societies to form a new society. Suppose we start with two disjoint societies S and S',
and preference profiles Rs for S and R' for S'. Suppose given an issue A, x is a
chosen alternative for the society S with profile Rs, as well as for the society S' with
profile R'. Then consistency requires that if we merge the two societies and the
individuals in S and S' retain their original preferences, then x must now be a chosen
alternative for the society resulting from the merger, given the unchanged original
preferences of the individuals. Further, if at all there is an alternative chosen by
both the societies in the original situation, then consistency requires that, after the
merger with unchanged preferences of the individuals, the new society must not
choose any alternative originally rejected by at least one of the two societies in
the original situation. Cancellation requires that, if for all x,y c X the number of
individuals who strictly prefer x to y equals the number of individuals who strictly
prefer y to x, then, for every issue, the society must choose all the alternatives in the
issue. Finally, faithfulness is a compelling property that postulates that, in a society
with one individual, the set of best alternatives, in any issue, for that individual must
be the set of socially chosen alternatives for that issue.
The following proposition follows from Theorem 1 of Young (1974) 16:
Proposition 5.2. [Young (1974)]. Every SDR with domain D(L) is the BDR (I)
with domain D(L) iff it satisfies independence of irrelevant alternatives, neutrality,
consistency, cancellation andfaithfulness.
Nitzan and Rubinstein (1981) give an interesting characterization of the BRR (I) for the
case where individual weak preference relations are assumed to be reflexive, connected
and antisymmetric, but not necessarily transitive [later I explain their definition of the
BRR (I) for this case]. The following definition introduces a number of properties of an
SRR, some of which are used by Nitzan and Rubinstein (1981) in their characterization
of the BRR (I):
Definition 5.3. Let G be an SRR with domain D(W'). Then G satisfies
(5.3.1) neutrality iff, for every society S, every issue A, all Rs, R E WIS, and every
permutation a on X, if [for all i C S and all x,y E X, xRiy iff a(x) R o(y)],
then [for all x,y E A,xR(A)y iff o(x)R'(u(A)) a(y)];
(5.3.2) anonymity iff, for all societies S and S' such that #S=#S', every issue A,
and all Rs, R, W's, if [there exists a one-to-one function y from S to S'
such that, for all i in S, Ri = R,(i)], then R(A) = R'(A), where R(A) = G(S, A, Rs)
and R'(A) = G(S, A, R,);

15 The properties of neutrality, anonymity and monotonicity were first introduced by May (1952) for
the case of exactly two alternatives.
16 Young proved his theorem in a framework where the issue was fixed.
Ch. 7: PositionalRules of Collectiue Decision-Making 383

(5.3.3) monotonicity iff, for every society S, every issue A, all Rs,R E W 's , and
all distinct x,y E X, if [for all y,z X- {x}, Rs/{y,z} = R/{y,z}] and
[for all z E X - {x} and all i E S, (xli z implies xR i z) and (xPiz implies xP i z)],
then [xI(A)y implies xR'(A)y, and xP(A)y implies xP'(A)y];
(5.3.4) semistrict monotonicity iff, for every society S, every issue A, all Rs,R sE
W'S, and all distinct x,y E X, if [xR(A)y] and [Rs and R are identical
except that, for some z E A - {x,y} and some i E S, (zPix and xP'iz)], then
[xP'(A)y];
(5.3.5) consistency iff, for all disjoint societies S and S', every issue A, all Rs, R, C
W'S, and all x,y E A, [if xR(A)y and xR'(A)y, then xR+(A)y] and [if
(xR(A)y and xR'(A)y) and (xP(A)y or (xP'(A)y), then xP+(A)y], where
R(A) = G(S,A,Rs),R'(A)= G(S',A,Rs,) andR+(A) = G(S U S',A,Rs + R,);
(5.3.6) cancellation iff, for all (S,A,Rs) e D(W'), [all x,y c X, #{i C S: xPiy}=
#{i S: yPix}] implies [for all x,y E A, xI(A)y].
The interpretations of neutrality, anonymity, monotonicity, consistency and cancella-
tion for SRRs are very similar to the interpretations of the corresponding properties
for SDRs. However, semistrict monotonicity, which was introduced by Nitzan and
Rubinstein (1981) and which they called monotonicity, deserves some comment. It
requires that, if, given a society, an issue A, and a preference profile for the society,
x belonging to A is socially at least as good as y belonging to A, and, if, subsequently,
x moves up in some individuals' preferences vis-a-vis some third alternative z in A
(the preference profile remaining the same otherwise), then, after the change in the
individual preferences, x must be strictly preferred to y. In terms of logical strength,
semistrict monotonicity is not comparable with monotonicity. Insofar as it lets x be
strictly preferred to y in the new situation when it was only weakly preferred to y
in the initial situation, it is more demanding than monotonicity. However, unlike
monotonicity, it does not by itself ensure that, if x moves up vis-a-vis y in some
people's preferences, the preference profile remaining the same otherwise, then the
social ranking of x vis-a-vis y must not go against x.
The following proposition follows directly from the theorem of Nitzan and
Rubinstein (1981). As noted earlier, Nitzan and Rubinstein assumed individual weak
preference relations to be reflexive, connected and antisymmetric, but not necessarily
transitive, over X. Recall that we defined the BRR (I) only for the case where the
individual preferences are linear orderings. However, for the purpose of the Nitzan-
Rubinstein theorem stated below, the BRR (I) is to be understood as being based on
a score system, where the score received by an alternative x from an individual is the
number of alternatives in the issue which are strictly less preferred than x, in terms of
that individual's weak preference relation 17.

17 Nitzan and Rubinstein (1981) use a framework where the issue is fixed. Since I allow the issue to
vary, the properties and the theorem of Nitzan and Rubinstein have been restated in this context.
384 PFK. Pattanaik

Proposition 5.4. [Nitzan and Rubinstein (1981)1. Let W* be the set of all reflexive,
connected and antisymmetric (but not necessarily transitive) binary relations over X.
Then a social orderingrule with domain D(W*) is the BRR (I) with domain D(W*) iff
it satisfies independence of irrelevant alternatives, neutrality, semistrict monotonicity,
consistency, and cancellation.

5.2. A Bayesianjustificationfor the BDR (I)

An ingenious justification for the Borda decision rule (I) is given by Young (1988,
1997), using probabilistic reasoning of a type that was initiated by Condorcet (1785)
in the context of the simple majority rule [for a lucid exposition of this strand of
Condorcet's thoughts, see Young (1986, 1988, 1997)]. First, suppose the issue contains
only two alternatives, a and b. Assume that one of the two alternatives is 'truly' better
from the social point of view. A priori, each alternative has the same probability of
being the correct social choice. In casting her vote, a voter exercises her judgement
about which of the two alternatives is socially better. The judgement of each person
has a fixed probability p of being right and, hence, a fixed probability, 1 -p, of being
wrong. Further, we assume that the judgement of each person is independent of other
persons' judgements. Now suppose, 6 out of 10 voters vote for a and 4 for b. Then
standard Bayesian reasoning tells us that the posterior probability of a being the better
social choice is k-p6 (l -p) 4 and the posterior probability of b being the correct social
choice is kp 4 .(1 _p) 6 , where k is a specific constant. Given that p > , it is clear that
a is more likely to be the correct social choice.
Now consider three alternatives, a, b and c. There are 6 possible linear orderings
over a, b and c: aPbPc, aPcPb, bPaPc, bPcPa, cPaPb, cPbPa. Assume that each of
these has the same a priori probability of being the correct social ordering of the three
alternatives. Now suppose that the observed orderings of 11 individuals are as follows:
aPibPic for i = 1,...,5; bPicPiafor i = 6,...,9; cP 1obPloa; and cP1 laPllb. Then
6 individuals will vote for a over b and 5 individuals will vote for b over a; 9 individuals
will vote for b over c and 2 individual will vote for c over b; and 6 individuals will vote
for c over a and 5 individual will vote for a over c. Given these voting patterns, the
posterior probabilities of each of the six linear orderings being the true social ordering
are as follows:
aPbPc: k' .p2 0 . (1 - p) 13 ;
aPcPb: k' p13 (1 - p)20;
bPaPc: k' .p19 (1 - p)'4;
bPcPa: k' .p 2 0 (1 - p)13 ;
cPaPb: k' .p14 (1 - p) 9 ;
cPbPa: k' p' 3 . (1 - p)20,
where k' is a specific constant. Then the posterior probability that a is the correct social
choice is k[p2 0 . (1 - p)13 + pl3 . (1 - p) 20 ]; the posterior probability that b is the
Ch. 7: PositionalRules of Collective Decision-Making 385

correct social choice is k'[p 19 . (1 - p)14 + p20 . (1 - p) 13]; and, finally, the posterior
probability that c is the correct social choice is k'[p 14 (1 - p)19 + p13 (1 - p)20].
It can be shown that, when (p - ) is sufficiently small, the posterior probability of
being the correct social choice is the highest for b among the three alternatives. Note
that b is also the choice under the Borda decision rule (I). In general, Young (1988)
shows that, if p is sufficiently close to , then the choice most likely to be 'correct'
is the choice yielded by the Borda decision rule (I). This conclusion depends on the
assumption that p is sufficiently close to , and on the intuition that there exists a
socially correct choice. As Young (1997, p. 182) points out, the latter intuition is
appropriate when the voters agree about the objective but cannot agree about the
appropriate policy for achieving the agreed objective, and the voting is intended to
settle this issue of which policy to adopt. However, the intuitive assumption that
there exists something that one can call the correct social choice becomes tenuous
when voters disagree about the social objective itself (this may be the case when, for
example, there is disagreement about what constitutes a just distribution of income)
and voting is a means of reconciling differences about the objective rather than a way
of deciding the right means to achieve some agreed objective. Despite this limitation,
Young's (1988) result is a striking one, and provides for the BDR (I) a type of
justification very different from those provided by the axiomatic characterizations.

5.3. A Paretianargumentfor the BDR (I)

As the BDR (I) is an SCDR with strictly monotonic scoring vectors, the outcomes
under the BDR (I) will be always Pareto-optimal. This was noted by Sen (1977).
However, as Farkas and Nitzan (1979) point out, there is also a deeper connection
between the BDR (I) and the principle of unanimity. The Farkas-Nitzan reasoning
involves the following stages:
(i) Assume that L is the set of all admissible preferences, and consider a society S,
an issue A, and a preference profile Rs for S.
(ii) For every individual i in S, consider the (linear) ordering Ri figuring in Rs, and
let its restriction to A be called Ti. For any given alternative x in A, find the
minimum number of 'steps' by which Ti can be converted into a linear ordering
T ' over A such that x is T'-greatest in A, where a 'step' consists of a transition
from one linear ordering over A to another linear ordering over A through an
inversion of the ranking of exactly two alternatives which are 'adjacent' in the
former linear ordering over A. Call this minimum number K(x, A, Ri). To give an
example, let X = {a, b, c, d}, A = {a, b, c} and aPidPi bPi c. In that case, we have
aTbTc. Then we need at least two steps to convert T into an ordering T' over A
such that c is T'-greatest in A. The first step will be to interchange the position of
b and c so as to get the linear ordering T" where we have aT"cT"b. The second
step will be to interchange the positions of a and c to go from T" to T' such that
cT'aT'b. Intuitively, K(x,A, Ri) may be viewed as a measure of the 'margin' by
which x misses being i's best alternative in A, given that Ri is the ordering of i.
386 PtK. Pattanaik

(iii) Now let K(x,A, Rs) denote the sum of all K(x, A, Ri) such that i E S. It is then
easy to see that K(x,A,Rs) represents the smallest total number of 'steps' by
which Rs can be converted to a preference profile R' for S, such that x is the
best option in A for every individual in the preference profile R'. Intuitively,
K(x, A, Rs) measures the 'margin' by which x misses being the unanimously best
alternative in A, given that the preference profile is Rs.
(iv) Farkas and Nitzan show that:

If, given the society S, the issue A. and the preference profile Rs,
x* belonging to A is a winner under the BDR (I), (5.1)
then we must have K(x*,A,Rs) > K(x,A,Rs) for all x E A.

(5.1) can be interpreted as showing that, among all the alternatives in issue A, a
winner under the BDR (I) falls short of being the unanimously best alternative in A by
the smallest margin. Thus, (5.1) provides another intuitive and interesting justification
for the BDR (I).

5.4. A general comparison of the Borda procedure and other positionalistvoting

rules
Saari (1990) has developed a powerful technique for analysing all the possible results
that can materialize for different issues under the BDR (I) for any given profile of
linear preference orderings. An important general conclusion that emerges from Saari's
(1989, 1990) analysis and that favours the BDR (I) is that "one can expect more
consistency among social choice mappings if their outcomes are based on the [Borda
count] rankings rather than the rankings of any other positional voting method" and
that "the Borda method is the unique positionalistmethod to minimize the kinds and
number of paradoxes that can occur". (It may be worth noting here that, not only
does Saari's analytical technique make it possible to establish such very general new
results, but it also provides an alternative way of deriving many results familiar from
other writers' contributions on positionalist voting procedures).

6. Score-based rules, dominance-based scoring rules, and runoff procedures

6.1. Score-based ranking and decision rules

The structure of score-based ranking rules and score-based decision rules, which
include, respectively, the BRR (I) and the BDR (I) as special cases, have been
explored by several writers. Smith (1973) gives an axiomatic characterization of score-
based ranking rules and an axiomatic characterization of a class of ranking rules
(called 'generalized point systems') which constitute extensions of score-based
ranking rules. Similarly, Young (1975) gives an axiomatic characterization for score-
based decision rules and also for a class of SDRs which include, besides SCDRs,
Ch. 7: PositionalRules of Collective Decision-Making 387

certain refinements of SCDRs. The contributions of Smith and Young are very similar:
they use similar axioms, and, in particular, both of them permit the society to vary.
Fishburn (1973a,b) gives a characterization of "summation social choice functions",
which, in many ways, constitutes a generalization of score-based decision rules. Unlike
Smith and Young, Fishburn assumes the society to be fixed. In Proposition 6.2 below,
I present a version of Smith's (1973) result 18.
A little reflection will show that SCDRs, as well as SCRRs, satisfy independence
of irrelevant alternatives, neutrality, anonymity and consistency (see Definitions
5.1 and 5.3). It is also clear that an SCDR satisfies monotonicity (see Definition 5.1)
if, and only if, for every positive integer m, the scoring vector for m-alternative issues
under the SCDR is monotonic, and similarly for an SCRR. What Smith (1973) shows
is that, if we add one more property, the 'Archimedean property', to independence of
irrelevant alternatives, anonymity, neutrality and consistency, then we have a set of
properties which characterizes the class of all score-based decision rules.
Let S and S' be two societies, let Rs and R be, respectively, the preference
profiles of S and S', and let t be any positive integer. We say that (S',R',) is a t-fold
replication of (S, Rs) iff there exists a partition {S,S , ... , S#s} of S' and a one-to-
one function V from S to {S, S2, ... , Ss}s, such that [#SI = #S2 = ... = #S s = t] and
[for all i G S, Ri = R for all j E V(i)]. Thus, intuitively, a t-fold replication of (S, Rs)
is arrived at by: (i) considering a different society where the number of individuals
is t times that in S, and (ii) taking a preference profile for this new society such
that it replicates t times each weak preference relation figuring in the original
profile Rs.
Definition 6.1. An SRR G satisfies the Archimedean property iff the following holds
for every pair of societies S and S', and all profiles of linear preference orderings
Rs and Rs , for S and S', respectively:

For every issue A and all x,y E A,

there exists a positive integer t such that, for all t, t > t,
whenever (S",Rs,,) is a t-fold replication of (S', R,), and S" and S are disjoint,
xP'(A)y implies xP"(A)y,
where R'(A) = G(S',A, R,) andR"(A) = G(S U S"',A,Rs + R,,).

Essentially, what the Archimedean property requires is this: Given an issue A, if one
society S', with some preference profile RS, strictly prefers x to y (x and y being two
alternatives in A), then for any other society S with any preference profile, one can
find a sufficiently large number of replications of S' (with corresponding replications
of the preference profile Rs , of S'), such that, when we add all these replications to S,

18 The main difference between the version presented here and the version proved by Smith (1973) is
that Smith works with a fixed issue, while I allow the issue to vary.
388 P8K. Pattanaik

the added individuals will 'overwhelm' whatever preferences the members of S had, so
that x will be strictly preferred to y by the enlarged society resulting from the addition.
The following proposition follows from Smith's (1973) Theorem 1:
Proposition 6.2. [Smith (1973)1. An SRR with domain D(L) is a score-based
ranking rule with domain D(L) iff it satisfies independence of irrelevant alternatives,
anonymity, neutrality, consistency and the Archimedean property.
Proposition 6.2 provides a characterization of the class of score-based decision rules
in terms of a set of very plausible properties of SRRs.

6.2. Score-based ranking rules and the principle of dominance

The class of score-based ranking rules is very wide, though it does exclude many
positionalist rules with a similar formal structure that do not satisfy anonymity
and neutrality. Even if one restricts the class further by requiring the property of
monotonicity (an SCRR that violates monotonicity can hardly be considered seriously),
the class still remains very wide. Let E be the class of all monotonic SCRRs with
domain D(L). Can one make any progress in deriving the social ranking of the
alternatives in an issue on the basis of the general intuition underlying the monotonic
SCRRs without, however, committing oneself to any specific SCRR in E? One possible
answer to this question can be to say that, given (S, A, Rs) E D(L), an alternative x in A
is socially at least as good as another alternative y in A iff, for the same (S,A,Rs),
every SCRR in E would lead to the conclusion that x is socially at least as good as y.
This intuition has been investigated in detail in a major contribution by Fine and Fine
(1974a,b).
Definition 6.3. The dominance-based scoring rule (DSR) with domain D(L) is
the SRR J with domain D(L) such that, for every (S,A, Rs) e D(L) and for all
x,y EA, xRJ(A)y iff xRG(A)y for all G in E, where RJ(A)=J(S,A,Rs) and
RG(A) = G(S,A, Rs).
Note that the condition for x to be socially at least as good as y under the DSR with
domain D(L) is extremely strong: x has to be socially at least as good as y under every
G belonging to E, the class of all monotonic SCRRs, before x can be considered to be
socially at least as good as y under the DSR. Thus, if at all there exists any G in E that
one considers acceptable, one would not have any hesitation in saying that x is socially
at least as good as y when the DSR declares x to be socially at least as good as y. This
attractive feature of the DSR, of course, has a price: the social binary relation yielded
by the DSR is not necessarily connected, though it is always reflexive and transitive.
Thus, for some (S,A,Rs) E D(L) and for some x and y in the issue A, the DSR may
not be able to compare x and y.
Example 6.4. Consider the society S= {1,2,3}, the issue A = x,y,z} and the
following profile, Rs, of linear orderings for this society:
Ch. 7: Positional Rules of Collective Decision-Making 389

R1 R2 R3

x y Y
y z z
z x x

Now consider a monotonic SCRR G such that, given A, G attaches the scores
7, 2 and 1, respectively, to the first, second and third positions. Consider also another
monotonic SCRR G' that assigns the scores 7, 6 and 1, respectively, to the first, second
and third positions. Given (S,A,Rs), x will be socially better than z under G, while
z will be socially better than x under G'. Therefore, x and z will turn out to be non-
comparable under the DSR with domain D(L). ·
There is an alternative, illuminating way of viewing the DSR. For every (S, A, Rs) E
D(L), every x in A, and every integer k (#A > k > 1), let Axk(S, A, Rs) denote
{i E Slr(x,A,Ri) < k}, i.e., the number of individuals in society S, for whom, given
the preference profile Rs, x has the t-th position in A for some t < k. Let A(S, A, Rs)
denote the vector (Al(S,A, Rs), A(S, A, Rs),... , AA(S, A, Rs)). Thus, AX (S, A, Rs) is a
#A-dimensional vector such that the first component of the vector is the number of
individuals in S who give x the first position in issue A, the second component of
the vector is the number of individuals in S who give x either the first or the second
position in issue A, and so on.
Example 6.5. Consider (S,A,Rs) as specified in Example 6.4. Then Ax(S,A,Rs)=
(x (S,A, Rs), A(S,A, Rs), X(S, A, Rs) = (1, 1, 3).
Fine and Fine (1974a) prove the following proposition 19 :
Proposition 6.6. Let J denote the DSR with domain D(L). For all (S,A,Rs) C D(L),
and all x,y C A, xR(A)y iff AX(S,A,Rs) > AY(S,A,Rs), where R(A) = J(S,A,Rs) and
[AX(S, A, Rs) > AY(S, A, Rs)] denotes that [every component of the vector AX (S, A, Rs)
is at least as great as the corresponding component of the vector AY(S,A,Rs)].
Thus, under the DSR with domain D(L), x is socially at least as good as y iff [the
number of individuals who give x the first rank is at least as great as the number of
individuals who give y the first rank; and the number of individuals who give x the
first or the second rank is at least as great as the number of individuals who give y
the first or the second rank; and so on].
Example 6.7. Let J be the DSR with domain D(L). Consider the triple (S, A, Rs) in Ex-
ample 6.4. There A5 (S, A, Rs) = (1, 1,3), AY(S, A, Rs) = (2, 3, 3), Az(S, A, Rs) = (0, 2, 3).

19 Strictly speaking, Fine and Fine (1974a) prove a more general result insofar as they did not assume
the individual preference orderings to be necessarily linear.
390 PK. Pattanaik

Therefore, by Proposition 6.6, we know that yR(A)x, not [xR(A)y], not xR(A)z,
not [zR(A)x], etc., where R(A) = J(S,A,Rs). U
It is clear that the DSR with domain D(L) satisfies the properties of anonymity,
neutrality and monotonicity for an SCRR. For other properties of the DSR, and also
for an axiomatic characterization, see Fine and Fine (1974a).

6.3. Runoff systems

In Section 3.2, I introduced a positionalist (I) SDR which I called the runoff
system based on the BRR (I). This SDR is just one member of a large class of
positionalist (I) SDRs which involve successive rounds of elimination of alternatives
from the issue under consideration. A tight, formal definition of the class of all such
procedures based on successive elimination of alternatives is rather cumbersome and
I shall not attempt it here. Instead of formally defining the class, I describe the general
structure of these SDRs, which I shall call runoff systems. As I present it, a runoff
system is an SDR. However, one can also articulate the basic intuition in terms of an
SRR [see Smith (1973)].
Given (S, A, Rs) E D(L), a runoff system H specifies the socially chosen alternatives
as follows: First, it applies a monotonic score-based ranking rule G to (S,A,Rs)
so as to get an ordering R(A) over A. If all the alternatives in A belong to one
indifference class defined in terms of R°(A), then A is declared to be H(S,A,Rs).
However, if R°(A) defines more than one indifference class in A, then the runoff
system requires some alternatives to be eliminated from A; a condition which such
elimination must satisfy is that one should not eliminate an alternative x from A
while leaving intact an alternative y in A such that xR°(A)y. Thus, intuitively, one
eliminates certain alternatives from A, starting from the bottom of R°(A) and going
upwards. Let A I be the set of alternatives in A which survive the process of elimination
in the first stage. Then one applies a monotonic score-based decision rule G to
(S,A 1,Rs) to get an ordering R'(Al) (it is possible that G is the same as GO). If
all the alternatives in Al are indifferent in terms of R 1(A1), then H(S,A,Rs)= A 1.
But, if Rl(Al) defines more than one indifference class in Al, then H requires
some alternatives to be eliminated from Al, the process of elimination being again
subject to the condition that one does not eliminate an alternative x from Al while
leaving intact an alternative y in Al such that xR'(A 1)y. This process is continued
up to a certain number of stages, at the end of which the alternatives in the highest
indifference class corresponding to the last stage are declared to be the socially chosen
alternatives; it is, of course, possible that the process is continued until there is only one
indifference class [as in the case of the runoff system based on the BRR (I), discussed
earlier].
Besides the runoff system based on the BRR (I), several other runoff systems have
been discussed in the literature. Consider the following examples:
Ch. 7: PositionalRules of Collective Decision-Making 391

Example 6.8.
(6.8.1) The runoff system based on the plurality ranking rule: At each stage,
one applies the plurality ranking rule which ranks alternatives on the basis
of the first positions secured by an alternative in the set of alternatives under
consideration, and eliminates the alternatives with the smallest number of first
positions (provided, of course, not all alternatives get the same number of first
positions). One continues like this until one arrives at a stage where all the
alternatives secure the same number of first positions, in which case they are
all declared to be socially chosen. This procedure is also known as the Hare
rule or alternative vote or single transferable vote.
(6.8.2) The runoff system based on the anti-plurality ranking rule: Under this
system, at each stage one eliminates the alternatives securing the largest
number of last positions among all the alternatives being considered in that
stage (provided, of course, not all the alternatives in that stage get the same
number of last positions). Otherwise, the structure is similar to that of the
runoff system based on the plurality rule.
(6.8.3) The Nanson runoff system: The structure of this runoff system is similar to
that of the runoff system based on the BDR (I), except that, in each stage,
instead of eliminating the alternatives with the lowest total Borda score at
that stage, we eliminate the alternatives with less than the average of the total
Borda scores of all alternatives at that stage. ·
The runoff systems have a complex structure, and we do not have axiomatic
characterizations of any class of runoff systems. However, Smith (1973) identifies
certain properties of different classes of runoff systems. One of the results of
Smith (1973) is of particular importance in assessing the appeal of runoff systems.
Smith (1973) shows that a large class of runoff systems violate the property of
monotonicity, so that, if an alternative x, chosen initially by the society, moves up in
some voters' preferences (the initial preference profile remaining the same otherwise),
then, x may get socially rejected after such change of the voters' preferences. To see
this, consider the runoff systems which satisfy the following conditions:
(i) if there are exactly three indifference classes in the first stage, then only the lowest
indifference class is eliminated to effect the transition to the next stage;
(ii) every ranking rule (assumed to be monotonic) involved in the runoff system is such
that, for every issue with at least two alternatives, the score or points specified for
the first position (defined in terms of an individual's linear ordering) in that issue
is strictly higher than the score or points specified for the last position (defined in
terms of an individual's linear ordering) in that issue.
Note that runoff systems based on the BRR (I) or the plurality ranking rule or the
anti-plurality ranking rule satisfy these two conditions.
The following example of Smith (1973) demonstrates that every runoff sys-
tem satisfying conditions (i) and (ii) must violate monotonicity. Consider a soci-
ety S= {1,2,... ,37}, an issue A = {x,y,z}, and a preference profile Rs such that
392 P9K. Pattanaik

individuals 1,...,24 are equally distributed over the three linear orderings xPyPz,
yPzPx, and zPxPy; individuals 25,...,33 are equally distributed over the linear
orderings zPyPx, yPxPz, and xPzPy; and, finally, individuals 34,...,37 have,
respectively, the orderingsxP 34 YP3 4 z, xP 35 zP 35y, xP 36 YP36 z, and YP 37 xP37 z. Let the
runoff system be denoted by H. Suppose, given a 3-alternative issue, the SCRR that is
used in the first stage assigns a score of st, s2 and s3 to the first, the second and the third
positions, respectively. Then, given condition (ii), sl > s2 > s3 and s > s3. Then, in the
first stage, the total scores received by x, y and z are, respectively, 14sl + 12s 2 + 1 Is 3,
12sl + 13s2 + 12s3, and 1lsl + 12s 2 + 14s 3. Given sl > 2 > s3 and s > s3, it is clear
that [14sl + 12s2 + 1is 3 > 12sl + 13s2 + 12s3 > 1 sl + 12s 2 + 14s 3], and, hence, given
condition (i), z will be eliminated after the first round, and we will be left with {x,y}.
Suppose, given {x,y}, the SCRR used in the second stage assigns score s for the first
position and s for the second position. Then, by condition (ii), s > s. In the second
round x gets the total score of 22s' + 15s' and y gets 15s' + 22s'. It is, therefore, clear
that H(S,A,Rs) = {x}.
Now construct a preference profile Rs which differs from Rs only in that the
individuals in the group 25,...,33} who had the ordering zPyPx now have the
ordering zP'xP'y; and those in the group {25,..., 33} who had the ordering yPxPz
now have the ordering xP'yP'z. Then, going through calculations similar to that given
above and involving sl, s2 and s3 in the first stage and sl and s in the second stage, it
can be checked that y will be eliminated in the first round, and, finally, we would have
H(S, A,Rs)= {z}. This, of course, violates monotonicity, since x is chosen to start
with, and, when we move from Rs to R', the only change in the preference orderings
that takes place is that x moves up in some individuals' preference orderings.
Intuitively, monotonicity is a very weak and compelling condition for an SDR.
The violation of this basic property by the very large class of all runoff systems
which satisfy conditions (i) and (ii) makes one quite uneasy about all these
decision procedures.

7. Concluding remarks

While the interest in positionalist group decision procedures goes back to Borda
(1781), the study of these procedures received a major fillip following Arrow's (1951,
1963) seminal contribution. Many writers have felt that the positionalistic approach to
group decision-making provides a plausible way out of Arrow's impossibility result.
Whether the positionalist group decision rules really provide such an escape route may
be a matter of debate. But there is no doubt that the literature on these rules that has
developed over the last fifty years or so has increased immensely our understanding of
the structure of many group decision procedures actually used in real life. In reviewing
this literature, I have focussed only on some central concepts and results. Even so,
two features of positionalist rules stand out. The first is the richness of the class of
positionalist rules: the structural variety within this class is truly remarkable. The
Ch. 7: PositionalRules of Collective Decision-Making 393

second feature that emerges, thanks especially to the axiomatic analysis of several
writers, is that many positional rules have highly attractive properties. Given these
properties, it is not surprising that they have had a strong intuitive appeal for people
over a long period of time.
There are some important issues relating to positionalist group decision-making,
which I have not discussed here. The problem of strategic voting under these rules is
one such problem. It may be recalled that the vulnerability of Borda's rule to strategic
manipulations was one of the earliest criticisms levelled against the rule (Borda's retort
to the criticism was that his procedure was intended for honest men! 20). I have not
discussed the issue of strategic manipulation under positionalist rules partly because it
is discussed by Brams and Fishburn (2002) (Chapter 4 of this volume), and also partly
because it is a part of the broader issue of strategic manipulation of social decision rules
in general, which will be taken up in a chapter in a later volume of this Handbook. The
problem of proportional representation in multi-member constituencies is another issue
that I have not taken up in this chapter where the focus has been on positionalist group
decision procedures in the context of the classical problem of the society's choice of
a single alternative out of several mutually exclusive alternatives, on the basis of the
preferences, assumed to be known, of the members of the society.

References

Arrow, K.J. (1951), Social Choice and Individual Values, Ist Edition (Wiley, New York).
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York).
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, London).
Black, D. (1976), "Partial justification of the Borda count", Public Choice 28:1-15.
Borda (J.-C. de Borda) (1781), "M6moire sur les 61lections par scrutin", M6moires de l'Acad6mie Royale
des Sciences annee 1781, pp. 657-665. Translated in English by A. de Grazia (1953), "Mathematical
derivation of an election system", Isis 44:42-51.
Brams, S.J., and P.C. Fishburn (2002), "Voting procedures", in: K.J. Arrow, A.K. Sen, K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 4, this volume.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris); facsimile published in 1972 by
Chelsea Publishing Company, New York.
Farkas, D., and S. Nitzan (1979), "The Borda rule and Pareto stability", Econometrica 47:1305-1306.
Fine, B., and K. Fine (1974a), "Social choice and individual ranking I", Review of Economic Studies
41:302-322.
Fine, B., and K. Fine (1974b), "Social choice and individual ranking II", Review of Economic Studies
41:459-475.
Fishburn, P.C. (1973a), "Summation social choice functions", Econometrica 41:1183-1196.
Fishburn, P.C. (1973b), The Theory of social Choice (Princeton University Press, Princeton).
Fishburn, PC., and W.V Gehrlein (1976), "Borda's rule, positional voting, and Condorcet's simple
majority principle", Public Choice 28:79-88.
G'rdenfors, P. (1973), "Positionalist voting functions", Theory and Decision 4:1-24.

20 See Black (1958) p. 182.

394 PK. Pattanaik

Gehrlein, WV, and D. Lepelley (1999), "Condorcet efficiencies under the maximal culture condition",
Social Choice and Welfare 16:471-490.
Goodman, L.A., and H. Markowitz (1952), "Social welfare function based on individual rankings",
American Journal of Sociology 58:257-262.
Hansson, B. (1973), "The independence condition in the theory of social choice", Theory and Decision
4:25-49.
Hansson, B., and H. Sahlquist (1976), "A proof technique for social choice with variable electorate",
Theory and Decision 13:193 200.
May, K.O. (1952), "A set of independent necessary and sufficient conditions for simple majority decision",
Econometrica 20:680-684.
Merrill III, S. (1984), "A comparison of efficiency of mutialternative electoral system", American Journal
of Political Science 28:23-48.
Murakami, Y. (1968), Logic and Social Choice (Routledge & Kegan Paul, London).
Nitzan, S., and A. Rubinstein (1981), "A further characterization of Borda Ranking method", Public
Choice 36:153-158.
Saari, D.G. (1989), "A dictionary for voting paradoxes", Journal of Economic Theory 48:443-475.
Saari, D.G. (1990), "The Borda dictionary", Social Choice and Welfare 7:279-317.
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San Francisco).
Sen, A.K. (1977), "Social choice theory: a re-examination", Econometrica 45:53-89.
Smith, J. (1973), "Aggregation of preferences with variable electorate", Econometrica 41:1027-1041.
Vandercruyssen, D. (1999), "Analysis of voting procedures in one-seat elections: Condorcet efficiency
and Borda efficiency", Discussion Paper Series DPS 99. 11 (Centre for Economic Studies, Katholieke
Universiteit Leuven, Leuven).
Young, H.P. (1974). "An axiomatization of Borda's rule", Journal of Economic Theory 9:43-52.
Young, H.P. (1975), "Social choice scoring functions", SIAM Journal of Applied Mathematics 28:
824-838.
Young, H.P. (1986), "Optimal ranking and choice from pairwise comparisons", in: B. Groffman and
G. Owen, eds., Information Pooling and Group Decision Making (JAI Press, Greenwich).
Young, H.P. (1988), "Condorcet's theory of voting", American Political Science Review 82:1231-1244.
Young, H.P (1997), "Group choice and individual judgments", in: D.C. Mueller, ed., Perspectives on
Public Choice (Cambridge University Press, Cambridge).
Chapter 8

GAME-THEORETIC ANALYSIS OF VOTING IN COMMITTEES

BEZALEL PELEG
Institute of Mathematics and Centerfor Rationality and Interactive Decision Theory,
The Hebrew University of Jerusalem

Contents
Abstract 396
Keywords 396
Introduction 397
1. Plurality voting, Borda count, and feasible elimination procedures: some
examples 398
1.1. The voting paradox 398
1.2. The failure of plurality voting in the presence of manipulation of preferences by
coalitions 399
1.3. The failure of the Borda count 399
1.4. Feasible elimination procedures 400
2. Basic concepts 401
2.1. Committees 401
2.2. Choice problems 403
2.3. Voting procedures 404
3. Representations of committees 407
3.1. The simple games associated with an SCC 407
3.2. Cores of simple games 409
3.3. Representations of committees by SCCs 410
3.4. Representations of committees by SCFs 411
4. Undistorted and consistent SCFs 412
4.1. Voting games and Nash equilibria 412
4.2. Exactly and strongly consistent SCFs 414
5. Strong representations of committees 415
5.1. Strong representations and effectivity functions 415
5.2. Strong representations of weak and symmetric games 417
5.3. The capacity of a committee 418
6. Continuation and generalizations 419
7. Concluding remarks 421
References 422

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K. Sen and K. Suzumura
( 2002 Elsevier Science B. V All rights reserved
396 B. PDeleg

Abstract

In this chapter we adopt the axiomatic approach in order to find (new) voting
procedures to committees that are immune against deviations by coalitions of voters.
We shall now describe our approach. Let G be a committee and let A be a finite set
of m alternatives, m > 2. Our problem is to find a social choice function F that will
enable the members of G to choose one alternative out of A. We insist that F will have
the following properties.
(i) F should be Paretian, monotonic, and preserve the symmetries of G;
(ii) the power structure induced by F should coincide with G;
(iii) for each profile RN of (true) preferences of N (i.e., the set of members of G),
F(RN) should be the outcome of a strong Nash equilibrium (in the strategic game
specified by F and RN).
Let, again, G be a committee and let A be a set of m alternatives, m > 2. The pair
(G,A) is called a choice problem. A social choice function that satisfies the foregoing
three criteria (i)-(iii), is called a strong representation of (G,A). If G is weak, that is,
G has a vetoer, then (G, A) has a strong representation for every value of m. If G does
not contain a vetoer, then there exists a natural number yu(G) > 2 (the capacity of G),
such that (G,A) has a strong representation if and only if 2 < m < P(G). A family of
algorithms, called feasible elimination procedures, produces a strong representation to
any choice problem (G, A) whenever such a representation exists. Feasible eliminations
procedures produce all the strong representations of symmetric committees.

Keywords

committee, social choice function, sincere outcome, strong Nash equilibrium

JEL classification:D71
Ch. 8: Game-Theoretic Analysis of Voting in Committees 397

Introduction

In this chapter we adopt the axiomatic approach in order to find (new) voting
procedures for committees that are immune against deviations by coalitions (of voters).
Our research started in 1975 [the first results appeared in Peleg (1978a); see also Dutta
and Pattanaik (1978)]. At that time no coalitionally stable voting rule was known.
Indeed, up to date, there is only one family of procedures, besides our class, which
consists of coalitionally stable procedures, namely, the family of voting by veto rules
[see Mueller (1978) and, for a recent survey, Moulin (1994)]. Voting by veto procedures
are specified by game forms (and not by social choice functions, a more restricted class
of voting rules). Therefore, they do not fit in our framework which focuses on securing
the sincere outcome as an equilibrium outcome. (In a strategic game induced by a game
form there may be no choice of "sincere" strategies).
We shall now describe our approach. Let G be a committee and let A be a (finite)
set of m alternatives, m > 2. Our problem is to find a social choice function F that
will enable the members of G to choose one alternative out of A. We insist that F will
have the following properties.
(i) Classical properties, namely:
(1) Pareto optimality;
(2) symmetry, that is, F and G should have the same symmetries;
(3) monotonicity-F should be monotonic.
Our next step is to associate with every social choice function F a committee G(F)
(that is, a simple game), by using the standard effectivity notions of (cooperative) game
theory. We can now formulate our second criterion:
(ii) G = G(F), that is, G coincides with the power structure induced by F.
If m > 3 and F is Paretian, then, by the Gibbard-Satterthwaite Theorem, F is
manipulable, that is, sincere voting may not be an equilibrium. Thus, we only ask
whether the sincere outcome can be secured as an equilibrium outcome. Moreover, we
are interested in strong Nash equilibria, that is, equilibria which are immune against
deviations by coalitions of players. This brings us to our third (and last) requirement:
(iii) For each profile RN of (true) preferences of the members of G, F(RN) is
an outcome of a strong Nash equilibrium (in the strategic game specified by
F and RN). (Here N is the set of members of G.)
Let, again, G be a committee and let A be a set of m alternatives, m > 2. The
pair (G,A) is called a choice problem. A social choice function that satisfies the
foregoing three criteria (i)-(iii), is called a strong representation of (G,A). If G is
weak, that is, G has a vetoer, then (G,A) has a strong representation for every value
of m. If G does not contain veto players, then there exists a natural number 8i(G),
the capacity of G, such that (G,A) has a strong representation if and only if
2 < m < ut(G) [see Holzman (1986b)]. Peleg (1978a) introduced a class of algorithms,
based on elimination procedures, which produced strong representations to (fully)
symmetric committees. Those algorithms, called feasible elimination procedures, were
generalized in Holzman (1986a,b), to the class of committees without vetoers. They
398 B. Peleg

produce a strong representation whenever such a representation exists. For a symmetric

committee they yield all strong representations.
It is important to distinguish our approach from the usual approach to the analysis
of strategic aspects of voting rules. In the standard approach [see, e.g., Moulin (1994)]
a choice problem is specified by the set of voters and the set of alternatives. No
a priori distribution of power among the voters is given. Also, after the voting rule
is constructed, the associated simple game is not computed or analyzed. In contrast,
we start with a given simple game and insist that it will be the one associated with
the constructed voting rule. For the analysis of the strategic aspects of voting rules in
the more general context the reader is referred to Chapter 4 of this volume.
We now review briefly the contents of this chapter. Section 1 is devoted for exam-
ples. In Section 2 we discuss the basic properties of committees and voting procedures.
We proceed, in Section 3, to analyze the power structures which are induced by
a voting rule. This analysis leads naturally to the definition of representations of
committees by social choice functions (or social choice correspondences). The strategic
aspects of a voting situation are examined in Section 4. The final outcome of this
examination is the definition of exactly and strongly consistent social choice functions.
A strong representation of a committee is a representation by an exactly and strongly
consistent social choice function. The problem of existence of strong representations
(to committees) is solved in Section 5. In Section 6, we report some generalizations
of our theory. We conclude in Section 7.

1. Plurality voting, Borda count, and feasible elimination procedures: some

examples

We shall show by means of examples that current voting methods cannot cope with
manipulation of preferences by coalitions. A new method which is immune against
profitable deviations by coalitions will be introduced via two examples.

1.1. The voting paradox

Let N = 1, 2,3} be a set of voters, and let A = {a, b, c} be the set of alternatives.
A preference ordering of A is a complete, transitive, and antisymmetric binary relation
on A. The preference ordering of voter i N is denoted by R'. For x,y A we
write xR'y, if (x,y) R'. Let RN = (R, R2 ,R3 ) be a profile of preferences. x A
is the result of majority voting (with respect to RN), if {i C N I xR'y}I > 2 (see
Notation 2.1.3, p. 402) for all y A \ x}. Consider now the following profile:

R' R2 R3
a c b
b a c
c b a
Ch. 8: Game-TheoreticAnalysis of Voting in Committees 399

Then the majority rule is undefined for this profile: a beats b by majority, b beats c,
and c beats a.
Remark 1.1.1. After reading Subsection 3.2 the reader will be able to see that the
emptiness of the core C((3, 2),A, RN) is equivalent to the Voting Paradox.

1.2. The failure of plurality voting in the presence of manipulation of preferences by

coalitions

Let N = { 1,2, 3,4, 5} be the set of voters and let A = a, b, c} be the set of alternatives.
If R is a preference ordering, then t(R) is the top alternative of R (i.e., x = t(R)
if xRy for all y A). Let RN be a profile of preferences and let x A. Then
w(RN,x) = I{i E N I t(R i) =x}l, and D(RN) = x E A w(RN,x) > w(RN,y) for all
y E A}. The choice by plurality voting with respect to RN, M(RN), is the first alternative
(in the alphabetical ordering), in D(RN). The reader is referred to Chapter 4 in this
Volume for a discussion of plurality voting.
Now consider the following profile:

R1 R2 R3 R4 R5

a c b a c
b a c b a
c b a c b

Then M(RN) = a. First we notice that 3 can manipulate the outcome: if Q3 = (c, b, a),
then M(RN\{ 3 }, Q3 ) = c and cR3a. Hence, we may not expect sincere voting by
individuals. A fortiori, we may expect deviations by coalitions from the sincere profile.
Clearly, every three-person coalition S can enforce every alternative x by choosing an
S-profile QS such that t(Q') = x for all i E S. Thus, in our case S = {2, 3, 5} can
enforce c, and for every i S, cRia. But then S2 = {1, 3, 4} can change the result
to b (and bRic for all i E S2). Finally, S3 = {1,4,5} can secure a again and aR'b for
all i C S3. Thus, as soon as deviations by coalitions are possible, the Voting Paradox
reappears.
Remark 1.2.1. In this example the core C((5, 3),A,RN) = 0 (see Subsection 3.2).

1.3. The failure of the Borda count

Let N = {1,..., 8} and let A = {a,b,c}. For a preference ordering R let ti(R)
be the ith alternative in the order R (see Notation 3.1.10, p. 409). Further, let
w(R,x) = 3 - i if x = ti(R),x E A. If RN is a profile of preferences and x A,
then w(RN,x) = i= 1 w(Ri,x). The Borda choice with respect to RN, B(RN), is the
400 B. Peleg

first letter in D(Rv ) = {x E A w(R"N,x) > w(RN,y) for all y E A}. (See Chapter 4 in
this Volume for a discussion of the Borda count).
Now consider the following profile:

RI R2 R3 R4 Rs R6 R7 R8

c b a c b a c b
a c b a c b a c
b a c b a c b a

Then B(RN) = c. Notice that 3 can manipulate the outcome: if Q3 = (b, a, c), then
B(RN\{ 3}, Q3) = b and bR 3c. Hence, deviations of coalitions from sincere voting cannot
be excluded. We claim that every five-person coalition can enforce every alternative.
We shall only prove that S = {1,2, 3,4, 5} can enforce c. Let QS be given by

Q1 Q2 Q3 Q4 Q5
c c c c c
a b a b a
b a b a b

Then, for every profile pN\S of N \ S, B(Qs, PN\s) = c. This claim implies, as in the
previous subsection, that RN is a Voting Paradox profile (with respect to the Borda
count).
Remark 1.3.1. The core C((8, 5),A,R N) 0 (see Subsection 3.2).

1.4. Feasible eliminationprocedures

Let N = {1,2,3,4, 5} and A = {a, b, c} as in Subsection 1.2. Further, let RN be a
profile of preferences. A feasible elimination procedure (f.e.p.) with respect to R is
a sequence (xl, C 1;x 2, C2;x 3) such that A = {x,x 2,x 3} and
(1) C c {i E N t3(Ri) = xl} and ICl =2;
(2) C2 C {i E N \ C 1 I x3Rix 2} and C21 = 2.
Clearly, for every profile RN there exists an f.e.p. with respect to RN. Indeed, if
RN is a profile, then there exists xl E A such that {i c N t3 (R') = x}l > 2. Let
C1 C i N I t3(R') = x} satisfy Cei1 = 2. Then, we may apply the majority
rule to the restricted profile RN\CI I A \ {xl} in order to obtain x2 and C2. For
example, the profile in Subsection 1.2 has two f.e.p.'s: (c,{l,4}; b,{2,5}; a) a) and
(b,{2,5}; c,{1,4}; a).
Now let RN be a profile. x e A is RN-maximal if there exists an f.e.p.
(x,C 1; X2, C2; X3) such that x = X3 . Denote by X(RN) the set all RN-maximal
alternatives. The alternative chosen by fe.p.k, ((RN), is the first letter in X(RN).
Ch. 8: Game-Theoretic Analysis of Voting in Committees 401

Consider again the profile RN in Subsection 1.2. The following claim is true:
(*) No coalition S C N has a profitable deviation from RN.
Assume, on the contrary, that S C N has a profitable deviation QS. Then
x = p(QS, RN\s) # a and xR'a for all i C S. Ifx = b, then S n {2, 5} = 0. Therefore b is
eliminated by any f.e.p. with respect to (QS, RN\s). A similar contradiction is obtained
if qp(Q s , RN\s) = c.
The social choice function Tp has the following surprising property:
(**) If RN is a profile of preferences and x = qp(RN), then there exists a profile QN
such that (i) (p(QN) = x; and (ii) no coalition S C N has a profitable deviation from
QN.

Indeed, let (z, C 1; y, C2; x) be an f.e.p. with respect to RN. If QN satisfies

(i) t3 (Qi) = z for i C C 1; and (ii) t3(Qi ) = y, i c C2, then QN has the desired properties.
(The reader may repeat the proof of (*)).
Intuitively, in (**) RN is the sincere profile and QN represents a possible (stable)
strategic behavior.
If QN is given by:

Q1 Q2 Q3 Q4 Q5
b c b c b
c b c a c
a a a b a

then X(QN) = {b, c}, and Thus, X is not single-valued.

qp(QN) = b.
For the example in Subsection 1.3 we also may apply feasible elimination
procedures. The only difference is that the quota (of bottom places) for elimination of
an alternative should be enlarged to 3. For a general treatment of feasible elimination
procedures see Subsection 5.2 of this chapter.

2. Basic concepts

2.1. Committees

A committee or a simple game, is a pair G = (N, W), where N is a (finite) society and
W is a set of coalitions (non-empty subsets of N) satisfying

[S W and Sc T] = Tc W (2.1.1)

The elements of W are called winning coalitions. Equation (2.1.1) is the monotonicity
property of committees. An example of a committee is the Knessett (the Israeli
402 B. Peleg

parliament). It has 120 members and every set of 61 members or more is a winning
coalition.
We shall be interested in the following properties of simple games.
Definition 2.1.1. A simple game G = (N, W) is proper if for every S c N

S W N \S W

G is strong if G is proper and for every S C N

S W=N\SE W.

The Knessett is a proper committee which is not strong. In order to present the next
example we need the following definition and notation.
Definition 2.1.2. Two simple games (N, W) and (N*, W*) are isomorphic if there exists
a bijection Jr: N - N* such that for all S c N, S E W if and only if r(S) C W*.
We shall not distinguish in the sequel between isomorphic committees. Hence, we
shall be able to consider mostly societies of the form N = {1, ... , n}, n = 1,2, ...
Notation 2.1.3. If K is a finite set, then K I denotes the number of members of K.
Example 2.1.4. Let n and k, n > k, be positive integers. The simple game
(n,k) = (N, W) is defined as follows: N = {1, ... , n} and

W = {SI S c N and SI > k}.

(n, k) is proper if k > n/2. If 1is a natural number, then (21 + 1, I + 1) is strong. Clearly,
(21 + 1,1 + 1) is the (21 + l)-person simple majority committee. U
We now define further useful properties of committees.
Definition 2.1.5. A simple game G = (N, W) is weak if

V = n(S S W
W} 0.

The members of V are called veto players or vetoers.

The UN security council is a weak committee: the Big Five are vetoers (see
Remark 2.1.10 below). We remark that a weak simple game is proper.
Definition 2.1.6. Let G = (N, W) be a simple game. A permutation of N is a
symmetry of G if for every S C W, Jr(S) = {ft(i) i C S} E W. The group of all
symmetries of G is denoted by SYM(G). G is symmetric if SYM(G) = S,, the group
of all permutations of N.

Remark 2.1.7. A simple game G = (N, W) is symmetric if and only if there exists
1 < k < n, where n = INI, such that G is isomorphic to (n,k).
Ch. 8: Game-Theoretic Analysis of Voting in Committees 403

Notation 2.1.8. Let N = {1, ... , n} and let x = (x l , ... , x") be an n-tuple of real
numbers. For S C N we denote x(S) = Ei s x .
Definition 2.1.9. Let N = {1, ... , n} and let G = (N, W) be a simple game. G is a
weighted majority game if there exist a quota q > 0 and weights w 1 > 0, ... , w' > 0,
such that for every S C N

S E W >w(S) > q.

The (n + )-tuple [q; wt,..., w"] is called a representation of G, and we write

G= [q; w, ... , w].

Remark 2.1.10. The UN Security Council is given by [39; 7,7,7,7,7, 1, 1, 1, 1, 1, 1,

1, 1, 1, 1]. The current Knessett under party discipline (i.e., when all the members of
the same party should vote in the same way), is a weighted majority committee of
13 parties.
We also remark that a weighted majority game G = [q; w1, ... , w ] is proper if
2q > w(N). G is weak if and only if there exists i, 1 < i < n, such that w(N \ {i}) < q.
An (n, k) game is given by [k; 1, ... , 1].
Remark 2.1.11. The U.S.A. Senate is given by the (101,51) game where the extra
player is the Vice President. The House of Representatives is a 435-person simple
majority game. The Congress is a 536-person committee (C, Wc) where S c C is
winning if it contains a (simple) majority in both the Senate and the House. (We ignore
the special role played by two-thirds majorities). As the reader may easily check, the
Congress is not a weighted majority game.
We now recall the following definition.
Definition 2.1.12. A committee (N, W) is dictatorial if there exists a player d E N
such that

SE W d ES.

The player d is called a dictator.

We remark that a committee is dictatorial if and only if it is both weak and strong.

2.2. Choice problems

A choice problem is a pair (G,A) where G = (N, W) is a committee and A is a set of

alternatives with at least two members. In order to resolve a choice problem ((N, W), A)
the members of N have to choose one alternative out of A. Usually, one may observe
several stages in the process of selection of a member of A. First, the committee
members have to learn to know the various alternatives, and compare them. At
404 B. Peleg

the end of the first stage the voters are (tentatively) partitioned according to their
preferences on A. Now each voter may explain and defend his position; furthermore,
he may persuade other voters to adopt his position. A third stage may be devoted
to coordination of strategies and coalition formation. However, the final stage in our
model always consists of actual voting. Our analysis is confined to the final stage.
(The reader may argue that the analysis of strategic voting cannot be separated from
the analysis of coordination of strategies and coalition formation. This is indeed the
case, and the relationship between the two stages will be completely clarified in
Subsection 4.1 where we discuss voting games).
We shall now make the following assumption.
Assumption 2.2.1. In this chapter we restrict ourselves to choice problems with afinite
set of alternatives.
Choice problems whose set of alternatives is a subset of some Euclidean space
are extensively discussed in other chapters in Volume 2 of this Handbook (see, e.g.,
Chapters 16, 17 and 25). Indeed, the main impossibility theorems of Arrow and
Gibbard-Satterthwaite were extended to economic domains. Also, the theory of the
core of spatial voting games is quite developed. However, the main results which we
report in this chapter have not yet been generalized to Euclidean spaces.
We proceed with the following remark.
Remark 2.2.2. Let ((N, W),A) be a choice problem. In our model the formation of
preferences by the members of N is completed before the final stage of voting. We
assume that for every i E N the preference ordering of i, R i, is a complete, transitive,
and antisymmetry binary relation on A. (This is a standard assumption of rationality.
The extension of our results to weak orders, that is, complete and transitive binary
relations, is straightforward).
Let N = {1, .. ., n}, let ((N, W),A) be a choice problem, and let Rv = (R l, ... , R")
be a profile of preference orderings. The triple ((N, W), A, RN) is an input to any voting
procedure that may resolve ((N, W),A). We shall elaborate on this point in Section 3:
any method of choice for ((N, W),A) must distinguish between losing and winning
coalitions.

2.3. Voting procedures

LetA be a set of m alternatives, m > 2. 2' denotes the set of all non-empty subsets of A,
and L = L(A) denotes the set of all linear orders on A. Furthermore, let N = { 1, ... , n}
be a society.
Definition 2.3.1. A social choice correspondence (SCC) is a function H: L -- 2'4.

Example 2.3.2. Let RN E L N . x is Pareto-optimal(with respect to RN), if for every

y A there exists i C N such that xR i y. The set of all Pareto-optimal alternatives is
denoted PAR(RN). PAR(RN) • 0 because A is finite. Thus PAR(.) is an SCC. N
Ch. 8: Game-TheoreticAnalysis of Voting in Committees 405

The unanimity principle for SCCs may be formulated in the following way.
Definition 2.3.3. An SCC H is Paretian if H(RN) C PAR(RN) for all RN E LN.
Symmetries of an SCC are defined in the following way.
Definition 2.3.4. Let H be an SCC. A permutation z of N is a symmetry of H if
H(R 1 , ... , R) = H(R(' 1), ... , R J (" )) for all (R', ... , R" ) LN. The group of all
symmetries of H is denoted by SYM(H).
Definition 2.3.4 enables us to introduce the following important class of SCCs.
Definition 2.3.5. An SCC H is anonymous if SYM(H) = S,, the group of all
permutations of N.
We remark that PAR(.) is anonymous.
Neutrality of SCCs with respect to the alternatives is defined as follows. Let a be
a permutation of A and let R E L. We denote by o(R) the linear order defined by the
following condition: for all x,y E A, a(x)a(R)ac(y) if and only if xRy.
Definition 2.3.6. An SCC H is neutral if for every permutation (r of A, and
for every RN = (R 1, ... , Rn) e LN,H(Ur(R), ... , (RN)) = o(H(RN)). (Here
a((H(RN)) = {(x) x G H(RN)}.)
We remark that PAR(.) is neutral. We shall now discuss the important property of
monotonicity.
Definition 2.3.7. Let RN LN and let x A. RN E LN is obtainedfrom RN by an
improvement of the position of x if
(i) for all a, b E A \ {x} and all i E N, aR b X aRib;
(ii) for all a G A and i C N, xRia = xR'ia.

Definition 2.3.8. An SCC H is monotonic if it satisfies the following condition. If

RN C LN,x C H(RN), and RN is obtained from RN by an improvement of the position
of x, then x C H(RN) and H(R'[) C H(RN).
Monotonicity, although essential, is sometimes violated. For example, plurality
runoff is not monotonic.
The following stronger version of monotonicity is used frequently.
Definition 2.3.9. An SCC H has the strong positive association (SPA) property if it
satisfies the following:
Let RN E LN andx C H(RN). If RN E LN and for ally C A and i C N, xR'y implies
xR' y, then x C H(RN).
SPA implies monotonicity.

Lemma 2.3.10. If an SCC H has the SPA property, then it is monotonic.

A proof of Lemma 2.3.10 is contained in Section 2.3 of Peleg (1984).
406 B. Peleg

As the reader may easily verify, PAR(.) has the SPA property.
In many practical choice problems the choice of a single alternative is required.
Therefore, we now turn to the study of social choice functions.
Definition 2.3.11. A social choice function (SCF) is a function F: L"N A.
An SCF F naturally defines an SCC HF in the following way: HF(RN)= fF(R'V)}
for every Rv LN.
Definition 2.3.12. An SCF F is Paretian (respectively anonymous, neutral, monotonic,
has the SPA property), if HF is Paretian (respectively, anonymous, neutral, mono-
tonic, has the SPA property). Also, a permutation r of N is a symmetry of F if it is a
symmetry of HF. Thus, SYM(F), the symmetry group of F, is equal to SYM(HF).
An SCF F is a selection from an SCC H if F(RN) H(RN ) for every R'" E LN.
A selection from a Paretian SCC is Paretian. However, a selection from an anonymous
(respectively, monotonic) SCC may not be anonymous (respectively, monotonic).
Nevertheless, certain tie-breaking rules preserve anonymity and monotonicity.
N N
Definition 2.3.13. Let H: LN 2 A be an SCC and let R G L. For every R E L let
,

F(RN; H, R) be the R-maximum of H(RN). This SCF, F( ; H, R) is called the selection

from H according to the order R.

Remark 2.3.14. Let H be an SCC, R E L, and F = F(.; H,R). Then

SYM(H) SYM(F). In particular, if H is anonymous, then F also is anonymous.
Furthermore, if H is monotonic, then F is monotonic.
Let F be an SCF. The range of F is the set

R(F) = x I x = F(RN) for some RN e L}.

Definition 2.3.15. An SCF F: LN - A is dictatorialif there exists d E N (a dictator)

such that for all RN LN, F(RN)Rdx for all x e R(F).
The main result of this section is the following theorem.
Theorem 2.3.16. Let F.: L v A be an SCF IfF has the SPA property and JR(F)I > 3,
then F is dictatorial.
Theorem 2.3.16 is due to Muller and Satterthwaite (1977).
We now briefly discuss game forms.
Definition 2.3.17. A multi-valued game form is a system F = (N; S, ... , S"; ;A)
where
(i) N = {1, ... , n} is the set of players;
(ii) S is the (non-empty) set of strategies of player i C N;
(iii) : S x ... x S" 2 A is the outcome correspondence;
(iv) A is the set of alternatives.
Ch. 8: Game-Theoretic Analysis of Voting in Committees 407

Clearly, every SCC H defines a multi-valued game form FH = (N; L, ... , L; H,A).
Thus, a multi-valued game form may be considered as a generalized SCC. The
following example is illuminating.
Example 2.3.18. For each i E N let Si = 2A . That is, each voter may approve of a set
B, B C A, of alternatives. For s' E S i, i E N, and a · A, let

y(a) = I{i I i E N and a E si}.

Finally, define
)
yt(s', ... , s = {a E A I y(a) > y(b) for all b E A}.

Then (N; S l , ... , Sn; pt; A) is approval voting, a voting procedure suggested in Brams
and Fishburn (1983). ·
We conclude this section with the definition of game forms.
Definition 2.3.19. A game form (GF) is a system F = (N; S, ... , Sn; J(; A),
where N, S 1, ... , S", and A have already been determined in Definition 2.3.17, and
;r: Sl x ... x Sn - A is the outcome function.
Every SCF F defines a GF rF = (N; L, ... , L; F; A). Thus, GFs may be considered
as generalized SCFs.

3. Representations of committees

3.1. The simple games associated with an SCC

Let ((N, W),A) be a choice problem and let H: LN - 2 A be an SCC. We are
interested in a solution to the following problem: when is H compatible with the
committee (N, W)? In order to solve this problem we shall now associate several simple
games with H.
Let N = {1, ... , n} be a set of voters and let A be a set of m alternatives, m > 2.
Definition 3.1.1. Let H: LN -* 2A be an SCC and let x E A. A coalition S is winning
for x (with respect to H) if

[RN E LN and xRiy for all i E S and y E A] X H(RN) = {x}.

The set of all winning coalitions for x is denoted by W*(H,x).

Let H: LN 2 A be an SCC and let x E A. If S G W*(H,x), then S can enforce x
without knowing the preferences declared by N \ S.
Definition 3.1.2. Let H: LN - 2A be an SCC. A coalition S is winning (with respect
to H) if S E W*(H,x) for all x E A. Thefirst simple game associated with H is the
408 B. Peleg

game G* = (N, W*(H)), where W*(H) is the set of all winning coalitions with respect
to H.
Let H: LN 2 A be an SCC. G (H) is monotonic (i.e., it is a committee). Also,
G*(H) is proper because m > 2.
Definition 3.1.3. Let H: LN -, 2A be an SCC and let x E A. A coalition S is
a a-effective for x (with respect to H) if there exists R s C Ls such that for all
QN\S G LN\s, H(Rs, QN\S) = {x}. The set of all a-effective coalitions for x is denoted
by Wa(H,x).
Let H: LN -- 2 be an SCC and let x A. If S E Wa(H,x) then S can enforce x
without knowing the preferences declared by N \ S. Also, W*(H,x) C Wa(H,x).
Definition 3.1.4. Let H: LN - 2A be an SCC. A coalition S is a-winning (with respect
to H) if S E Wa(H,x) for all x E A. The second simple game associated with H is
the game Ga = (N, Wq(H)), where Wq(H) is the set of all a-winning coalitions with
respect to H.
Let H: LN . 2A be an SCC. Ga(H) is monotonic and proper.
Definition 3.1.5. Let H: LN -- 2 A be an SCC and let x c A. A coalition S is ft-effective
for x (with respect to H) if for every QN\S LN\S there exists Rs L s such that
H(Rs, QN\) = {x}. The set of all fi-effective coalitions for x is denoted by W/i(H, x).
Let H: LN 2 A be an SCC and let x A. If S W[3 (H,x) then S may not be
able to obtain x without knowing the preferences declared by N \ S. Thus, if a choice
of a profile RN G LN is made by a secret ballot, then -effectivity considerations
are inappropriate. However, in open voting processes, /3-effectivity arguments may be
valid. Clearly, Wa(H,x) c W(H,x).
Definition 3.1.6. Let H: LN -- 2A be an SCC. A coalition S is -winning (with respect
to H) if S E Wf3 (H,x) for all x c A. The third simple game associated with H is the
game Gf(H) = (N, Wry(H)), where W,(H) is the set of all -winning coalitions with
respect to H.
Let H: LN -- 2A be an SCC. G/3(H) is monotonic. However, it may not be proper
[see Example 3.1.13 in Peleg (1984)].
Remark 3.1.7. Let H: L' --* 2 A be an SCC. If H is Paretian, then N E W*(H). If
H is anonymous, then the games G*(H), Ga(H), and G/s(H) are symmetric. Finally,
if H is neutral, then for every x E A, W*(H) = W*(H,x), Wa(H) W(H,x), and
W(H) = Wfl(H,x).

Definition 3.1.8. An SCC H: LN - 2 is tight if W*(H,x) = Wp(H,x) for every

x E A.

Example 3.1.9. Let N = {1, ... , n} be a set of voters and let A be a set of
Ch. 8: Game-Theoretic Analysis of Voting in Committees 409

m alternatives, m > 2. Further, let M: LN 2 be the choice by plurality voting.

Then M is tight, and G*(M) = G(M) = GO(M) = (n, [] + 1). ·
Example 3.1.9 may be generalized. We start with a notation.
Notation 3.1.10. Let A be a set of m alternatives and let R L(A). Then we denote
by t(R) the ith alternative in the order R. Thus,

tI(R)Rt 2 (R) .. t- (R)Rtm(R).

Lemma 3.1.11. Let H: LN - 2A be a monotonic SCC. If H depends only on top

alternatives (i.e., there exists a function h: A - 2A such that

H(RN) H(R 1 , ... , R n ) = h(tl(R'), ... , tl(Rn))for all RN C LN),

then H is tight. [For a proofsee Section 3.1 in Peleg (1984)].

We conclude this section with the following important example.
Example 3.1.12. Let N = {1, ... , n} be a set of voters, and let A be a set of
m alternatives, m > 2. Further, let B = B(m, n) be the Borda rule. The three simple
games G*(B), Ga(B), and GO(B) are symmetric, because B is anonymous. It is possible
to prove that G* = (n, [n(m 1)]1),
+

Ga (n m = + and G (n,[2] + 1).

Example 3.1.12 is investigated in Gardner (1977), Moulin (1982) and Peleg (1984).

3.2. Cores of simple games

The core correspondence of a committee will play an important role in the sequel.
Hence, we make a small digression to study it. Let G = (N, W) be a committee and
let A be a set of m alternatives, m > 2. Furthermore, let RN E LN(= LN(A)).

Definition 3.2.1. Let x,y A, x y, and let S W; x dominates y (with respect

to RN) via S, written xDom(RN,S)y, if xRiy for i E S; x dominates y (with respect
to RN), written xDom(RN)y, if xDom(RN, T)y for some T E W. The core of the
choice problem ((N, W), A) with respect to RN is the set of undominated alternatives
in A; it is denoted by C(N, W,A,RN) = C(RN).
We observe that C(N, W,A,RN) 0 for all RN E LN, if and only if Dom(RN) is
acyclic for all RN E LN. Hence, we have the following result.
Remark 3.2.2. Let G = (N, W) be a weak simple game and let A be a finite set of
alternatives. Then C(N, W,A, RN) 0 for all RN C LN.
410 B. Peleg

We now proceed to investigate the non-emptiness of the core of simple games

without vetoers. First we need the following notation.
Notation 3.2.3. Let (N, W) = G be a simple game without vetoers. Nakamura r number
of G, v(G), is defined by
v(G)=minIol I ac Wand n{S S a}=0}.

The following result was proved in Nakamura (1979).

Theorem 3.2.4. Let G = (N, W) be a simple game without veto players and let A be
a finite set of alternatives. Then C(N, W,A,RN) X 0 for all RN LN(A), if and only
if JA < v(G).
Finally, we observe that if G = (N, W) is a committee without vetoers, then v(G) < n
(where n = INJ).

3.3. Representations of committees by SCCs

We shall now answer the question which was formulated at the beginning of
Section 3.1. Let G = (N, W) be a proper committee and let A be a set of m alternatives,
m > 2. An SCC H: LN - 2A is compatible with (N, W) if the power structure induced
by H coincides with G. Formally, we have the following definition.
Definition 3.3.1. An SCC H: Lv - 2A is a representation of G of order m if
G*(H) = G.
We are interested in the existence of "nice" representations. Thus, we need the
following definition.
Definition 3.3.2. Let H: L N - 2 A be a representation of G. H is a faithful
representation of G if SYM(H) = SYM(G).
The "nicest" representation of a committee is its core, provided that it is always
non-empty.
Theorem 3.3.3. Assume that C(N, W,A,RN) = C(RN) 0 for all RN C LN. Then
C(.) is a tight, faithful, and neutral representation of G which has the SPA property.
Furthermore, if G is non-null (i.e., N C W), then C(.) is also Paretian.
The reader should notice that the three simple games of the core correspondence
coincide: G*(C(.)) = G,(C(.)) = G 3(C(.)). The following result complements
Theorem 3.3.3.
Theorem 3.3.4. Assume that G is not weak and that m > v(G). f H. L'V - 2A is a
representation of G of order m, then H does not have the SPA property.
Theorems 3.3.3 and 3.3.4 are proved in Section 3.2 of Peleg (1984). A "nice"
representation for each choice problem is given, for instance, by the following
example.
Ch. 8: Game-TheoreticAnalysis of Voting in Committees 411

Example 3.3.5. For RN G LN and x c A denote

Dom-'(x, RN) = {y I y C A and yDom(RN)x}.

Let 0: LN , 2 A be defined by

O(RN) = {x E A I Dom-l(x, RN)| < Dom-I(y,RN) for ally E A}.

Then 0(.) is a tight, faithful, neutral, and monotonic representation of G of order m.
If N W then 0(.) is also Paretian. (See, again, Section 3.2 of Peleg (1984) for a
proof). 0

3.4. Representations of committees by SCFs

For the analysis of strategic voting in committees we need to study the relationship
between committees and single-valued voting procedures, that is, SCFs or GFs. For
reasons that will later become clear, we shall consider only SCFs.
Let G = (N, W) be a proper simple game and let A be a set of m alternatives,
m > 2. Further, let F: LN ,A be an SCF, and let HF be the SCC which is given by
HF(RN) = {F(RN)} for all RN e LN.
Definition 3.4.1. The three simple games associated with F are given by:
G*(F) = G*(HF), Ga(F) = G(HF), and GO(F) = Gfi(HF). Also F is tight if HF
is tight. Finally, F is a representation of G if G*(F) = G.
We now notice the following result. Let N be a society and let A be a set of
m alternatives, m > 2.
Lemma 3.4.2. Let F: LN -* A be an SCF If G*(F) is dictatorial,then F is dictatorial.
Conversely, if F is dictatorialand R(F) = A, then G*(F) is dictatorial.
A simple game G = (N, W) is non-null if N C W. Henceforth, we shall only consider
non-null games. The first impossibility result in this section is the following theorem.
Theorem 3.4.3. Let an SCF F: LN -, A be a representation of the proper simple
game G = (N, W). If m > 3 and F has the SPA property, then G is dictatorial.
Proof: N E W = W*(F) implies that R(F) = A. By Theorem 2.3.16 F is dictatorial
because AI = m > 3. Hence, by Lemma 3.4.2 G = G*(F) is dictatorial. D
Let G = (N, W) be a proper committee. An SCC H: LN -- 2A is a core-extension if
H(RN) = C(N, W,A,RN)(= C(RN)) if C(RN) 0. The SCC 0(.) of Example 3.3.5 is
a core-extension. Thus, the following Theorem proves the existence of representations
by SCFs.
Theorem 3.4.4. Let H: LN 2 A be a core-extension of the game G. If F = F(H,R)
,

is a selection from H according to the order R, then F is a tight representation of G.

412 B. Peleg

Furthermore, if H is faithful (respectively, monotonic, Paretian), then F is faithful

(respectively, monotonic, Paretian).
By Example 3.3.5 and Theorem 3.4.4, every proper committee has tight, Paretian,
faithful, and monotonic representations by SCFs of every order m > 2. For certain
proper committees it is impossible to add neutrality to the foregoing list of properties
of representations. Indeed, it is well known that for some pairs (m, n) (where m = 121
and n = IN[), there exists no anonymous and neutral SCF F: L N --+ A [see,
e.g., Moulin (1983, p. 25)]. We recall that faithful representations of symmetric
games are anonymous. Thus, for example, every game (n, k), where n/2 < k < n,
has no faithful and neutral representation of order m = tn, where t is a natural
number.
Section 3.3 of Peleg (1984) contains a detailed investigation of the existence of
"nice" representations by SCFs that are also neutral. That section also contains proofs
of Lemma 3.4.2 and Theorem 3.4.4.

4. Undistorted and consistent SCFs

4.1. Voting games and Nash equilibria

Let N = {, ... , n} be a set of voters and let A be a set of m alternatives, m > 2.

Further, let F: LN -- A be an SCF, and let RN be the profile of true preferences of the
voters. When the voters choose one alternative out of A according to the rule F they
actually play the following n-person game in strategic form.
Definition 4.1.1. The game associatedwith F and RN is the n-person game in strategic
form g(F,RN) = (L, ... , L; F; A; R', ... , RN), where
(i) L is the set of strategies for every i C N;
(ii) F is the outcome function;
(iii) A is the outcome space;
(iv) R' is the preference relation of voter i E N on the outcome space A.
We emphasize that, according to Definition 4.1.1, every R E L is a legal
strategy for each i E N. Also, there are no (exogenous) restrictions on preplay
communication between the voters. Hence, coordination of strategies is possible.
However, binding agreements between the players (i.e., agreements whose violation
entails high monetary penalties) are not allowed. Notice that (a) if the outcome
of g(F,RN) is determined by a secret ballot, then binding agreements are impos-
sible, and (b) binding agreements are, usually, not permitted in (political) voting
games.
The following notation will be used in the sequel.
Notation 4.1.2. Let T C S C N, T • 0, and let Rs L s. We denote by RT the
restriction of Rs to T.
Ch. 8: Game-Theoretic Analysis of Voting in Committees 413

Definition 4.1.3. QN LN is an equilibrium point (e.p.) of g(F,RN) if for every

i for all T L.
F(Q,)RiF(Ti,QN\ {i}) for all T i e L.

It is now possible to define manipulability of SCFs.

Definition 4.1.4. Let F: LN A be an SCF. F is nonmanipulable if for each
RN C LN, RN itself is an e.p. of g(F,RN).

The main result of this section is the Gibbard-Satterthwaite Theorem.

Theorem 4.1.5. Let F: LN - A be an SCE If F is nonmanipulable and R(F)I > 3,
then F is dictatorial.
Theorem 4.1.5 is due to Gibbard (1973) and Satterthwaite (1975). It implies that
(almost) all SCFs do not guarantee sincere voting. A natural question at this point is:
can we secure (at least) the sincere outcome? In order to do this we may insist, for
example, that the sincere outcome is the outcome of every e.p. The precise definition
is as follows.
Definition 4.1.6. An SCF F: LN A is undistortedif for each RN C LN the following
condition is satisfied

{a I a = F(QN) for some e.p. QN of g(F, RN)} = {F(RN)}.

Unfortunately, there are no interesting undistorted SCFs.

Theorem 4.1.7. Let F: LN A be an SCF 1fF is undistortedand IR(F) > 3, then
F is dictatorial.
Proof: By Maskin (1985), if F is undistorted, then F has the SPA property. Hence,
by Theorem 2.3.16, F is dictatorial. D
Thus, Definition 4.1.6 must be weakened.
Definition 4.1.8. An SCF F: LN -4 A is exactly consistent if for each RN E LN there
exists an e.p. QN of g(F, RN), such that F(QN) = F(RN).
Definition 4.1.8 leads to a general possibility result. Let G = (N, W) be a proper
committee, and let A be a set of m alternatives, m > 2. Then G has a tight, Paretian,
faithful, and monotonic representation of order m, which is also exactly consistent.
We shall only prove the foregoing claim for simple games without vetoers. (For weak
games we shall later prove a stronger result).
Lemma 4.1.9. Let G = (N, W) be a proper committee without veto players, and let A
be a set of m alternatives, m > 2. If F: LN - A is a representation of G, then F is
exactly consistent.
414 B. Peleg

Proof: Let R L and let a = F(RN). Choose QN L such that tl(Q') = a

for all i N. N \ {i} e W for every i N, because G has no vetoers. Hence,
N \ {i} C W*(F) for every i e N, because F is a representation of G. Thus, for every
N = for all T L.

F(Q ; \ (i, T) = F(Q) for all T' E L. []

Our existence result for exactly consistent representations, namely Lemma 4.1.9,
is unsatisfactory: the e.p. Q which was constructed in its proof is "uninteresting",
because it is not immune to deviations by coalitions of voters. Furthermore, under the
assumptions of Lemma 4.1.9, every alternative is an equilibrium outcome for every
profile of (true) preferences.
In order to avoid the dilemma of choosing between the impossibility result, namely
Theorem 4.1.7, and the uninteresting existence result, namely Lemma 4.1.9, we shall
strengthen our equilibrium concept to prevent (profitable) deviations by coalitions. This
will be done in the next section.

4.2. Exactly and strongly consistent SCFs

Let N = {1, ... , n} be a set of voters, and let A be a set of nmalternatives, m > 2.
Definition 4.2.1. Let F: L N -X.A be an SCF and let R'N e LN. e LN is a strong
QN
e.p. (s.e.p.) of g(F,RN) if for every coalition S and for every Ps E L s , there exists
i S such that F(QN) RiF(Ps, QN \S).
Let F: LN -t A be an SCF. The sincere outcome can always be secured by means
of s.e.p.'s if F is coalitionally undistorted. The precise definition is as follows.
Definition 4.2.2. F is coalitionally undistorted if for each R' L N the following
condition is satisfied

{a a = F(Q N) for some s.e.p. QN of g(F, RN)} = {F(RN)}.

Again, there are no interesting coalitionally undistorted SCFs: if F is coalitionally

undistorted and R(F) > 3, then F is dictatorial. The proof of this claim is the same
as the proof of Theorem 4.1.7. This leads us to the following definition.
Definition 4.2.3. Let F: LN -- A be an SCF. F is exactly and strongly consistent (ESC)
if for each RN E LN there exists an s.e.p. QN of g(F,RN) such that F(Q N) = F(RN).
We remark that an ESC SCF is exactly consistent. Clearly, the outcome of an
ESC SCF is not necessarily distorted. As far as we can see, Definition 4.2.3 provides
the strongest possible way of securing the sincere outcome for manipulable SCFs. We
chose to restrict ourselves to SCFs, and not to consider generalized voting procedures,
that is GFs, in order to be able to focus on the sincere outcome, and thus obtain
Definition 4.2.3.
Ch. 8: Game-Theoretic Analysis of Voting in Committees 415

We shall now clarify the relationship between implementation theory and Defini-
tion 4.2.3. We start with the following definition.
Definition 4.2.4. Let F = (N; S1, Sn; r; A) be a GF and let RN E LN. Yn C SN
... ,
is an s.e.p. of the (strategic) game g(F,RN) = (N; S l, ... , S"; Sr; A; RN) if for each
coalition B c N and every 6B C SB, there exists i c B such that r(yN)Riy(6B, yN \B).
We denote
SEO(F, RN) = {a e A I a = (7yN) for some s.e.p. 7N of g(,RN)}.

We proceed with the following definition.

Definition 4.2.5. Let F = (N; S 1, ... , S"; ~r; A) be a GF and let F: LN -->A be an
SCF. F (partially) implements F by strong equilibriaif for every RN LN
(F(RN) E SEO(F,RN)){F(RN)} = SEO(F,RN).

We now make the following remarks.

Remark 4.2.6. Let F be a GF and F an SCE If F implements F by s.e.p's and
IR(F) > 3, then F is dictatorial. (The proof of this claim is the same as the proof
of Theorem 4.1.7).
Remark 4.2.7. An SCF F is exactly and strongly consistent if and only if
it partially implements itself. Thus, as far as the strategic aspects of voting
procedures for committees are concerned, we rely on the property of (partial) self-
implementability (by s.e.p's).

5. Strong representations of committees

5.1. Strong representationsand effectivity functions
Let G = (N, W) be a proper and non-null simple game and let A be a set of
m alternatives, m > 2.
Definition 5.1.1. An SCF F: LN --+A is a strong representation of G of order m if
(i) F is a representation of G, that is, G*(F) = G;
(ii) F is ESC.
Strong representations are tight in a very strong sense. For the precise formulation we
need the following definitions.
Definition 5.1.2. Let F: LN - A be an SCF, B C A, B a 0, and S c N, S • 0. S is
winning for B if
[RN E LN,xRiy forx E B,y B, and i S] ~ F(RN) E B.
S is a-effective for B if there exists Rs Ls such that for every QN\S L N\S,
F(Rs , QN \ S) E B. S is fi-effective for B if for every QN \ s E LN \ s there exists Rs Ls
such that F(Rs, QN \S) E B.
416 B. Peleg

Notation 5.1.3. Let K be a set. We denote by P(K) the set of all subsets of K.
Definition 5.1.4. Let F: LN - A be an SCF. The first effectivity function associated
with F, E* E*(F): P(N) --+P(P(A)), is given by

E*(S) = {B I B C A, and S is winning for B},

if S 0, and E*(0) = 0. The a-effectivity function associated with F,Ea = E(F):

P(N) - P(P(A)), is given by

Es(S) = {B I B c A, and S is a-effective for B},

if S 0, and Ea(0) = . The /3-effectivity function associated with F,E 3 = E(F):

P(N) - P(P(A)), is given by

Ef3(S) = {B B C A, and S is fi-effective for B},

if S # 0, and EF(0) = 0.
Let F: LN - A be an SCE If S 0, S C N, then A E E*(S) n EA(S) n E/(S),
and 0 if E*(S) U E,(S) U Ef(S). Furthermore, if F is Paretian, then B E E*(N)n
E,(N) n E:(N) for all B c A, B X 0. The following result is important for our theory
of representations of committees.
Theorem 5.1.5. Let F. LN - A be an SCE If F is ESC, then F is strongly tight, that
is, E*(F) = E,(F) = E/3(F) [see Corollary 4.1.29 in Peleg (1984)].

Remark 5.1.6. Let F: L - A be an SCF and let S C N. Then E*(S) C E(S) C Efi(S)
Also, if F is strongly tight, then F is tight.
We shall now describe the connection between strong representations and cores of
effectivity functions.
Definition 5.1.7. Let E: P(N) -i P(P(A)), B c A, B 0, x A \ B, and RI C L
B dominates x via a coalition S, written BDom(RN , S)x, if (a) B G E(S); and (b) yR'x
for all y e B and i E S. B dominates x, written BDom(RN) x, if there exists a coalition T
such that BDom(RN, T) x. The core of E (with respect to RN ) is the set of undominated
alternatives in A, and it is denoted by C(E, RN).
Let F: LN -- A be an SCE Then, for every RN C L N
(a) C(E*(F),RN) D C(Ea(F),RN) D C(E,(F),RN);
(b) C(E*(F),RN) C C(N, W*(F),A,RN);
(c) C(Ea(F),RN) C C(N, Wa(F),A,RN);
(d) C(Ef(F),R ' ) C(N, Wfi(F),A,RN).
We are now ready for the following result.
Theorem 5.1.8. Let F: LN - A be an ESC SCE and let RN E LN. Then
F(RN) C C(Ef(F),RN).
Ch. 8: Game-Theoretic Analysis of Voting in Committees 417

Proof: Assume, on the contrary, that F(RN) is not a member of C(E,RN), where
El = EO(F). Denote x = F(RN). Then there exist B c A \ {x}, B 0, and S c N,
S • 0, such that B E ES(S) and yRix for all y E B and i E S. Let QN be an s.e.p.
of g(F,RN) such that F(QN) = x. By Definition 5.1.4, there exists pS E Ls such that
F(Ps, QN \ S) e B. Denote y = F(Ps, QN \ s). Then y X x and yRix for all i e S. Thus,
QN is not an s.e.p. of g(F, RN) and the desired contradiction has been obtained. E
Corollary 5.1.9. Let G = (N, W) be a proper (non-null) simple game and let the SCF
F: LN - A be a strong representation of G. Then, for every RN LN, F(RN) is in
C(N, W,A, RN) (in particular, C(N, W,A, RN) • 0).

5.2. Strong representationsof weak and symmetric games

The results on existence of strong representations for weak and symmetric simple
games can be easily formulated.
Theorem 5.2.1. Let G = (N, W) be a weak simple game. Then G has faithful and
monotonic strong representationsof every order greater than or equal to two.
We remark that if G = (N, W) is a simple game, N E W, and F is a strong
representation of G, then F is Paretian.
Theorem 5.2.2. Let G = (n, k) be a symmetric simple game, let < k < n, and let
b = n - k + 1. Then G has a faithful and monotonic strong representation of every
order m, where 2 < m < [(n + 1)/b].
Theorem 5.2.2 may be elaborated to yield a solution to the problem of mass elections
[see Peleg (1991)]. Proofs of Theorems 5.2.1 and 5.2.2 are contained in Peleg (1984).
Remark 5.2.3. Holzman (1986a) proved that [(n + 1)/b] is a sharp bound for the
order of a representation of a symmetric committee.
We shall now describe the set of all strong representations of a symmetric simple
game. Let (n, k), where " < k < n be a symmetric simple game, and let
2 < m < [(n + l)/b], where b = n-k + 1. Then, mb < n + 1. Let A be a set of
m alternatives. Choose natural numbers b(x), x C A, such that (i) Ex CA b(x) = n + 1;
and (ii) min EA b(x) = b. We proceed with the following notation and definition.
Notation 5.2.4. If R C L and B C A, then we denote by R I B the restriction of R
to B.
Definition 5.2.5. Let RN E LN. A feasible elimination procedure (f.e.p.), with respect
to RN, is a sequence (xi,,C 1; ... *; xi,,, , Cm-l; Xi,,), where A = {xi, . . ., xi,,} and
Ci C N, i = 1, . .., m - 1, such that the following conditions are satisfied:
(i) If 1 s <t <m-
l,then C n C = ;
(ii) ICjl =b(xj),j= 1, ... , m- l;
(iii) x is the worst alternative with respect to Rs {x ..., xi,, } for all s Cj and
forj= 1,..., m-1.
418 B. Peleg

Let RN C L N . If (xi,, C; *· ; xi, , C- ; xi,) is an f.e.p. with respect to RN, then xi,,/
is called RN-maximal. We denote
M(RN) = {x c A I x is RVN-maximal}.
(The proof that M(RN) 0 is straightforward). Now let F: L ' - A be an anonymous
SCF. For x C A let
k(F,x) = min{lS1 I S is winning for x},
and b(F,x) = n + 1 - k(F,x). The following theorem characterizes all faithful
representations of (n, k):
Theorem 5.2.6. Let F: L N -- A be anonymous. Then F is ESC, and b(F, x) = b(x)
for all x E A, if and only ifF(RN) C M(RV) for all RN E LN .
A proof of Theorem 5.2.6 is contained in Chapter 5 of Peleg (1984). It implies that all
strong representations of symmetric games are obtained by f.e.p.'s. Holzman (1986a)
generalized the foregoing theorem in the following way. Let G = (N, W) be a non-
weak proper committee. Then, if G has a strong representation of order m, than it also
has a strong representation of order m which is obtained by means of f.e.p.'s (with
respect to G).

5.3. The capacity of a committee

Let G = (N, W) be a proper, non-null simple game. A coalition B E P(N) is blocking if
N \ B C W. We denote by b the set of all blocking coalitions. (N E W implies 0 ~ b.)
B b is minimal blocking if B c B and B* B, imply that B* b. b denotes the
set of all minimal blocking coalitions.
Definition 5.3.1. A coalition S is meager if it is either non-blocking (S 9 b) or minimal
blocking (S e b,,).
For example, {i} is meager for every i c N, because 0 b.
Definition 5.3.2. Let p > r be positive integers. S P(N) is of type (p, r) if S is a
disjoint union ofp meager coalitions, at least r of which are non-blocking.

Definition 5.3.3. Let p be a positive integer. A coalition S is p-blocking if S is the

disjoint union ofp blocking coalitions.
Now we are ready for the following definition.
Definition 5.3.4. A committee G = (N, W) is singular if there is a 2-blocking coalition
of type (2, 1). Otherwise, G is regular.

Example 5.3.5. Let G = [7; 1, 1, 1,2,2, 2]. Clearly, S b if w(S) > 3. Let
S = {2,3,4,5}. Then, on the one hand S = {2,4} U {3,5} is 2-blocking. On the
other hand S = {2, 3} U {4, 5} is of type (2, 1). Hence, G is singular. U
Ch. 8: Game-TheoreticAnalysis of Voting in Committees 419

We remark that symmetric games are regular. For the analysis of a larger family
of regular games the reader is referred to Holzman (1986b). We proceed with the
following definition.
Definition 5.3.6. Let p > 2 be a natural number. A coalition S is p-critical if
(i) S is of type (p, 2); and
(ii) t N \ S > S U t} is p-blocking.
We remark that N is p-critical for p > n + 2 ((ii) is vacuous for N).
Example 5.3.7. Let G = [5; 1, ,1, 1, 1,1,2].
1, Then S = 1, 2, 3, 4, 5, 6} is of type (2, 2)
whereas N is 2-blocking. Thus, S is 2-critical. The reader may check that G is
regular. ·
Notation 5.3.8. Let G be a committee. Then

c(G) = min{p I there exists ap-critical coalition in G}.

The central concept of this section is given by the following definition.

Definition 5.3.9. Let G be a proper non-weak committee. The capacity of G, Pu(G),
is given by

y(G) = maxim I G has a strong representation of order m}.

By Peleg (1978b), Mu(G) > 2 and (G) < n. The precise value of (G) was
determined in Holzman (1986b).
Theorem 5.3.10. Let G be a proper non-weak committee. Then the capacity of G is
given by

2, if G is singular;
t(G) = r(G) - 1, if G is regular and N is of type ((G), 2);
K(G), if G is regular and N is not of type (c(G), 2).

Furthermore, G has a monotonic and faithful strong representation of order m for

every 2 < m < u(G).

6. Continuation and generalizations

Effectivity functions: Let N = {1, ... , n} be a set of players and let A be a set of
m alternatives, m > 2.
Definition 6.1. A correspondence E: P(N) P(P(A)) is an effectivity function if it
satisfies the following conditions: (i) E(N) = P(A) \ 0}; (ii) E(0) = 0; (iii) A G E(S)
for all S • 0; and (iv) 0 E(S) for all S c N.
420 . Peleg

Effectivity functions were introduced in Moulin and Peleg (1982). They were used
extensively in Moulin (1983), Peleg (1984) and lIolzman (1986a,b). Since 1982
numerous applications of effectivity functions to problems in game theory and social
choice have been found. Some of them are summarized in Abdou and Keiding (1991).
A recent application of the results on effectivity functions to social choice is Peleg
(1998).
Generalizations to topological spaces of alternatives: Kim and Roush (1981)
investigated the existence ESC SCFs where the set of alternatives is the real
line. BarberA and Peleg (1990) extended the Gibbard-Satterthwaite Theorem to the
(restricted) domain of continuous (utility) functions on a metric space. That paper was
followed by several extensions of the Gibbard-Satterthwaite Theorem to economic
domains (see Sprumont (1995) for a survey). However, our results on representations
of committees have not yet been generalized to economic domains. We remark that
the main results on cores of effectivity functions have been generalized to topological
spaces of alternatives.
Incomplete information: The theory of representations of committees has been
systematically generalized to committees with an information structure, that is,
committees where the preferences of the members are private information. The
reader is referred to d'Aspremont and Peleg (1988) for the details. The main result
of that paper is that all weak games have "nice" Bayesian incentive compatible
representations.
Choice problems (G,A) where yu(G) < Al: Let G = (N, W) be a proper (and non-
null) committee, and let A be a set of m alternatives. If m > p(G) then G has no strong
representation of order m. A solution to this problem is suggested in Peleg (1992). First,
we associate a simple game with each game form. [Effectivity functions of game forms
were already introduced in Moulin and Peleg (1982)]. Then we provide representations
of simple games by extensive game forms with perfect information, of any (finite)
order. This enables us to focus on subgame perfect equilibria of representations (instead
of Nash equilibria). However, the treatment in Peleg (1992) is incomplete: faithfulness
and monotonicity of representations (by extensive game forms) are not discussed.
Representations of committees by game forms on economic domains: Let A be a
convex and closed subset of a Euclidean space, and let G = (N, W) be a proper and
non-null committee. A game form F = (N; S1, ... , S; rT; A) is a strong representation
of G on A if: (i) the simple game associated with F coincides with G; (ii) F has a strong
Nash equilibrium for every profile of continuous and convex preferences on A. Keiding
and Peleg (2001) investigate the existence of strong representations of committees in
the foregoing model. The results depend both on the Nakamura number of G and the
geometric properties of A.
Representations of effectivity functions: Let N be a society and let A be the set of all
possible social states. Following Peleg (1998) we define a constitution as an effectivity
function E: P(N) -- P(P(A)). If S C N, S X 0, and B E E(S), then S has the right to
Ch. 8: Game-Theoretic Analysis of Voting in Committees 421

enforce the social state to be a member of B. A game form F = (N; S1, ... , S"; r; A) is
a representation of E if the effectivity function of r coincides with E. A representation
of E enables the members of N to exercise their rights simultaneously without violating
the constitution. A desirable property of a representation is strategic consistency, that is,
the existence of equilibrium for each profile of preferences. Keiding and Peleg (2002)
investigate the existence of coalition-proof consistent representations of effectivity
functions, whereas Peleg, Peters and Storcken (2002) characterize Nash consistent
representations.

7. Concluding remarks

The Gibbard-Satterthwaite Theorem has stimulated several lines of research of voting

systems. We shall list some of them. One possibility to avoid the Gibbard-Satterthwaite
Theorem is to restrict the set of preferences which can be reported by the voters.
Chapter 24 in Volume 2 of this Handbook deals with strategy-proof voting schemes
on restricted domains of preferences. The paper "voting by committees" [Barber,
Sonnenschein and Zhou (1991)], which falls in the foregoing category, is somewhat
related to this chapter.
A strategy-proof voting scheme possesses an equilibrium in dominant strategies
for every profile of preferences of the voters. One way to relax this property is to
require the existence of a unique strategy (up to payoff equivalence), which survives
iterated elimination of (weakly) dominated strategies. This approach was initiated
by Farquharson (1969), generalized by Moulin (1979), and followed by many other
investigators (see Moulin (1994) for a detailed survey). The usage of survivors of
iterated elimination of (weakly) dominated strategies is called sophisticated voting.
Voting schemes which allow sophisticated voting are called dominance solvable.
A remarkable family of dominance solvable voting schemes is voting by veto schemes
[Mueller (1978)]. For a survey of the results on voting by veto see Moulin (1994).
Analysis of cooperative solutions of voting schemes has also appeared in print
(besides the analysis of strong representations of committees). For example, Moulin
and Peleg (1982) contains a characterization of strong equilibrium outcomes of voting
by veto (see Moulin (1994) for a survey).
We now briefly recall our approach. Let A be a set of m alternatives, m > 2, and let
G = (N, W) be a (non-null) proper simple game. The pair (G, A) describes a typical
choice problem. A solution to the choice problem (G,A) is an SCF F: LN -3 A with
the following properties
(i) F is Paretian, monotonic, and preserves the symmetries of G;
(ii) G*(F) = G (the power distribution induced by F is G);
(iii) F is exactly and strongly consistent (for every profile of preferences the sincere
outcome is an outcome of a strong Nash equilibrium).
Our survey describes a complete solution to the foregoing problem of existence of
strong representations of committees.
422 B. eleg

References

Abdou, J., and H. Keiding (1991), Effectivity Functions in Social Choice (Kluwer Academic Publishers,
Dordrecht, The Netherlands).
Barberit, S., and B. Peleg (1990), "Strategy-proof voting schemes with continuous preferences", Social
Choice and Welfare 7:31-38.
Barbera, S., H. Sonnenschein and L. Zhou (1991), "Voting by committees", Econometrica 59:595 609.
Brams, S.J., and P.C. Fishburn (1983), Approval Voting (Birkhiuser, Boston).
d'Aspremont, C., and B. Peleg (1988), "Ordinal Bayesian incentive compatible representations of
committees", Social Choice and Welfare 5:261-279.
Dutta, B., and PK. Pattanaik (1978), "On nicely consistent voting systems", Econometrica 46:163-170.
Farquharson, R. (1969), Theory of Voting (Yale University Press).
Gardner, R. (1977), "The Borda game", Public Choice 30:43-50.
Gibbard, A.E (1973), "Manipulation of voting schemes: a general result", Econometrica 41:587-601.
Holzman, R. (1986a), "On strong representations of games by social choice functions", Journal of
Mathematical Economics 15:39-57.
Holzman, R. (1986b), "The capacity of a committee", Mathematical Social Sciences 12:139-157.
Keiding, H., and B. Peleg (2001), "Stable voting procedures for committees in economic environments,"
Journal of Mathematical Economics 36:117-140.
Keiding, H., and B. Peleg (2002), "Representation of effectivity functions in coalition proof Nash
equilibrium: A complete characterization", Social Choice and Welfare, forthcoming.
Kim, K.H., and EW Roush (1981), "Properties of consistent voting systems", International Journal of
Game Theory 10:45-52.
Maskin, E. (1985), "The theory of implementation in Nash equilibrium: a survey," in: L. Hurwicz,
D. Schmeidler and H. Sonnenschein eds., Social Goals and Social Organization: Essays in Memory
of Elisha Pazner (Cambridge University Press) pp. 173 204.
Moulin, H. (1979), "Dominance-solvable voting schemes", Econometrica 47:1337-1351.
Moulin, H. (1982), "Voting with proportional veto power", Econometrica 50:145-162.
Moulin, H. (1983), The Strategy of Social Choice (North-Holland, Amsterdam).
Moulin, H. (1994), "Social choice," in: R.J. Aumann and S. Hart, eds., Handbook of Game Theory,
Vol. 2 (North-Holland, Amsterdam) pp. 1091-1125.
Moulin, H., and B. Peleg (1982), "Cores of effectivity functions and implementation theory", Journal
of Mathematical Economics 10:115-145.
Mueller, D. (1978), "Voting by veto", Journal of Public Economics 10:57-75.
Muller, E., and M.A. Satterthwaite (1977), "The equivalence of strong positive association and strategy-
proofness", Journal of Economic Theory 14:412-418.
Nakamura, K. (1979), "The vetoers in a simple game with ordinal preferences", International Journal
of Game Theory 8:55-61.
Peleg, B. (1978a), "Consistent voting systems", Econometrica 46:153-161.
Peleg, B. (1978b), "Representations of simple games by social choice functions", International Journal
of Game Theory 7:81-94.
Peleg, B. (1984), Game Theoretic Analysis of Voting in Committees (Cambridge University Press).
Peleg, B. (1991), "A solution to the problem of mass elections," in: K.J. Arrow, ed., Issues in Contemporary
Economics, Vol. 1, Markets and Welfare (Macmillan, London) pp. 287-294.
Peleg, B. (1992), "On perfectly coalition-proof Nash equilibria", in: M. Majumdar, ed., Equilibrium and
Dynamics: Essays in Honor of David Gale (Macmillan, London) pp. 259-268.
Peleg, B. (1998), "Effectivity functions, game forms, games, and rights", Social Choice and Welfare
15:67-80.
Peleg, B., H. Peters and T. Storcken (2002), "Nash consistent representations of constitutions: a reaction
to the Gibbard Paradox", Mathematical Social Sciences 43:267-287.
Ch. 8: Game-Theoretic Analysis of Voting in Committees 423

Satterthwaite, M.A. (1975), "Strategy-proofness and Arrow's conditions: existence and correspondence
theorems for voting procedures and social welfare functions", Journal of Economic Theory 10:
187 217.
Sprumont, Y. (1995), "Strategy-proof collective choice in economic and political environments", Canadian
Journal of Economics 28:68-107.
Chapter 9

REPRESENTATIVE DEMOCRACY AS SOCIAL CHOICE*

NORMAN SCHOFIELD
Center in Political Economy, Washington University in St. Louis

Contents
Abstract 426
Keywords 426
1. Introduction: Constitutional choices 427
2. Voters as a "committee" 432
3. A "committee" of politicians 436
4. Elections as methods of belief aggregation 443
5. Electoral risk-taking and economic or political quandaries 448
6. Concluding remarks 452
References 452

* The empirical analyses mentioned in this chapter are based on co-authored work, with Andrew Martin,
Gary Miller, David Nixon, Robert Parks, Kevin Quinn, Itai Sened, and Andrew Whitford. This work was
supported by NSF Grant SBR-98-18582. Amartya Sen, Jon Elster and Arthur Lupia kindly made some
helpful remarks on an earlier version of the paper. Thanks are due to Tsvetan Tsvetkov for drawing
Figures 1 and 7, and to Lida de Maaijer-Hoek for her help with the final preparation of the chapter.
Figures 2, 5 and 6 are based on the research of Andrew Martin, Kevin Quinn and Andrew Whitford. A
version of this chapter was presented at the European Public Choice Meeting, Siena, Italy, May, 2000.

Handbook of Social Choice and Welfare, Volume 1, Edited by KJ Arrow, A.K Sen and K. Suzumura
© 2002 Elsevier Science B. V All rights reserved
426 N. Schofield

Abstract

Social Choice traditionally employs the preferences of voters or agents as primitives.

However, in most situations of constitutional decision-making the beliefs of the
members of the electorate determine their secondary preferences or choices. Key
choices in US political history, such as the ratification of the Constitution in 1787 and
the election of Lincoln in 1860, were conditioned by changing beliefs as regards the
truth of propositions about the political universe. Preference-based models of election
tend to conclude that candidates, or parties, converge to a vote-maximizing policy
position at the "electoral center". Empirical work suggests that such a conclusion is
invalid. This chapter argues, on the contrary, that parties or candidates adopt positions
that optimize, in a Nash equilibrium sense, with respect to both their beliefs over
electoral response, and their beliefs over appropriate policy choices. The analysis
indicates that political choices will be different depending on whether plurality ("first
past the post") or proportionality is used as the method of electoral representation.

Keywords

belief and preference aggregation, constitutional quandary, plurality and proportional

electoral systems, political equilibrium

JEL classification: C71, C78, D72

Ch. 9: Representative Democracy as Social Choice 427

1. Introduction: Constitutional choices

To inquire into the best form of government in the abstract ... is not a chimerical, but a highly
practical employment of scientific intellect. .S. Mill (1861)

The deepest issue of social choice concerns the nature of the constitutional "bargain"
made within a society. The arguments of Hobbes suggest the necessity of a
Constitution, the Leviathan, to mitigate the consequences of anarchy or "Warre",
within which "the life of man [is] solitary poor, nasty, brutish and short". [Hobbes
(1651) Ch. 18, p. 234]. Hobbes may well have constructed his argument in response to
the chaos of the civil war in Britain. But the people of Britain rejected the strong form
of Leviathan in the person of Oliver Cromwell, Lord Protector. On Cromwell's death,
Charles Stuart was welcomed back as monarch. Even then the nature of the contract
between the people and the monarch was deeply contested. In his Two Treatises of
Government arguing against Filmer's Patriarcha(1652), John Locke developed further
the notion of the contract. "And thus every Man, by consenting with others to make
one Body Politik, under one Government, puts himself under an Obligation to every
one of that Society, to submit to the determination of the majority ... " [Locke (1690)
pp. 197-198].
The institutional innovations put in place in Britain after the "Glorious Revolution"
and the assumption of the throne by William and Mary in 1688 led to the creation
of a strong fiscal state that better equipped Britain to compete with France during
the entire period until 1815 [Brewer (1988), North and Weingast (1989)]. France, in
contrast, was unable to complete similar fiscal innovations. The distinguished French
theorist and financier, Turgot, attempted to redesign the French fiscal apparatus prior
to the American War of Revolution. However, aid provided by Louis XVI of France, in
alliance with his minister of state, Vergennes, to the American colonists was enough to
bankrupt France and force the calling of the Estates-General in 1789 [Norberg (1994)].
In the charged milieu thus created, the great social mathematician, Condorcet, hoped
to create a constitutional monarchy, based on a Declaration of the Rights of Man. The
faction supporting such a constitution, the Girondins, were expelled in the Jacobin
Terror of 1794. As Schama (1989, p. 856) notes: "The great exponent of a state in
which science and virtue would be mutually reinforcing, the Marquis de Condorcet,
died in abject defeat".
It is well known that Condorcet in his Essai of 1785 [McLean and Hewitt (1994)]
laid the foundations of social choice theory. His famous "Jury Theorem" 1 [Rae
(1969), Schofield (1972)] has recently excited considerable interest [Ladha (1992,
1993, 1995), Ladha and Miller (1996), Austen-Smith and Banks (1996), Feddersen and

1 The theorem asserts that, in a situation of risky social choice between two alternatives, if the average
probability, p, that a juror chooses the true option exceeds one half, then the jury, using majority rule,
will choose the true option with probability greater than p. Moreover, this latter probability will approach
unity as the size of the jury increases.
428 It Schofield

Pesendorfer (1998), Duggan and Martinelli (1999)]. Even better known is the notion of
a Condorcetian "voting cycle". This work by Condorcet was brought to the attention of
political scientists and economists by Duncan Black (1958): see also the Introduction
in Black's Collected Works in 1998. The most famous result in this tradition of formal
social choice theory is, of course, due to Kenneth Arrow (1951).
The two distinct ideas of belief aggregation associated with the Jury Theorem, and
of preference aggregation, as utilized in social choice theory, seem to generate entirely
different interpretations of the functioning of a democracy.
I have suggested elsewhere [Schofield (2002a)] that Condorcet's result on the Jury
Theorem influenced Madison's arguments in his essay on "The Vices of the Political
System of the United States" of 1786, and in his Federalist X of 1787 [Madison
(1999, pp. 69-80 and pp. 160-167)]. Madison argued in these two essays that well-
intentioned representatives could act as a "jury" in determining superior choices for
the republic. [See Beer (1993) for discussion]. To better understand the process of
the ratification of the American Constitution in 1787-88, it is also important to make
sense of the decision by the colonial representatives in the Continental Congress to
declare independence from Britain in 1776.
My view of these two crucial decisions in 1776 and 1787 was that they initially
involved deep constitutional quandaries. In the early 1770s it was unclear to the
colonists whether the British did indeed threaten tyranny. For their part the British
cabinet was concerned with the effects on the Indian population of settler intrusion
into the Ohio Valley. After the seven-years war of 1756-1763, a rebellion by a
coalition of Indian tribes under Pontiac took two years to settle. The British tried
first of all by an Act of Proclamation to close the Ohio Valley to settlement. When
this proved ineffectual, the Quebec Act of 1774 transferred control of the whole
trans-Appalachian region to Quebec. By this act, colonial expansion to the West
was to be constrained. Although this provocation may have inflamed the colonial
elite, the obvious military capability of Britain initially deterred a declaration of
independence. However, the secret promise of aid to the colonists by Louis XVI
changed the decision problem facing the Continental Congress from one of uncertainty
to one of risk. While the members of the Congress may be regarded as risk-averse,
in the dilemma presented between independence and acquiescence, they were willing
to choose the risky option of war. Of course, not all of the colonial population
agreed with this assessment. Nonetheless, this risky choice led to the successful
outcome of independence. From the point of view of the American elite in 1786-87,
the historical fact that the choice in 1776 proved to be "correct" would have
given some empirical basis for believing in the ability of a representative body to
choose wisely.
As Riker (1964) suggested many years ago, the weak form of confederation adopted
by the American states after the Revolutionary War had proved to be ineffectual in
dealing with the significant threat already posed by Spain in the Mississippi Valley
in the early 1780s. Although a stronger federation was the obvious way to deal
with this threat, such a constitutional transformation created an intellectual quandary.
Ch. 9: Representative Democracy as Social Choice 429

A fundamental belief at this time was that democratic choice was beset by the problem
of factional turbulence. Adam Smith (1776) had, for example, alluded to such a belief
in the closing pages of Wealth of Nations.
Madison's argument in Federalist X dealt directly with this quandary. While
recognizing the possibility of such turbulence, Madison argued that representatives of a
heterogeneous republic could indeed make choices for the public good. If we interpret
Madison's argument in terms of the Jury Theorem, then it only seems to be valid,
however, if the choices to be made are "binary" (left-right or yes-no, for example).
Although it was not made explicit by Madison, an institutional apparatus that creates
"binary" political competition may be a necessary condition for the validity of the
argument. It has indeed been noted by many scholars [Schattschneider (1942), Riker
(1953)] that politics in the USA nearly always collapses to two-party competition. As
Riker observed, however, plurality rule 2 may be necessary for two-party concentration,
but does not appear to be sufficient (consider the Liberal Democratic Party in present-
day Britain). Indeed, probably the most crucial presidential election in US history
took place in 1860, but involved four candidates. Lincoln (the Republican candidate)
competed against Douglas (the "northern" Democrat), Bell (the "Whig"), and
Breckinridge (the "southern" Democrat). Although Lincoln gained approximately 60%
of the northern vote, he took only 40% of the total popular vote. Nonetheless, because
of the plurality nature of the electoral college, Lincoln gained 180 electoral-college
votes out of 303. The lead-up to this election also involved the deep quandary over the
resolution of the compromise over slavery. In a sense this quandary was transformed
into a dilemma for the northern electorate, as a result of the Dred Scott Decision by
the Supreme Court in 1857. I have argued [Schofield (2002b)] that this decision was
interpreted by Lincoln as a threat by the southern states to extend slavery to the North.
Acquiescing to this potential threat could result in a great expected cost to northern
voters. On the other hand, choosing Lincoln increased the probability of war, also
inducing a terrible expected cost.
In a sense the institutional feature of plurality rule forced a choice, one way or
another. As a thought experiment, imagine that the election of 1860 took place under
proportional representation 3 of some kind. Then it is plausible that the result would
have been a "coalition government" of northern and southern Democrats, and Whigs.
Under a "parliamentary government" of this kind, it is unlikely that war would have
occurred. Instead, some kind of compromise would have been made, and the issue of
slavery resolved in a different fashion.
This thought experiment is offered as a way of emphasizing different outcomes from
different electoral laws. Arguments over the superiority of one form of representation

2 Plurality rule means simply that the party or candidate with more votes than any other wins the
political prize. Presidential elections in the USA are based on plurality rule in the electoral college.
3 Proportional representation (PR) means that representation of parties or candidates is approximately
proportional to vote shares.
430 N Schofieid

over another date back, certainly, to the nineteenth century. John Stuart Mill (1861) for
example, suggested that a "proportional" electoral rule led to fairer outcomes, because
all interests would be represented. In contrast Lawrence Lowell (1896) argued that the
legislature must contain "two parties only, in order that the parliamentary form of
government should permanently produce good results". The arguments have persisted
to the present time. Riker (1953), for example, put forward the case that the use
of proportional representation tends to lead to increasing political fragmentation. In
the Weimar Republic of Germany, the number of significant parties in the Reichstag
increased from seven in 1919 to sixteen in 1930 [Mellen (1943)]. It is not implausible
that this paved the way for Hitler. Karl Popper (1945) also wrote of the dangers of
proportional representation, and much later [Popper (1988)] agreed with Duverger
(1984) that plurality rule had the advantage of offering a clear choice to the electorate.
[See Schofield (2001) for a review of Riker's work].
One can, of course, contrast plurality and proportional electoral rules by a "macro-
political" examination of any different effects that can be observed. A second, possibly
more illuminating, procedure is to construct theoretical models designed to distinguish
between different electoral rules. Unfortunately, most formal social choice models
focus on two-party competition. The results in this research program are mixed. The so-
called "chaos theorems" in spatial voting theory suggest that anything can happen. [See
Riker (1980) for such an argument]. On the other hand a large body of literature on the
spatial model of elections [Calvert (1985), Enelow and Hinich (1984), for example]
suggests that candidates, under two-party competition, converge to an electoral center.
The same result obtains moreover, under proportional representation, when candidates
attempt to maximize "expected vote share" [Lin et al. (1999)]. Electoral models in
this Downsian tradition [Downs (1957)] cannot therefore distinguish between the
consequences of plurality and proportional rule. However, it is not at all obvious that
candidates, in the "multi-party" situation characterized by proportional representation,
will necessarily adopt policies simply to maximize votes. it seems much more likely
that candidates come to the political arena with well-specified beliefs that they wish
to implement.
This brings us to a more general theoretical point that is worth emphasizing. Almost
all models of elections assume that voters are characterized simply by their political
preferences, and that candidates desire to "win" (by maximizing votes perhaps).
However, it is worth distinguishing between the fundamental preferences of a voter,
and "secondary" preferences of the voter over acts or choices.
The connecting relationship between the two is created by the set of beliefs that the
voter, or agent, has over the state of the world. These beliefs are subjective probability
estimates concerning the truth values of various propositions [Savage (1954)]. Most
social choice theory focuses on situations where the preferences are fixed. Clearly,
however, new information about the state of the world will change beliefs, and thus
secondary preferences. If Bayes' Law is utilized, as the procedure by which beliefs
are transformed, then one might expect beliefs and secondary preferences to change
continuously, even slowly. The instances, mentioned above, of the transformation of
Ch. 9: Representative Democracy as Social Choice 431

constitutional quandaries into choices, suggest, on the contrary, that beliefs can change
dramatically under certain circumstances.
Implicit in the concluding section of Keynes' General Theory of Employment
[Keynes (1936)] is the assertion that fundamental constitutional decisions must often
be made in an uncertain environment. As Keynes observed:
By "uncertain" knowledge, let me explain, I do not mean merely to distinguish what is known
for certain from what is only probable .... Even the weather is only moderately uncertain. The
sense in which I am using the term is that in which the prospect of a European war is uncertain,
or the price of copper and the rate of interest twenty years hence. Keynes (1937)

[See Schofield (1999a) for an elaboration on the relevance of uncertainty in Keynes'

thought].
Uncertainty is more than just some variance in subjective probability estimates.
In situations that I have described as quandaries, the costs of incorrect choices are
high, while individuals will be unwilling to derive secondary preferences from their
uncertain estimates of probabilities. How, then, can social choice be made? One
feature that appears to characterise the constitutional choices described above is that a
compelling argument was presented which had the effect of transforming uncertainty
into risk. For example, in the Continental Congress in 1776, uncertain concerns over
the intentions of the British and French were changed into coherent beliefs. Although
the expected costs of acquiescence or rebellion were both high, the decision became
one of risk, not uncertainty. It is reasonable to identify the consequent social decision as
a dilemma. Although not all members of the Congress would necessarily have precisely
the same probability estimates, it is clear that a majority chose rebellion. This outcome,
based on secondary preferences, can be interpreted as a core (or unbeaten outcome
under the institutional rule adopted). By analogy, the set of beliefs that underlay this
decision can be described as a core belief.
In the same way, the set of beliefs over foreign threat, and over the possible
resolution of the dangers of factionalism (held by the members of the Constitutional
Convention in Philadelphia in 1787) led to a core belief in the wisdom of ratification.
Finally, the election of 1860 can be seen as the culmination of a process leading
from electoral uncertainty, to a more precise appreciation of the dilemma (involving
anticipated costs and probabilities) to a final choice. It is important to note that the
set of beliefs, held in the northern electorate, was transformed, by the institutional
mechanism of the electoral college, into a core belief that sustained the choice of
Lincoln. [See Schofield (2002b) for further discussion].
As I have suggested above, when beliefs of the voters are the most significant
primitives to be studied, then decision-making may be described in terms of a jury.
In contrast, when the beliefs of the agents are held fixed, and preferences of the
agents are taken as the primitives, then the object of study will be termed committee
decision-making. In the body of the chapter that follows I shall review various
models of elections that are derived from the theory of committee decision-making.
Section 2 surveys the traditional view of an election as a committee choice. Typically,
432 V. Schofield

this literature assumes that there are only two parties, or candidates, who attempt
to adopt positions in order to win. When there are more than two parties, then
coalition formation is necessary for the creation of government. Bargaining between
parties (after the election) can also be interpreted in terms of committee decision-
making [in Chapter 8 of this Volume, Peleg (2002) presents a somewhat different
perspective on this process].
However, the choices made by political parties over leaders and policy positions are
made prior to elections. Section 3 presents a formal perspective on the existence of
equilibria in such a "political game". Such equilibria depend on the beliefs held by
the party principals over the nature of the electoral response to party declarations.
One way to model an election is to describe each voter by a stochastic choice
function, which (given party positions) assigns a probability vector to the voter. The
probability that voter i chooses party j can be interpreted as an estimate of the belief
by i that party j is the superior choice (given the recent history of party actions).
Econometric analyses can then be used to estimate the relationship between the vector
of party positions and the expected vote shares of the various parties. A number of
examples are offered which strongly indicate that parties do not adopt policy positions
simply to maximize expected vote. Indeed, the results of Section 4 suggest that the
elite members of parties choose their political leaders subject to complex beliefs over
electoral response and preferable policy choices. In multiparty systems it is clear that
the logic of such choices can lead to high political fragmentation.
The theoretical and empirical analyses presented in these three sections are not
conclusive but they do suggest that plurality and proportional electoral systems
generate quite different political choices. Because of the need to bargain over coalition
government, proportional electoral systems may support political choices that are, in
a sense, "risk-averse". The concluding section of this chapter suggests that plurality
electoral methods may endanger risky choices by some of the parties. It is possible
that these "endogeneous" risk postures in different political systems can be connected
to the quite different present-day economic characteristics of the United States and
Britain on the one hand, and the multiparty polities of Continental Europe, on the
other.

2. Voters as a "committee"

It has been traditional in the literature on the spatial voting model to suppose that the
electorate comprises a committee. That is to say, each voter, i, has a "bliss" point xi
in a "policy" space, Z, say. It is usual to assume that Z is a compact, convex subset
of Euclidean space 9'. The "utility", ui, for voter i from some policy point y E Z is
typically assumed to be Euclidean, of the form ui = -l xi -yll22, where 1111is the norm
on 9. Almost all the early work on this spatial voting model studied the induced
strict majoritarian preference Qf on Z. Letf = {xi: i E N} denote the distribution of
voter ideal points for the society N of size n N=
IN. Let Qi: Z Z denote voter i's
Ch. 9: Representative Democracy as Social Choice 433

strict preference correspondence, given by z E Qi(y) iff liz -xil < Ily -xill. Qf is
given by z E Qf(y) iffz G Qi(y) for all i belonging to some majority coalition A of
size A > n + 1. The social equilibrium, or majority rule voting core, is

E(Qf) = {X: Qf (X) = 0}. (2.1)

Results by Plott (1967), McKelvey and Schofield (1986), Banks (1995) and Saari
(1997) can be interpreted to imply that E(Qf) is empty for almost allf E 9Wfl, as
long as w > 3. Moreover, the majoritarian preference Qf will generally be chaotic. For
almost any pair of points yo,Yr G W there exists a finite voting trajectory yl ... ,Y,- I
with y, E Qf(yt-1) for t = 1,..., r [McKelvey (1976, 1979), Schofield (1977, 1978,
1985), Austen-Smith and Banks (1998, 1999)].
In a very influential book, William Riker (1982) discussed the relevance of
these social choice results for democratic theory. Riker contrasted the populist
(or Rousseauian) view of democracy, whereby liberty is obtained "by embodying the
will of the people in the action of officials ... " [Riker (1982) p. 11], with the liberal
(or Madisonian) view that a sufficient condition for democracy is that "the government
... is administered by people holding their offices during pleasure, for a limited
period, as during good behavior". [FederalistXXX, in Bailyn (1993)]. In Riker's
opinion, the chaos results from social choice theory made the populist view untenable.
However, the liberal view was still compatible with these results. Nonetheless it did
appear possible that even liberal democracy would be beset by instability or factional
turbulence, of the kind discussed in the previous section in relation to Madison's fears
of the possible outcome of Union. It is worth mentioning that throughout the 1970s
many scholars had also drawn pessimistic conclusions about the long-term survival of
democracy in the face of the economic disorder induced by monetary instability and
the oil crises of 1974 and 1979. Indeed, these fears can be described as a "democratic
quandary". I shall discuss this quandary in Section 5 below.
A number of scholars adopted a "neo-institutional" approach arguing that political
institutions generated rules of the political game and that these rules would lead to
equilibrium behavior by politicians [see for example Shepsle (1979), Shepsle and
Weingast (1981) and Fiorina and Shepsle (1982)]. Riker's response was that in any
institutional setting, the "losers" would have an incentive to change the rules, in the
hope of gaining power [Riker (1980)]. Overall, he argued that "The most important
conclusion of [this] line of reasoning is that, in the long run, nearly anything can
happen in politics" [Riker (1980)]. In his later work on the Ratification of the
Constitution and the lead-up to the Civil War, Riker, in a sense, set out to explore
this hypothesis [Riker (1984, 1986, 1996)].
However, arguments about the stability of democracy based on the chaos theorem
have to make more precise the nature of the calculus of politicians themselves. The
simplest assumption of a two-party (or two-candidate) model of elections is that each
of the two candidates (j, k, say) simply adopts positions zj, Zk E Z in an effort to defeat
434 VN.
Schofield

Mv 2

Fig. 1. Median Lines and the yolk under ma-

jority rule with a uniform voter distribution
on me pentagon. figuree rawn y I svetan
A B Tsvetkov.)

the other. For example we could assume that the utility Uj of candidate j at the policy
profile zj, Zk was given by

I if Zj C Qf (Zk),
Uj(zj,zk) = -1 if Zk Qf(zj), (2.2)
0 otherwise.
A pure-strategy Nash (1950, 1951) equilibrium (PSNE) in this two-agent game is a
pair (,z) such that Uj(zj,z*) > Uj(z*,z*) for no zj C Z and Uk(zJ,zk) > Uk(Z<,z)
for no Zk E Z.
Obviously enough if z E E(Qf) then a possible PSNE is the pair (z,z) C Z x Z.
On the other hand, the chaos theorem implies that PSNE in this game generally do
not exist. However, there may be situations in which mixed-strategy Nash equilibria
(MSNE) exist [Kramer (1978)]. In the 1980s interest focused on characterizing MSNE
in such spatial election models.
To illustrate these results consider Figure 1. Assume that the distribution,f, of voter
ideal points is given by the uniform distribution on the pentagon, Z in the figure.
A median line (such as M1M2 in Figure 1) exactly divides the distribution in two. If
the infinite family of median lines were to intersect, then E(Qf) would be non-empty.
[For example if one had, instead of the pentagon, the disc with the uniform distribution,
then E(Qf) would be non-empty. See Arrow (1969), for an illustration.] In Figure I
the medians do not intersect. But they do generate a disc, called the yolk: the smallest
disc that touches all medians. The positioning of the yolk in Figure 1 suggests that its
location and size provide a measure of the spherical symmetry of the distribution, f.
Various alternative "attractors" of the voting process have been considered. One of
these is the uncovered set of Qf [Miller (1980)]. The general idea is to modify Q so
that it exhibits no cycles. Thus, the covering relation, Qf induced by Qf is defined by
z C Qf(y) iffz E QJ(y) and Qf(z) C Qf(y). The uncovered set, E(Qf) of Qf, is given
Ch. 9: Representative Democracy as Social Choice 435

0.6

0.4
0
.0

5 0.2
*Carter
o * Johnson
* Kennedy
o0 0
E E z~* Clinton
0
U
-0.2
r
* Bush
* Eisenhower * Reagan
©
* Nixon -0.4
* Ford

-0.6
-0.6 -0.4 -0.2 0 0.2 0.4 0.6
Nominate Common Space Dimension one

Fig. 2. "Positions" of US presidents from Eisenhower to Clinton in the two-dimensional policy space of
Poole and Rosenthal (1997).

by E(Qj) = E(Qf) = {z c Z: QJ(z) = ¢} It is known that the support of MSNE for

the two-candidate game just described, under the electoral distributionf, is contained
within the uncovered set of Qf [Banks, Duggan and Le Breton (1998, 2002)]. Earlier
studies of the uncovered set [McKelvey (1986), Cox (1987)] show that the uncovered
set is "centrally located", and converges to the social equilibrium E(Q/), when this
exists. For example in Figure 1 the uncovered set is bounded by a disc with radius
approximately twice the yolk radius.
An alternative notion, the "heart" [Schofield (1999b-d)], is a local version of the
uncovered set. In Figure 1 the heart, defined by Qf, is the set bounded by the family of
median lines. The heart contains both the yolk and the uncovered set, and can usually
be easily calculated. More importantly, the heart is "lower hemi-continuous" in all
parameters. I shall use the idea of the heart in the next section.
The conclusion of the general Downsian model [Downs (1957)] of two-party
competition is that parties, or candidates, "converge" to a domain, centrally located
with respect to the electoral center [Calvert (1985)]. However, there is very little
evidence that two-party competition induces convergence of this kind.
For example, Figure 2 presents estimates of US presidential policy positions from
Eisenhower to Clinton. The space, Z, is obtained from estimates of Poole and Rosenthal
(1991, 1997) and from National Election Survey Data. The left-right dimension is a
436 N. Schofield

usual economic dimension, while the north-south dimension concerns social policy/
civil rights. Clearly there is little indication that Democrat and Republican candidates
converge to the electoral origin.
This convergence phenomenon is a theoretical property of a committee of voters,
choosing "deterministically" in the manner described. An alternative voter model is
"probabilistic" or stochastic. Here the "utility" for voter i from party j at position Zj
takes the form ui(zj) = -z - xill2 + j, where ej is a normal random variable,
with mean 0 and variance a2. Voter i chooses party j, at zj, over party k at zk iff
ui(zj) > Uk(zk). Thus voter choice is described by a probability pair (-ip, lPik), where
ipij, the probability that i chooses, is given by Prob(-llzj -xi 112
+ j > -lzk -xi112+ Ek)
It is usual to assume that (J, Ek) are both normal and independent, and that party j,
say, adopts a position to maximize the expectation cG
i N i.
As Enelow and Hinich (1984) and Coughlin (1992) have shown, for such a model
there will typically exist a PSNE (z*,z*) where each party adopts an identical
position, z*, at the mean of the voter distribution. Figure 2 suggests that both the
deterministic and the probabilistic voter model are inappropriate as theoretical devices
to understand elections under plurality, or majority rule. We shall return to this point
in Section 4 below, where I shall argue that modeling voter choice as a committee
(based on preferences) is an invalid approach.

3. A "committee" of politicians

The deterministic voting model described in the previous section assumes that
elections are riskless: both parties know the relationship between the pair of party
declarations (zj,zk) and the electoral outcome. Even the probabilistic model assumes
that each party attempts to maximize the expectation of the vote share. This is
tantamount to ignoring the variance in the "stochastic" vote share variables.
Cox (1984) proposed an alternative model that emphasized the stochastic nature of
the electoral response. First note that there are three states of the world resulting from
the election when just two parties or candidates compete. Let us use Do to denote
a draw (where both parties or candidates gain equal shares of votes, or seats),
Dj to denote the state in which party j wins and Dk to denote the state where k wins.
Each party,j say, is assumed to have a "Euclidean" utility from the policy z of the form
-liz -yjll2 , where yj is party j's bliss point. In state Dj party j also gains government
perquisites, say ajSc.
Prior to the election, j, k declare policy positions (z,zk). If state Dj ensues, then
j implements zj, and takes the government perquisite ajs.
If state D0 occurs, then a lottery, say go(zJ, Zk) results: that is, (I, k} bargain together
and implement a randomized policy (Zj + Zk) dividing the perquisite equally. Let
sot, I, srk: 2 [0, 1] denote the electoral probability functions, so, for example, the
probability of a draw when (zj, zk) is declared is ro(zj, zk).
Ch. 9: Representative Democracy as Social Choice 437

Table I
The Election in Britain, May 1, 1997

Party Seats % Seats % Vote

Labour 419 63.6 44.4

Conservatives 165 25.0 31.4
Liberal Democrats 46 7.0 17.2
Scottish National Party 6 0.9 2.0
Plaid Cymru 4 0.6 0.5
Indep./Others 19 a 2.9 4.5

Northern Ireland (of which 18)

Ulster Union 10
UK Union 1
Democratic Union 2
SDLP 3
Sinn Fein 2

Effective Number 2.12 3.12

a Of these nineteen, 1 is an independent (i.e., a constituency in Cheshire). The other 18 are in Northern
Ireland, as indicated.

Then the utility to j, say, of the underlying game form g is:

Uj(g(Zj,Zk)) = (zj,k) [a6 - lz -Yil12]

+ :rk(Zj,Zk) [-JIZk -Yjll2 ] (3.1)
+ (Zj,Zk) [ aj6- I-(zj+ k)-Yj 12] -

In this formulation, when party k wins, then partyj receives no government perquisites.
Cox (1984) suggested that, under fairly natural conditions, including continuity of
the probability functions, there would exist a PSNE for this game form. Of course,
most parliamentary elections involve more than two parties. Indeed, many years ago
Duverger (1954) suggested that electoral systems based on proportional representation
would result in numerous parties. Conversely, he argued that so called "first past the
post", or plurality elections, based on single member districts, would tend to result in
two-party politics. Britain is the obvious example of a plurality electoral system. As
Table 1 illustrates, in the election in May 1997 in Britain, about 44% of the popular
vote for Labour was sufficient to give it over 60% of the seats. However, the vote for
the small center Liberal Democrat Party (at 17%) was by no means negligible. The
empirical question of determining "the probability of winning" in such an electoral
system is extremely difficult. Using survey techniques it is of course possible to
438 N. Schofield

estimate the popular vote shares, but these do not translate directly into seat shares.
I return to this question in the next section.
By contrast, Table 2 presents the outcomes in the last four elections in Israel. The
electoral system is extremely proportional, in that vote shares and seat shares are
almost identical. (Small parties with less than 2% of the popular vote received no
seats).
To understand political motivations in such an electoral system, it is necessary to
note that no party may expect to win the election (if this is interpreted to mean "gain
a majority of the seats").
However, it is possible to model post-election bargaining between parties. To
understand the nature of the policy space in Israel, surveys conducted by Arian and
Shamir (1995, 1999) for the 1992 and 1996 elections were utilized. Factor analysis of
responses by a sample of voters (of size approximately 1000) gave a two-dimensional
space, Z. The left-right dimension correlated with attitudes to the PLO, and the north-
south dimension described beliefs regarding the proper relationship between the state
and the Jewish faith. The pre-election declarations (or manifestos) of the various parties
were processed using the Arian-Shamir questionnaire. The "positions" or declarations
of the various parties are marked in Figure 3 for 1992 and in Figure 4 for 1996.
[See Schofield, Sened and Nixon (1998) and Schofield and Sened (2002) for further
details].
To extend Cox's model of electoral competition, let us regard the situation after the
1992 election as a committee comprising the ten parties with seat strengths as given
in Table 2. Let us use DI to denote the family of "winning" or majority coalitions
after the election of 1992. Obviously Labor, together with the three small parties (on
the left), with 61 seats, form a majority. Likud, and the religious parties, together with
Tzomet, have only 59 seats, and so need another party, such as Meretz, to attain a
majority.
Further, let us ignore government perquisites for the moment and assume that each
party has Euclidean utility based on an ideal policy point given by its declaration
(as illustrated in Figure 3). With these party preferences and voting weights, the
"social preference" in the Knesset can be denoted (D,. It should be obvious that
the "parliamentary" core, or voting equilibrium E(oD,) is non-empty, and located at
the Labor party declaration. To see this, note that the "compromise set" for each
coalition is simply the convex hull of the ideal points of the coalition members. Clearly
all "compromise sets" for the various winning coalitions intersect at a single point,
namely that of the Labor party. A model of political bargaining proposed by Banks and
Duggan (2000) suggests that the only possible outcome of bargaining in the coalition
situation D1 is precisely this equilibrium or core point. Moreover Laver and Schofield
(1990) argue that Labor would form a minority government. In fact, Labor under
the leadership of Rabin did form a minority government, with implicit support from
Meretz. As a consequence, the peace accords with the PLO were initiated.
After the election of 1996 a very different coalition structure, D2, was created. It
should be clear from Table 2 that under D2, the coalition excluding Labor and the
Ch. 9: Representative Democracy as Social Choice 439

Table 2
Elections in Israel

Party Knesset Seats

1988 1992 1996 1999

Left:
Labor 39 44 34 28
Meretz - 12 9 10
Shinu 2 - - 6
Others 9 - -
Dem Arab 1 2 4 5
Commun. 4 3 5 3
Balad - - - 2

Subtotal 55 61 52 54

Center:
Olim - - 7 6
Yisrael Beiteinu - - - 4
Third Way - - 4
Center - - - 6

Right:
Likud 40 32 32 19
Tzomet 2 8 -

Subtotal 42 40 43 35

Religious:
Shas 6 6 10 17
Aguda 5 4 -
Yahadut 2 - 4 5
NRP 5 6 9 5
Moledet 2 3 2 4
Techiya 3 -

Subtotal 23 19 25 31

Total 120 120 120 120

440 V. Schofield

eq'
Yahadut
00
Shas )

Dem-Arab l I
.V o-
Y Corr munists ;,

I-

I I I I
-2 -1 0 1 2
Security

Fig. 3. Party Positions in the Israeli Knesset in 1992, also showing the 95%, 75%, 50% and 10% contour
lines of the highest-density plot of the voter sample distribution [from Schofield and Sened (2002)].

-2 -I 0
Security

Fig. 4. Party Positions in the Israeli Knesset in 1996, also showing the 95%, 75%, 50% and 10% contour
lines of the highest-density plot of the voter sample distribution [from Schofield and Sened (2002)].
Ch. 9: Representative Democracy as Social Choice 441

"left" controlled a majority of 68 seats. In this case, the party preferences and voting
weights give a different "social preference", cr2 say. It is evident that the core E(crD2) is
empty. Essentially there are, in principle, three different coalition governments possible
after 1996: one based on Likud, Shas, and the religious parties; a national unity
government of Likud and Labor; and a coalition of Labor, and the "left" together
with Shas. Since the equilibrium, or core, E(a(D 2 ) is empty there is no possibility of
a minority Likud government. In fact, Netanyahu (leader of Likud) won a separate
prime-ministerial election against Peres (leader of Labor). Although Labor was the
larger of the two parties, Netanyahu constructed a majority coalition that depended on
Shas.
I suggest that bargaining between the parties is constrained to a set known as the
"heart" 4 . Under Euclidean policy preferences, the "parliamentary" heart after 1996
is simply the convex hull of the declarations of Labor, Likud and Shas. Because
Netanyahu won the separate election, the outcome after this election can be seen to be
a point on the arc joining the Likud and Shas positions. However, because the Likud
position was not an equilibrium, the Likud-led government was eventually defeated
under a vote of no-confidence. (One aspect of this defeat was the defection of Gesher,
a small centrist party, out of its alliance with Likud. Gesher is not listed in Table 2,
since its seats counted as part of Likud.) After the 1999 election, a majority coalition
led by Barak, of Labor, was constructed that collapsed two years later.
It is possible to construct a Nash equilibrium model that can be used to explain how
parties choose declarations prior to an election. Firstly, suppose that {D 1,... ,DT}
denotes the collection of all possible coalition structures that can occur at the time
of the election. Let P denote the set of parties, { 1,... ,p}, contesting the election.
Suppose that the various parties have selected their leaders, and that the vector of
leader positions is denoted z = (zl,...,z, .z., zp).
Assume that each leaderj has a smooth, strictly convex preference correspondence
q,(zj): Z x A - Z x Ap.
This is derived from a utility function of the form u(y, 6) = -ly-zjl/2
+ aj6,
where 6 represents government perquisites. The symbol Ap denotes the (p - 1)-
dimensional simplex. Let qa(z) denote the profile of leader preferences, which we can
regard as a profile on W = Z x Ap.
Once the election outcome (in terms of the coalition structure Dt) is known, then
this, together with the profile of leader preferences, generates a committee preference,
which I shall now denote as o, (z). It may be the case (as in the example of the Knesset
in 1992) that the parliamentary core E(u,(z)) is non-empty. Even when the core is
empty, however, it has been shown [Schofield (1999d)] that the more general solution,
the "parliamentary heart", is non-empty. Because the preferences of the committee are

4 A point x is in the "heart" defined by the "parliamentary" preference relation aD iff there is a
neighborhood V of x such that V n &D(x) -_ . Here 0a is the covering relation induced from the
preference D.
442 N. Schofield

defined on the space W = Z x Ap, the heart 7-t(z), say, is a subset of W. For the fixed
coalition structure, D,, the heart can be regarded as a correspondence f7-l: ZP - W.
Moreover, this correspondence admits a continuous selection [Michael (1956)]. I now
assume that for each z, and D,, the outcome of coalition bargaining is a lottery g7'(z),
whose support is contained within the heart. I further assume that for each fixed D,,
this outcome function ga: Zp - ~F is smooth. Here W denotes the space of lotteries
on W, endowed with the weak topology [Parthasathy (1967)]. The function g' is meant
to represent the beliefs that the political elite have about the nature of the bargaining
game, at the fixed coalition structure, Dt. However, prior to the election the elite cannot
be sure which coalition structure will occur. I assume, however, that at the vector, z,
of party-leader positions, the political elite can estimate the probability ,r(z) that D,
occurs. I also assume that each electoral probability function :t,: Zp - [0, 1] is smooth.
With these smoothness assumptions on g = {gt}, and on n = {z,}, it is possible to
show that party principals can rationally choose party positions.
To be more precise, I assume that the leader of party j is chosen by a principal
of the party (that is one of the elite party members). The principal cares about final
government policy, and about government perquisites. The game form g and the
electoral probability, r, are assumed to be common knowledge. The preference of the
principal of party j is described by a von Neumann-Morgenstern utility function Uj,
which represents the preference held by the principal over both policy and government
perquisites. At a vector, z, of declarations, the induced utility, lUj, of principal j, is
given by
T

Ujf(z)= >
t1l
(z) Ui(g(z)). (3.2)

Equation (3.2) can be seen to be an extension of Cox's Equation (3.1) to the more
complex case of multiparty competition. A local Nash equilibrium (LNE) of the game
({Uj: j E P},ga) is a vector z* e Z P , such that for each j there is a neighborhood V
of zj, with

Zj = argmax{Uj(z*,... ,z,... ,): j C Vj}. (3.3)

Schofield and Sened (2002) have shown that a LNE "generically" exists, under the
assumptions that have been made on ga and mr.
The model that has just been proposed attempts to incorporate a number of
features that seem important for political decision-making. First of all, any policy
declaration made by a party must be the result of compromise of some sort among the
heterogeneous party elite. I assume above that this compromise party preference can be
identified with that of a party principal. This principal then chooses a leader to present
to the electorate. The leader's policy preferences will be credible to that electorate. In
choosing the leader, the principal must balance the effect the leader will have on both
the electoral response, and on post-election coalition bargaining. Obviously enough, if
Ch. 9: Representative Democracy as Social Choice 443

government perquisites are highly valued, then the principal may focus on choosing a
leader who is centrally located, or electorally popular. If policy is more highly valued,
then the principal may choose a "radical" leader, capable of vigorous bargaining with
other parties over government policy.
To implement such a model requires both an estimation of the electoral risk
functions {rt}, as well as a determination of party-leader positions, and of principals'
preferences (or at least the policy preferences within each party).
The next section reports on recent work that has been undertaken on estimating
7r in different countries over a number of elections. The results suggest that elections
more closely resemble jury choice, in that the choices of voters are determined by
changing beliefs, rather than fixed preferences. Moreover, the choices of party positions
(or leaders) seem not to be determined by attempts to maximize expected vote, but
rather to influence government policy.

4. Elections as methods of belief aggregation

The existence of a "political equilibrium" proposed in the previous section depends

on the smoothness of the electoral probability functions {rt}. Since the model is one
of political choice prior to an election, these probability functions should be regarded
as beliefs of the political elite, concerning the nature of electoral response. To model
these beliefs, we can use survey data to estimate an electoral model in the following
fashion. Given a vector z = (zi,..., zp) of party declarations, we can assume that the
utility ui of voter i is represented by a (px 1) vector

Ui(z) = Vi +Xfi + Syi + . (4.1)

In a general multinomial probit (MNP) model, the (px 1) vector consists of a

disturbance term, which is multivariate normal, with mean 0, and a variance/covariance
matrix Z. In the simpler multinomial logit model (MNL) the disturbances are
iid (independent and identically distributed). The (pxm) matrix X can incorporate
socio-structural/individual features (such as class, education). The (px 1) vector Vi
describes policy aspects. In the pure spatial model Vi is simply the vector (-d),
where d = lzj - xill is the policy distance between party j and voter i's ideal
point. The matrix Si includes choice features such as the "likability" of party leaders,
past economic performance of parties, etc. Voter i chooses party j with probability
Wij(z) = Prob[uij(z) > uik(z) for all k j]. Thus the model is stochastic and voter i's
behavior is described by a probability vector i (z) = ( il(z), .. ip(z)).
If this empirical model is based on a sample, N, of size n, then the estimated vote
share of party j at Z is obviously CN Wij (z). However, because v = { i}N are
random variables, the vote shares will have significant variance. Intrinsic to this model
is the existence of electoral risk. By modeling the electoral response, q', it is possible
to construct estimates of the probability functions {rtt} used in the previous section.
Such a model generates smooth functions {t}.
444 N. Schofield

Because the Arian-Shamir survey data for 1992 and 1996 on Israel included
voter intentions, it was possible to construct a pure spatial model of the 1992 election.
In order to do so, it was necessary to incorporate constant terms (in the matrix S)
for each of the parties. As indicated above, these terms described non-policy aspects
of the parties (such as popularity of the party leaders). In 1992 the constant was
largest for Labor, smaller for Likud, and much smaller for Meretz and the other parties
[Schofield, Sened and Nixon (1998)]. We can infer from this that voters understood
that Labor and Likud were the principal protagonists in the 1992 election. Voters whose
preferred policy positions were closer to one of the religious parties apparently voted
for Likud [see also Cox (1997), for an analysis of strategic voting]. An examination
of the expected vote shares for Likud and Labor suggested that these two parties were
close to positions that maximized expected vote shares. However, the smaller religious
parties, according to our estimation, could have dramatically increased their expected
vote share by moving to the electoral center. The model proposed in the previous
section suggests a reason why parties may adopt local Nash equilibrium positions that
do not maximize expected vote share.
Consider for example the religious party Shas. As I indicated previously, under the
coalition structure D2, Shas could expect to belong to government. The bargaining
model proposed by Banks and Duggan (2000) suggests that Shas, by adopting a policy
position far from the center, could bargain more effectively with Likud or Labor. Thus,
if the subjective probability 7r2 held by Shas increases, the model suggests that Shas
will adopt a more radical position. [A more detailed formal analysis of this observation
is offered in Schofield and Parks (2000).]
Figures 3 and 4 do indeed suggest that the parties adjust their policy positions from
one election to another. Note, however, that these adjustments are small, which is in
keeping with the emphasis on local Nash equilibria made in the previous section.
It has been argued that, under vote maximization, parties will converge to an
electoral center [Lin, Enelow and Dorussen (1999)]. There is no evidence from the
Israeli examples presented here that convergence in the Downsian sense occurs. In
fact, the density plot of the sample voter distribution in Figure 4 suggests that this
distribution is strongly "bipolar". The "center" of the electoral distribution is located
between two peaks of the distribution, and only small parties (such as Olim or Gesher)
adopted positions near this center.
Indeed, Israel appears to exemplify a hypothesis offered by Duverger. As he wrote
"the centre does not exist in politics: there may well be a Centre party, but there is
no centre tendency, no centre doctrine ... Every Centre is divided against itself..."
[Duverger (1954) p. 215].
A further inference can be drawn from Figures 3 and 4. It appears that the sample
distribution of voter ideal points changes over time. In a sense the religious dimension
has become more important, as the empirical correlation between attitudes on security
and religion has declined. We can infer that voter choices are derived not so much
from preferences, which are relatively stable over time, but from beliefs that change
in accordance with a changing understanding of the nature of the political world. It
Ch. 9: Representative Democracy as Social Choice 445

E
.. 0-
2
2 o-

0
0

Fig. 5. Party Positions in the British

|D1 _ _ . A 1.... . Is .... .........Aa
rarlamenr in I /, also snowing
" - a highest-density plot of the voter
_ sqmlo Aitrihetinnr4r^m,--- hi~o
I I I I I Sllllp; UIltLI IUU LIIlUl UlpuullJbllCU
-2 -1 O 1 2 research by Quinn, Martin and Whit-
General (L-R) ford (1999)].

should also be observed that recent elections in Israel illustrate Riker's (1953) point that
proportional representation may increase political fragmentation. While compromise
is necessary in such a PR system, increasing fragmentation may make such agreement
very difficult to attain.
Other empirical work on voting in proportional electoral systems suggests that
voter choice is best understood as a method of belief aggregation rather than
preference aggregation. Larger parties in countries such as the Netherlands and
Germany tend not to converge to the electoral center. Voter choice is partially
determined by policy considerations, but also by non-policy considerations such as
leader dependability, etc. [Schofield, Martin, Quinn and Whitford (1998)].
Modeling voter choice in a plurality electoral system such as Britain is made difficult
because the electoral model gives vote shares, rather than seat shares. However, a recent
MNP analysis [Quinn, Martin and Whitford (1999)] is suggestive.
Figure 5 presents a picture of British politics in 1979. The background represents
the distribution of voter ideal points (the contour lines give the 95%,75%,50%
and 10% density regions). The first left-right dimension describes a general economic
factor, and the second delineates preferences over the "scope of government". This
distribution is estimated from a factor analysis of a voter sample undertaken by Rabier
and Inglehart (1981). To obtain party positions a survey of political elites [ISEIUM
(1983)] was utilized. This latter sample gave party-specific distributions of political
elite responses. The two-dimensional median within each party was utilized to estimate
each party position. Note that this estimated position can be interpreted as the position
of the party "principal" (as utilized in the previous section), not the party "leader".
A classical socio-structural model of voting utilizes individual voter properties to
446 ,X Schofieid

Table 3
Elections and Estimations for Britain, 1979

Party Election vote Election seats Sample % Estimation 95%

(%)a (%)b Con. Int. c

Labour 36.9 (39.0) 43.45 35.41 35.24 (30.2, 40.5)

Conservatives 43.9 (46.4) 54.75 53.58 53.55 (48.0, 59.0)
Liberals 13.8 (14.6) 1.75 11.01 11.21 (7.7, 15.0)
Scot Nationals 1.6
Welsh National 0.4
N. Ireland, etc. 3.9
Effective Number 2.80 2.15

a Numbers in parentheses are vote shares ignoring small parties and independents.
b For simplicity, these are recalibrated as percentage of seats of the total number held by the three
parties. Other, regional, parties only controlled 2.5% of parliamentary seats.
c Abbreviation: Con. Int., Confidence Interval.

estimate voter choice. In Equation (4.1), this is equivalent to assuming the vector
Vi = 0. Utilizing variables such as religion, income, education, it is possible to estimate
how the sample (of size 426) voted. Actual voter choice was known, and the estimation
correctly predicted 49% of the votes. A pure spatial model assumes X - 0, and
estimates Vi making use of voter and party positions. In the estimate for 1979 a small
positive constant term for the Conservatives, and a negative term for the Liberals, were
incorporated. Table 3 gives the actual election votes share together with the estimated
vote response.
Because the model is fitted to the sample response, the estimated expected
vote shares are very close to the sample vote shares. What is perhaps more relevant
is that approximately 49% of the sample choices were correctly predicted. In a joint
model, utilizing both spatial and socio-structural variables, the correct prediction
increased to 51%. Analysis of the Bayes factors [Kass and Raftery (1995)] suggested
that the pure spatial model was the best statistical estimator of voter response in
Britain.
This analysis raises a number of interesting questions about voter choice under a
plurality electoral system such as Britain's.
Firstly, constant terms [as designated by S in Equation (4.1) were utilized. However,
estimated confidence intervals on these constants gave no indication that they were
statistically different from zero.
Secondly, the spatial model did correctly predict about 50% of the voter choices,
but was poor at correctly assessing the choice of voters near the center of the electoral
distribution. Such voters would be expected to exhibit a high probability of voting
for the Liberals. It is probable, of course, that "centrist" voters strategically chose
one of the two large parties. However, the wide confidence intervals on the party-
Ch. 9: Representative Democracy as Social Choice 447

specific constant terms suggest that modeling the strategic component of voter choice
is extremely difficult.
Thirdly, the modeling exercise did not utilize party declarations or leader positions,
but rather estimates of the average position of the elite in each party. It was obvious,
however, that the two large parties could potentially have increased their "expected
vote shares" by modifying their policy positions. Since there is no evidence that they
did this, it seems implausible that the usual Downsian vote-maximization interpretation
can be valid.
To further explore the nature of party choice in Britain, the British General Election
Studies of the elections of 1992 and 1997 were utilized. The questions on the
surveys covered topics such as taxes, nationalization, redistribution, the European
Community, women's rights and Scottish Nationalism (for Scottish voters). Moreover,
the questionnaire asked voters to (subjectively) locate the three major parties (Labour,
Conservative and Liberal Democrat). Factor analysis gave scores to each of the
questions, so that a policy space could be constructed. Unlike the analysis for 1979,
a single dimension was constructed. A key aspect of this policy dimension was the
extent of redistributive policies.
Figure 6 presents an estimate of the distribution of the sample of voters (excluding
those from Scotland) in this policy space, together with the perceived positions of the
three main parties, in 1992 and 1997.
It is, of course, difficult to directly compare the figures for 1992 and 1997, since the
composition of the policy space may have changed slightly. Nonetheless, the figures
suggest that between 1992 and 1997 Labour was perceived to have moved nearer the
center of the distribution, while the Conservative Party moved further away.
However, again there is no evidence whatsoever that the two main parties
"converged" towards an electoral center. Clearly leadership choice in Britain involved
an element of risk-taking.
The empirical analyses discussed in this section give some hints about the nature of
political choice in electoral systems based on proportional and plurality systems. One
point should perhaps be made on the differences between plurality electoral systems,
such as Britain's, and those based on proportional representation. The game described
by Equation (3.2) in the previous section is intended to be relevant to both types
of political systems. However, the nature of electoral risk is different under plurality
and proportionality. Relatively small changes in electoral support under plurality can
dramatically change the probability functions, so in some sense political choice under
plurality is riskier than under proportionality. Although it has not been formally derived
in any way here, it seems likely that there are circumstances under which the political
elite in a plurality system will be more willing to take political risk than the elite in
a political system based on proportional representation. The next section will offer
a number of speculations about the degree to which such differences in electoral
institutions may lead to different political and economic choices.
448 i. Schofield

0a,
CO
2C
0

¢ 0
0

'8

-2 -1 0 1 2
Economics
C

c)

=-
0 01;
to

0
(/3

a
0

6j
0

-2 -1 0 1 2
Economics

Fig. 6. Estimated density function of voter ideal points and the party positions of Labour (Lab), Liberal
Democrats (Lib) and Conservatives (Con) in a one-dimensional policy space, for 1992 and 1997.

5. Electoral risk-taking and economic or political quandaries

As noted in the Introduction, it is well understood that plurality electoral systems

tend to give rise to two-party systems. Indeed, the use of a plurality electoral system
would seem to reduce the degree of fragmentation of political support [Rae (1967)].
For example, if we use the "effective number" (the inverse of the Herfindahl index
of concentration) as a measure of fragmentation, then the effective number based on
vote shares in Britain was 2.8 in 1979 and 3.12 in 1997. Using seat shares, the effective
numbers were 2.15 in 1979 and 2.12 in 1997. (See Tables 1 and 3). In contrast, the
Ch. 9: Representative Democracy as Social Choice 449

effective number (in both vote and seat shares) in Israel increased from 4.62 in 1988
to 9.61 in 1999.
There is a "correlation" between effective number of seat shares and government
duration in European polities based on proportional representation. The correlation is
quite weak however. In Italy, for example, the average effective number was 3.5 (for
the period 1945 to 1987) while government duration averaged 13 months. Nonetheless
the Christian Democrat party was in every government. In the Netherlands, with
an average effective number of 4.5, average duration was 27 months [Schofield
(1993)]. With plurality rule, and single-party majorities, Britain has had relatively
long-lived governments. Duverger (1954) has argued that it is not the fragmentation
of polities based on proportional representation that is of concern. Rather, under
"proportionality", governments must necessarily be based on coalitions, and such
coalitions require compromise. In contrast, under plurality rule voters are given a clear
choice. The resulting majority party can thus obtain a mandate to pursue its declared
economic or political objectives.
In the 1970s it seemed entirely possible that electoral systems based on plurality rule
were subject to the factional turbulence feared by Madison. Many authors observed
that attempting to adjust inflation-unemployment levels to electorally advantageous
values could induce inflationary expectations. In particular, since a plurality electoral
system magnifies small swings in voter preferences, governments in Britain and the
USA could be held hostage by small, but powerful groups [Beer (1982)]. In contrast,
under proportional electoral rule, it was suggested that different factions in the polity
would be represented by different parties and that these could bargain more effectively
with each other so as to attain economically efficient outcomes [Crouch (1985)].
A second line of argument, based perhaps more on economic reasoning, suggested
that both Britain and the USA were undergoing relative economic decline vis
vis the economies of western Europe and Japan. Figure 7 presents estimates of
GDP per capita (in constant 1985 dollars) of six OECD countries. It is obvious enough
that the gap of approximately 2000 dollars (in 1985 values) between Britain and
Germany as of 1950 has disappeared, and indeed been reversed. Moreover, the gap
of over $5000 between the USA and Germany (as of 1950) has decreased to about
$3200 (in 1985 values). However, Figure 7 suggests that GDP/capita in the West-
European countries is flattening out, or reaching an asymptote, at a lower level than
in the USA.
Indeed, relatively high levels of unemployment in Germany, Italy and France have
persisted for some time. (As of February 2002 these levels were approximately 9.6%,
9.3% and 9.0%, respectively). In Japan a low average level of unemployment of
2.5% (for the 1980s) has increased to 5.6%. In the USA and Britain unemployment
is currently 5.6% and 5.2% (respectively), much lower than the average levels of
7.1% and 9.5% for the 1980s [see Garrett (1998), and the discussion in Schofield
(2000)]. It would seem plausible that the nature of the political system in these various
countries can induce macro-economic effects of this order of magnitude. I shall offer
450 N. Schofield
4.Sh e

-- Germany
A France
-x- Japan
-*- Italy
-- Britain
-- USA

Fig. 7. Estimates of GDP/capita (in 1985 dollars) for six OECD countries, 1950-1992. (Figure drawn
by Tsvetan Tsvetkov.)

a number of speculative hypotheses, that derive from Duverger's discussion, in an

attempt to account for these differences.
Firstly, consider polities based on proportional representation.
In a country such as Israel, with a low electoral threshold, it is relatively easy for new
parties to form and enter the political arena. As this occurs and political fragmentation
increases, it may become increasingly difficult to construct and maintain coalitions.
Although an equilibrium model for party choice was proposed in Section 4, this does
not necessarily imply political stability. In fact, the game form, denoted g, may become
much more complicated as the number of possible coalition structures increases. The
complexity of coalition bargaining suggests why Netanyahu was eventually defeated,
and an election required in 1999.
In both Germany and France the electoral system is approximately proportional,
although the effective thresholds are quite high. In a sense this imposes a bound on the
level of political fragmentation. Nonetheless coalition governments are the norm. For
example, in Germany after the last election, Gerhard Schr6der of the Social Democrat
Party has had to depend on the Greens for a majority. Similarly, Jospin, leader of the
Socialists in France, depends on support from the far left. It seems plausible that the
necessity for coalition restricts the ability of political leaders to implement policies that
are deemed risky by significant portions of the electorate. Attempting to deal with high
unemployment might necessitate restricting the intervention of the state, or dealing
with budget deficits. Both Schr6der and Jospin have found these strategies difficult to
deploy.
Ch. 9: Representative Democracy as Social Choice 451

Now consider the USA and Britain. Obviously their political institutions are different
from each other. However, both presidential elections in the USA and parliamentary
elections in Britain are based on plurality electoral systems. This chapter has argued
that all elections are risky, in the sense that they involve significant variance. Moreover,
the degree of risk (or variance) under plurality rule is, in a sense, much greater than
under proportional rule. As we have seen, a party in Britain may gain an overwhelming
Parliamentary majority with approximately 40% of the vote. Indeed, an election for
president in the USA, or government in Britain, may in fact be won by offering what
may indeed be a risky strategy. Both Britain in 1979 and the USA in 1980 appeared to
be in an economic quandary, a situation of extreme uncertainty over appropriate policy
to pursue in an attempt at reducing inflationary expectations. Reagan and Thatcher
both offered relatively untested and risky policies that they believed would eventually
lead to a significant transformation in economic behavior. In a sense, they invited the
electorates of the two countries to act as juries, to perform a judgement on the relative
merits of different kinds of policies. This does not mean to say that the Condorcet
Jury Theorem necessarily applies to the elections of 1979 and 1980. However, it is
clear that the judgements of the two electorates brought about what can be viewed as
constitutional change in these two polities. Both economies had to undergo a period
of high interest rates, and unemployment. Eventually their labor markets and their
economic structures were transformed.
Clearly, presidential choices do not entirely determine US politics. Since po-
litical choice in Congress necessarily involves compromise, the committee-based
model of decision-making would suggest that congressional choices will tend to
generate a greater degree of risk-aversion than that displayed by US presidents.
Efforts by Franklin Roosevelt, Harry Truman, John Kennedy, Lyndon Johnson and
Ronald Reagan to address fundamental economic or political quandaries were often
blocked or made more difficult by Congressional decisions. Part of the genius of
the US Constitutional design may have been to balance presidential risk-taking with
Congressional risk-aversion.
This remark is not, of course, intended as unqualified praise for the US system of
representative government. The presidential election of 2000 obviously left much to
be desired.
European political systems also display undesirable features. Corruption scandals
have plagued France, Germany and Italy. The move towards European Union has
also created possibilities for corruption and bureaucratic incompetence. On the one
hand, qualified (or weighted) rule in the Council of Ministers, and a weak European
Parliament suggests European-wide decision-making will be unambitious in any
attempt at making their markets more competitive [Schofield (2000)]. On the other
hand, there is continuing risk-preferring pressure to deepen the Union, and to extend
it to the countries of Eastern Europe. How Europe will respond to these constitutional
quandaries is unclear.
452 . Schofield

6. Concluding remarks

Societies often face constitutional quandaries: should a Declaration of Independence

be made? should slavery be resisted? should Hitler be opposed? should states federate?
It would seem that the beliefs of the members of the society are called into question
in such situations. Both Condorcet and James Madison raised the question about how
a "fit choice" could be made by the representatives of the society. It is this question
that this chapter pursues, in an attempt to see whether methods of social choice theory
can contribute to our understanding of representative democracy.

References

Arian, A., and M. Shamir, eds. (1995), The Elections in Israel 1992 (Michigan University Press,
Ann Arbor, MI).
Arian, A., and M. Shamir, eds. (1999), The Elections in Israel 1996 (SUNY Press, Albany, NY).
Arrow, K.J. (1951), Social Choice and Individual Values (Wiley, New York).
Arrow, K.J. (1969), "Tullock and an existence theorem", Public Choice 6:105-111.
Austen-Smith, D., and J. Banks (1996), "Information aggregation, rationality and the Condorcet jury
theorem", American Political Science Review 90:34-45.
Austen-Smith, D., and J. Banks (1998), "Social choice theory, game theory, and positive political theory",
Annual Review of Political Science 1:259-287.
Austen-Smith, D., and J. Banks (1999), Positive Political Theory 1: Collective Preference (Michigan
University Press, Ann Arbor).
Bailyn, B. (1993), Debate on the Constitution (Library of America, New York).
Banks, J.S. (1995), "Singularity theory and core existence in the spatial model", Journal of Mathematical
Economics 24:523-536.
Banks, J.S., and J. Duggan (2000), "A bargaining model of collective choice", American Political Science
Review 94:73-88.
Banks, J.S., J. Duggan and M. Le Breton (1998), "The core and the uncovered set with an arbitrary
distribution of voters", Unpublished typescript (University of Rochester, New York).
Banks, J.S., J. Duggan and M. Le Breton (2002), "Bounds for mixed strategy equilibria and the spatial
model of elections", Journal of Economic Theory, forthcoming.
Beer, S. (1982), Britain Against Itself (Faber and Faber, London).
Beer, S. (1993), To Make a Nation (Harvard University Press, Cambridge, MA).
Black, D. (1958), The Theory of Committees and Elections (Cambridge University Press, Cambridge).
Reprinted in 1998 in: I. McLean, A. McMillan and B. Munroe, eds., Collected Writings of Duncan
Black (Kluwer, Boston).
Brewer, J. (1988), The Sinews of Power: War, Money and the English State, 1688-1783 (Harvard
University Press, Cambridge).
Calvert, R. (1985), "Robustness of the multidimensional voting model: candidates, motivations,
uncertainty and convergence", American Journal of Political Science 29:69-85.
Condorcet (M.J.A.N. de Condorcet) (1785), Essai sur l'application de l'analyse la probability des
decisions rendues la plurality des voix (Imprimerie Royale, Paris). Translated extracts (1994), in:
1.McLean and E Hewitt, eds., Condorcet: Foundations of Social Choice and Political Theory (Edward
Elgar, Aldershot, England) pp. 120-158.
Coughlin, P.J. (1992), Probabilistic Voting Theory (Cambridge University Press, Cambridge).
Cox, G.W. (1984), "An expected-utility model of electoral competition", Quality and Quantity 18:
337-349.
Ch. 9: Representative Democracy as Social Choice 453

Cox, G.W. (1987), "The uncovered set and the core", American Journal of Political Science 31:408-422.
Cox, G.W (1997), Making Votes Count (Cambridge University Press, Cambridge).
Crouch, D. (1985), "Corporatism in industrial relations: A formal model", in: W Grant, ed., The Political
Economy of Corporatism (Macmillan, London) pp. 63 88.
Downs, A. (1957), An Economic Theory of Democracy (Harper and Row, New York).
Duggan, J., and C. Martinelli (1999), "A Bayesian model of voting in juries", Unpublished typescript
(University of Rochester, New York).
Duverger, M. (1954), Political Parties: Their Organization and Activity in the Modem State (Wiley, New
York).
Duverger, M. (1984), "Which is the best electoral system?" in: A. Lijphart and B. Grofman, eds.,
Choosing an Electoral System (Praeger, New York) pp. 31-39.
Enelow, J.M., and M.J. Hinich (1984), The Spatial Theory of Voting: An Introduction (Cambridge
University Press, Cambridge).
Feddersen, T., and W. Pesendorfer (1998), "Convicting the innocent: the inferiority of unanimous juries",
American Political Science Review 92:23-35.
Filmer, R. (1949), in: Patriarcha and Other Political Writings, edited and with an introduction by P
Laslett (Basil Blackwell, Oxford); originally published in 1652.
Fiorina, M., and K.A. Shepsle (1982), "Equilibrium, disequilibrium, and the general possibility of a
science of polities", in: P. Ordeshook and K. Shepsle, eds., Political Equilibrium (Kluwer, Boston)
pp. 46-64.
Garrett, G. (1998), Partisan Politics in the Global Economy (Cambridge University Press, Cambridge).
Hobbes, T. (1651), Leviathan: or the Matter, Forme, and Power of a Common-Wealth, Ecclesiastical
and Civill. Republished in 1968, edited and with an introduction by C. MacPherson (Penguin,
Harmondsworth, UK).
ISEIUM (1983), European Elections Study: European Political Parties; Middle Level Elites (Europea
Institut, Mannheim).
Kass, R., and A. Raftery (1995), "Bayes factors", Journal of the American Statistical Association
90:773-795.
Keynes, J.M. (1936), The General Theory of Employment, Interest and Money (MacMillan, London).
Keynes, J.M. (1937), "The general theory of employment", Quarterly Journal of Economics 51:209-223.
Kramer, G.H. (1978), "Existence of electoral equilibrium", in: P Ordeshook, ed., Game Theory and
Political Science (New York University Press, New York) pp. 375-390.
Ladha, K. (1992), "Condorcet's jury theorem, free speech and correlated votes", American Journal of
Political Science 36:617-634.
Ladha, K. (1993), "Condorcet's jury theorem in light of de Finetti's theorem: majority rule voting with
correlated votes", Social Choice and Welfare 10:69-85.
Ladha, K. (1995), "Information pooling through majority voting: Condorcet's jury theorem with correlated
votes", Journal of Economic Behavior and Organization 26:343-372.
Ladha, K., and G. Miller (1996), "Political discourse, factions and the general will: correlated voting
and Condorcet's jury theorem", in: N. Schofield, ed., Collective Decision Making (Kluwer, Boston)
pp. 393-410.
Laver, M., and N. Schofield (1990), Multiparty Government: The Politics of Coalition in Europe (Oxford
University Press, Oxford). Reprinted in 1998 (Michigan University Press, Ann Arbor).
Lin, T., J. Enelow and H. Dorussen (1999), "Equilibrium in multicandidate probabilistic spatial voting",
Public Choice 98:59-82.
Locke, J. (1690), Two Treatises of Government. Republished in 1988, edited and with an introduction
by P. Laslett (Cambridge University Press, Cambridge).
Lowell, L. (1896), Governments and Parties in Continental Europe (Houghton Mifflin, Boston).
Madison, J. (1999), Writings, edited and with an introduction by J. Rakove (Library of America, New
York).
454 N. Schofield

McKelvey, R.D. (1976), "Intransitivities in multidimensional voting models and some implications for
agenda control", Journal of Economic Theory 12:472-482.
McKelvey, R.D. (1979), "General conditions for global intransitivities in formal voting models",
Econometrica 47:1086-1112.
McKelvey, R.D. (1986), "Covering, dominance and institution-free properties of social choice", American
Journal of Political Science 30:283-314.
McKelvey, R.D., and N. Schofield (1986), "Structural instability of the core", Journal of Mathematical
Economics 15:179-198.
McLean, I., and F. Hewitt (1994), Condorcet: Foundations of Social Choice Theory (Edward Elgar,
Aldershot, UK).
Mellen, S. (1943), "The German people and the post-war world", American Political Science Review
63:40-56.
Michael, E. (1956), "Continuous selections I", Annals of Mathematics 63:361-382.
Mill, J.S. (1861), Representative Government. Reprinted in 1975, in: Three Essays, edited and with an
introduction by R. Wollheim (Oxford University Press, Oxford).
Miller, N. (1980), "A new solution set for tournaments and majority voting: further graph-theoretic
approaches to the theory of voting", American Journal of Political Science 24:68-96.
Nash, J.F (1950), "Equilibrium points in N-person games", Proceedings of the National Academy of
Sciences USA 36:48-49.
Nash, J.F (1951), "Non-cooperative games", Annals of Mathematics 54:286-295.
Norberg, K. (1994), "The French fiscal crises of 1788 and the financial origins of the revolution of
1789", in: P. Hoffman and K. Norberg, eds., Fiscal Crises, Liberty and Representative Government,
1450-1789 (Stanford University Press, Stanford) pp. 253-298.
North, D., and B. Weingast (1989), "Constitutions and commitment: the evolution of institutions governing
public choice in 17th century England", Journal of Economic History 49:803-832.
Parthasathy, K.R. (1967), Probability Measures on Metric Spaces (Academic Press, New York).
Peleg, B. (2002), "Game-theoretic analysis of voting in committees", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 8,
this volume.
Plott, C.R. (1967), "A notion of equilibrium and its possibility under majority rule", American Economic
Review 57:787-806.
Poole, K., and H. Rosenthal (1991), "Patterns of congressional voting", American Journal of Political
Science 35:228-278.
Poole, K., and H. Rosenthal (1997), Congress: A Political Economic History of Roll Call Voting (Oxford
University Press, Oxford).
Popper, K. (1945), The Open Society and its Enemies (Routledge, London).
Popper, K. (1988), "The open society and its enemies revisited", The Economist 307:19-22.
Quinn, K., A. Martin and A. Whitford (1999), "Voter choice in multiparty democracies: a test of
competing theories and models", American Journal of Political Science 43:1231-1247.
Rabier, J.-R., and R. Inglehart (1981), Eurobarometer II April 1979. The Year of the Child in Europe
(Inter-University Consortium for Political and Social Research, Ann Arbor, Michigan).
Rae, D. (1967), The Political Consequences of Electoral Laws (Yale University Press, New Haven).
Rae, D. (1969), "Decision rules and individual values in constitutional choice", American Political
Science Review 63:40-56.
Riker, WH. (1953), Democracy in the United States (Macmillan, New York).
Riker, W.H. (1964), Federalism: Origin, Operation, Significance (Little Brown, Boston).
Riker, W.H. (1980), "Implications from the disequilibrium of majority rule for the study of institutions",
American Political Science Review 74:432-446. Reprinted in 1982, in: P. Ordeshook and K. Shepsle,
eds., Political Equilibrium (Kluwer-Nijhoff, Boston) pp. 3-24.
Riker, W.H. (1982), Liberalism Against Populism (San Francisco, Freeman).
Riker, W.H. (1984), The Development of American Federalism (Kluwer, Boston).
Ch. 9: Representative Democracy as Social Choice 455

Riker, WH. (1986), The Art of Political Manipulation (Yale University Press, New Haven, CT).
Riker, W.H. (1996), The Strategy of Rhetoric: Campaigning for the Ratification of the Constitution (Yale
University Press, New Haven, CT).
Saari, D.G. (1997), "The generic existence of a core for q-rules", Economic Theory 9:219-260.
Savage, L. (1954), The Foundations of Statistics (Dover, New York).
Schama, S. (1989), Citizens: A Chronicle of the French Revolution (Knopf, New York).
Schattschneider, E. (1942), Party Government (Farrar and Rinehart, New York).
Schofield, N. (1972), "Ethical decision rules for uncerrtain voters", British Journal of Political Science
2:193-207.
Schofield, N. (1977), "Transitivity of preferences on a smooth manifold", Journal of Economic Theory
14:149-172.
Schofield, N. (1978), "Instablity of simple dynamic games", Review of Economic Studies 45:575-594.
Schofield, N. (1985), Social Choice and Democracy (Springer, Heidelberg).
Schofield, N. (1993), "Political competition and multiparty coalition governments", European Journal of
Political Research 23:1-33.
Schofield, N. (1999a), "The Heart of the Atlantic Constitution", Politics and Society 27:173-215.
Schofield, N. (1999b), "The Cl topology on the space of smooth preference profiles", Social Choice
and Welfare 16:347-373.
Schofield, N. (1999c), "A smooth social choice method of preference aggregation", in: M. Wooders,
ed., Topics in Mathematical Economics and Game Theory: Essays in Honor of R. Aumann, Fields
Institute Communications, Vol. 23 (American Mathematical Society, Providence, RI) pp. 213-232.
Schofield, N. (1999d), "The heart and the uncovered set", Journal of Economics: Zeitschift fir
Nationaldconomie, Supplement 8:79-113.
Schofield, N. (2000), "Constitutional political economy: on the possibility of combining rational choice
theory and comparative politics", Annual Review of Political Science 3:277-303.
Schofield, N. (2001), "Constitutions, voting and democracy: a review", Social Choice and Welfare
18:571-600.
Schofield, N. (2002a), "Evolution of the constitution", British Journal of Political Science 32:1-20.
Schofield, N. (2002b), "Quandaries of war and of union in North America: 1763 1861", Politics and
Society 30:5-49.
Schofield, N., and R. Parks (2000), "Nash equilibrium in a spatial model of coalition bargaining",
Mathematical Social Sciences 139:133-147.
Schofield, N., and I. Sened (2002), "Local Nash equilibrium in multiparty politics", Annals of Operations
Research, forthcoming.
Schofield, N., A. Martin, K. Quinn and A. Whitford (1998), "Multiparty electoral competition in the
Netherlands and Germany: a model based on multinomial probit", Public Choice 97:257-293.
Schofield, N., I. Sened and D. Nixon (1998), "Nash equilibrium in multiparty competition with 'stochastic'
voters", Annals of Operations Research 84:3-27.
Shepsle, K.A. (1979), "Institutional arrangements and equilibrium in multidimensional voting bodies",
American Journal of Political Science 23:27-60.
Shepsle, K.A., and B.R. Weingast (1981), "Structure induced equilibrium and legislative choice", Public
Choice 37:503-519.
Smith, A. (1776), An Inquiry into the Nature and Cause of Wealth of Nations. Reprinted in 1976 (Oxford
University Press, Oxford).
Chapter 10

SOCIAL WELFARE FUNCTIONALS AND INTERPERSONAL

COMPARABILITY *

CLAUDE d'ASPREMONT
CORE, Louvain University, Louvain-la-Neuve, Belgium

LOUIS GEVERS
University of Namur and CORE

Contents

Abstract 461
Keywords 461
1. Introduction 462
2. Social welfare fiunctionals and related concepts 465
2.1. Definitions 465
2.2. Examples 468
2.2.1. Imposed SWFLs 468
2.2.2. Dictatorship 468
2.2.3. Maximin and leximin 469
2.2.4. Pure utilitarianism, weighted utilitarianism and relative utilitarianism 470
2.2.5. Weighted rank utilitarianism and the generalised Gini family 471
2.2.6. Nash's bargaining solution 472
2.2.7. Borda's method of voting 472
2.2.8. Majority voting 473
2.3. Domain interpretation 474
2.4. Some related concepts 476
3. Axioms and their use 477
3.1. Preliminary 477
3.2. Invariance axioms 478
3.2.1. Comparisons of evaluation levels 479
3.2.2. Comparisons of evaluation differences 481
3.2.3. Comparisons of evaluation indicator ratios 486
3.2.4. Other invariance axioms 487

* We thank an anonymous referee for many useful remarks and W Bossert, A. Dhillon, M. Fleurbaey,
E Gaspart, F Maniquet, J.E Mertens, K. Suzumura and J. Weymark for helpful discussions.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K Sen and K. Suzumura
( 2002 Elsevier Science B. V All rights reserved
460 C. d'Aspremont and L. Gevers

3.3. Other information-filtering axioms and formal welfarism 489

3.3.1. Pareto indifference, Independence and Neutrality 489
3.3.2. Formal welfarism and invariance properties 494
3.3.3. Separability, Continuity and Convexity properties 495
3.3.4. Alternative approaches to formal welfarism 497
3.4. Pareto dominance principles and weak welfarism 499
3.5. Equity axioms 503
4. Independence and invariance-based characterisations 507
4.1. Restricting interpersonal level comparability 507
4.2. Full comparability: from cardinal to ordinal measurability 513
4.2.1. No Separability 514
4.2.2. A modicum of Separability 516
4.2.3. Full Separability 517
4.3. Homothetic vs. translatable social welfare functionals 519
5. Discarding neutrality or invariance 520
5.1. Uncertainty and risk: from Harsanyi to Relative Utilitarianism 521
5.2. On some egalitarian social rankings 531
6. Conclusion 534
References 537
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 461

Abstract

This chapter reviews the SWFL approach to social choice. It does not attempt to be a
complete and systematic survey of existing results, but to give a critical assesment
of the main axioms and their role in filtering the ethically relevant information,
in particular the measurability and comparability properties of individual evaluation
functions. Social welfare functionals are defined formally together with closely related
concepts. After adducing a good number of examples, we elaborate on the meaning of
the SWFL domain of definition and we sketch some alternative approaches. Several
types of axioms are considered; some of them are used to filter the relevant information
while others express collective efficiency or equity requirements. Then, to illustrate the
various tradeoffs among these axioms, selected characterisation results are presented;
most of them are cast in what we call the formally welfarist framework. Finally, we
have assembled some other characterisations which eschew either invariance properties
or the formally welfarist framework. We discuss the treatment of two sets of social
alternatives endowed with an enriched structure, viz. the set of classical exchange
economies and the complete set of lotteries one can define on an abstract set of
pure alternatives. As an introduction to the latter discussion, we elaborate on the
difficulties raised by social evaluation when risks and uncertainty are taken explicitly
into account.

Keywords

social welfare functionals, social welfare orderings, invariance axioms, utilitarianism,

welfarism

JEL classification: D71, D81

462 C dAspremont and L. Geuers

1. Introduction

Insofar as it probes the foundations of political constitutions, social choice theory deals
with the relationship collective decisions or preferences ought to bear with individual
preferences. Arrow (1951) launched the first systematic attack on this problem at
a formal level. He aimed at generality, paid no attention to the specifics of usual
economic models, and assumed that individual preferences could be of any shape
whatsoever.
Although the set of problems raised by political decisions and those raised by
social-evaluation judgements share the same basic formal structure, they ought to be
distinguished sharply from each other. Political decisions are usually arrived at in
groups, and individual preferences cannot be filtered; they must be accepted as they
are while manipulation attempts cannot be excluded from communication channels
with the center. Therefore, such things as election procedures and assembly rules are
likely to be of central importance. Game-theoretic equilibrium concepts have been
found relevant in this context, and equilibrium correspondences are a key link between
individual preferences and collective decisions. Political equilibrium may often be
interpreted as some form of compromise; in case new information becomes available
before it is enacted, one does not expect the revised compromise to evolve from its
predecessor in a very rational fashion.
In contrast, when an ordinary citizen attempts to take the standpoint of an ethical
observer in order to formulate social-evaluation judgments, perhaps as an input towards
establishing eventually an optimal voting strategy, he or she is in a position to treat new
information more rationally; his or her attention is likely to be focused on the content
of social outcomes and their consequences, at least as much as on the procedures
followed to arrive at them. Moreover, the appropriate summary statistics describing
what is ethically relevant from each individual's viewpoint do not necessarily have a
utility interpretation, unless one is persuaded by the welfarist tradition. Even if this
is the case, there is nothing schizophrenic when a voting procedure is approved by
someone who is morally objecting to the consequences of some piece of legislation
adopted in accordance with it. At least two kinds of reason may account for this.
On the one hand, voting procedures have a more permanent character than ordinary
legislative output; on the other hand, the ethical observer may be relying on subjective
prior information or on external information without paying a great deal of formal
attention to manipulation.
Between the political body and the common citizen, public officials occupy
an intermediate position: each of them should ideally be concerned with social
evaluation, but, as Bergson (1954) writes, "the values to be taken as data are
not those which might guide the official if he were a private citizen ... His one
aim in life is to implement the values of other citizens as given by some rule
of collective decision-making". Bergson stresses that most pre-Arrovian welfare
economists were discussing the ordinary citizen's problem, whereas Arrow's analysis
of social choice is more relevant for public officials. Arrow (1963, p. 107), from
Ch. 10: Social Welfare Functionals and InterpersonalComparability 463

whom we borrow the above quotation, marks his agreement with this interpretation.
Indeed, his exclusive reliance on lists of individual preferences as informational basis
of collective preference may be well suited for discussing the formal aspects of
political decisions. But their economical use can hardly accommodate any concern
for equity. For instance, when comparing income distributions, this concern might
be expressed by the following doctrine: (1) individual utilities are all which matters
for social evaluation, (2) they are concave functions of income, which is their single
argument, and (3) their sum ought to be maximized by society. This simplified
utilitarian approach is out of reach of the seminal Arrovian model. Indeed, it is
based on individual preference relations which do not lend themselves to ethically
meaningful interpersonal comparisons. Any one of two features may account for
this shortcoming: too little structure is imposed on the set of social decisions or
alternatives, and moreover, social preference between any two alternatives is required
to depend only on individual preferences restricted to this pair. The latter principle
was called by Arrow independence of irrelevant alternatives. It greatly contributes to
informational parsimony. However, if it is weakened or deleted, the formal construction
of interpersonal utility comparisons can be obtained as the by-product of a voting
procedure, but the social ranking associated with it is unlikely to be adequate for
social evaluation. A more meaningful construction can also be found in the literature;
it relies on a set of alternatives endowed with a structure which is richer and less
abstract: for instance, uncertainty is made explicit and social preferences are required
to admit of an expected utility representation, or context-specific domain restrictions
and equity arguments are brought in, as in the recent literature on axiomatic allocation
theory.
Quite some years before these developments, Sen (1970) opened several paths
branching out of Arrow's trail. Along the one we plan to follow, Sen reconsiders the
problem of social evaluation from the viewpoint of an ethical observer who might
be a private citizen. In contrast with the more recent approach we just sketched, he
refrains from giving the set of alternatives any specific structure, and his innovation
pertains to the informational basis of social-evaluation judgments: it is assumed to
consist of all logically possible lists of individual utility functions. Except for the
hardly significant case of individual preferences failing to be numerically representable,
the above formal approach, which is crystallized in the concept of social welfare
functional (SWFL), is more general than Arrow's, since individual utility levels or
gains can be considered interpersonally comparable a priori, unless specific axioms
restrict, for informational or moral reasons, the ethical observer's discriminating ability.
However, generality involves a cost: there is a preparation stage, at which the ethical
observer is to select an adequate list of a priori comparable utility representations
of individual preferences, about which others do not necessarily agree. In the simple
example already alluded to, the social evaluation of individual income vectors is
very dependent on the degree of inequality aversion embodied in the concavity of
each individual utility function, and thoroughly rational observers can be in total
disagreement about this value judgment. Furthermore, SWFL theory can hardly help
464 C. d'Aspremlont and L. Gevers

them solve their conflict, even though it does often provide assistance for finding out
which value judgments are compatible and which are not. For example, it contains
much clarification of the debate opposing utilitarianism with competing principles,
but it cannot prescribe any complete ready-made recipe for every social-evaluation
problem.
The major part of our survey is devoted to a review of developments of Sen's
concept. The individual utility interpretation of its informational basis has been
criticized by Sen himself and by several political philosophers who stress that
social-evaluation judgments should not be concerned exclusively with arbitrating the
individuals' conflicts of interests as they are narrowly modelled by positive economic
theories. But this criticism does not necessarily diminish the usefulness of the formal
SWFL concept. Indeed, the latter can be reinterpreted as an application of multi-
objective decision theory to the ethical observer's problem. It attempts to make explicit
the link between the social evaluation of an alternative and its appraisal from the
viewpoint of every individual in turn; the latter appraisals may themselves be linked
with individual preferences, but they do not necessarily duplicate them. For instance,
they could be represented by numerical indicators summarizing the individuals' doings
and beings or their life expectancy or their set of opportunities. Stretching the use of an
expression sparsely referred to in the literature, we shall call them individual evaluation
functions or indicators. Aggregating them over the set of individuals is our main task
in the present chapter.
Thus, the informal preparatory stage required before making use of the SWFL
apparatus is not without advantage, as it makes for a versatile tool. Since the set of
possible alternatives lacks any particular structure, the SWFL can accommodate the
needs of an ethical observer who is more interested in appraising the instrumental
value of social rules and political institutions than in focusing on single outcomes.
It can also be found useful by someone persuaded by the piecemeal engineering
approach of axiomatic allocation theorists; as they confine their analysis to a variety
of specific economic environments, the SWFL can be thought of as a kind of residual
tool for evaluating situations not yet adequately covered by the existing theoretical
corpus.
However, it is also legitimate to adopt a positive interpretation of this body of
literature: most people may be assumed to take up at least occasionally the position of
an ethical observer. Empirical studies of professed evaluation judgments or decisions
1.
have started cropping up, and some of them are based on laboratory experiments
A theory capable of structuring social-evaluation judgments is not only interesting in
its own right but it can also usefully interact with the empirical part of an ambitious
research program. There is a clear analogy with the theory of individual decisions
under uncertainty.

l See for instance Bar-Hillel and Yaari (1984).

Ch. 10: Social Welfare Functionals and InterpersonalComparability 465

To sum up, as they are defined by Sen (1970), social welfare functionals are maps
determining the social ordering of the set of alternatives with help of a complete
list of individual numerical indicators which are not necessarily interpreted as utility
functions 2 .
This chapter reviews the SWFL approach to social choice. It does not attempt
to be a complete and systematic survey of existing results (this has been done
elsewhere3), but to give a critical assessment of the main axioms and their role
in filtering the ethically relevant information, in particular the measurability and
comparability properties of individual evaluation functions. In Section 2, social welfare
functionals are defined formally together with closely related concepts. After adducing
a good number of examples, we elaborate on the meaning of the SWFL domain
of definition and we sketch some alternative approaches. Section 3 is essentially
devoted to the analysis of axioms considered in isolation; some of them are used
to filter the relevant information while others express collective efficiency or equity
requirements. Finally, to illustrate the various tradeoffs among these axioms, selected
characterisation results are presented in Section 4; most of them are cast in what we
call the formally welfarist framework. In Section 5, we have assembled some other
characterisations which eschew either invariance properties or the formally welfarist
framework. We discuss at length the treatment of two sets of social alternatives
endowed with an enriched structure, viz. the set of classical exchange economies and
the complete set of lotteries one can define on an abstract set of pure alternatives.
As an introduction to the latter discussion, we elaborate on the difficulties raised by
social evaluation when risks and uncertainty are taken explicitly into account. Section 6
concludes.

2. Social welfare functionals and related concepts

2.1. Definitions

A society made up of a finite set N = {1,..., n} of individuals faces a set of possible

social decisions or alternatives X consisting of at least three elements. Whether we
interpret X as the set of all conceivably feasible alternatives or as the set of actually
feasible social decisions in a narrowly defined situation, perhaps after excluding some
alternatives because they violate legal or human rights, it is our task to evaluate the
respective merits of its elements and to rank them from society's viewpoint. Any

2 A similar point of view is defended in Mongin and d'Aspremont (1998).

3 The most recent survey, closest to this chapter, is Bossert and Weymark (2000). See also Sen (1977,
1979, 1986a), Blackorby et al. (1984), Lockwood (1984), d'Aspremont (1985), Moulin (1988), Roemer
(1996), Mongin and d'Aspremont (1998).
466 C. d'Aspremnont and L. Gevers

ranking (or preference ordering) of X is required to be rational, i.e. a complete and

transitive binary relation over X:
Rationality of R.
Vx,y E X, xRy or yRx, and Vx,y,z X,
xRy and yRz implies xRz.

The set of all rankings one can define over X is denoted R. For any R E 7?
and x,y E X, xly means indifference (xRy and yRx) and xPy strict preference (xRy
and not yRx). The simplest ranking is the trivial one: Vx,y X, xly. In general,
however, a ranking R may be a complicated object. Its handling is often facilitated
if it admits of a faithful translation in the language of real numbers. In more formal
terms, R is said to be representableby a numerical function u defined on X if and only
if Vx,y e X, xRy 4 u(x) > u(y). To avoid pre-committing the interpretation of u, we
call it an evaluation function. To ensure its representation, we sometimes (when X is
nondenumerable and has a topological structure) assume in addition

Continuity of R:
Vx,y E X, the sets {x E X I xRy} and {x E X yRx} are closed in X.
In the sequel, any pair (x, i) will be called a station if it is an element of X x N.
Social evaluation has to rest somehow on information pertaining to this set, either
directly or through individual evaluations. Any label x E X is assumed to convey
directly a full description of all ethically relevant aspects of the social decision it
designates, except for the other elements involved in the construction of the social
ranking. On the other hand, the observer is assumed to be fully informed of individual
evaluations by a real-valued function defined on X x N and called hereafter individual
evaluation profile or, for short, profile. A typical profile is denoted U; if X is finite,
it can also be thought of as an IXI x NI matrix with generic element U(x, i) lying at
the intersection of row U(x, ) = Ux with column U(., i) = U. In any case, Ui will be
called individual i's evaluation function or indicator, whereas U, the restriction of U
to {x}, will be called an individual evaluation vector 4 of x, a point in evaluation space
sN,
9 where 9l stands for the real line.
Since we may want to accommodate every profile in a universal set, we define

U = {U U: X x N - 91}.
We are now ready to define formally the main subject of this survey. Given any
profile U in a subset5s D C U, we are to recommend the social ranking of X that

4 In this chapter, vector inequalities are distinguished as follows: if a = (al,..., a,) and b = (bl,..., b,,),
(1) 'a < b' means 'for every i, ai < bi', (2) 'a <; b' means 'for every i, ai < bi and there exists k such
that ak < bk' and (3) 'a < b' means 'for every i, ai < bi'.
5 In this chapter, the notation 'C' means 'is a subset of or equal to' and not 'is a strict subset of'.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 467

is best adapted to it. Following Sen (1970), who first fully articulated this concept, a
socialwelfarefunctional (SWFL) is a map F: D --+ R with generic image Ru = F(U).
If x is ranked socially at least as high as y whenever the relevant profile is U, we shall
write xRuy.
According to the context, various assumptions may apply to the domain D of F.
Usually, they ensure that (1) D is not a singleton and (2) the set 7-(X, D) = {r E N
I 3x C X, 3U C D such that U = r} fills the whole individual evaluation space, i.e.
H(X, D) = N. The strongest one is:

Domain universality (UD). 1) = U.

But, in many cases this will be too demanding. A weaker assumption will then be
used 6 :

Domain attainability(AD):
Vu,v,w 9l ,
3x,y,z X, 3UECD such that U=u, Uy = v and U = w.

Interesting sufficient conditions for Domain attainability have been developed,

adopting what we shall call the single preference profile approach to formal
welfarism 7, and following the work of Hammond (1976b), Kemp and Ng (1976), Parks
(1976), Pollak (1979) and Roberts (1980b). To make the approach plausible, X is given
a modicum of structure: for instance, Pollak (1979) assumes that X can be partitioned
in separate subsets; within each subset, the alternatives differ only with respect to the
allotment of one transferable private good among individuals, and the latter entertain
self-oriented preferences with at least three distinct levels of satisfaction. Although
individual preferences are assumed to remain fixed over the domain ofF, every n-tuple
of utility functions representing them is admitted. We call this a Pollak domain.
In some contexts, zero appears naturally as the greatest lower bound (resp. least
upper bound) of every individual evaluation function so that 7-(X, D) = 9K+ or 9++
(resp. 9N or 9N ). To illustrate, we rely on what can be called the pure income
distribution model. Our n agents are assumed to be capable of consuming only one
single good in non-negative amounts. As every individual's consumption set is 9,
society's possibility set is 9'+N. It is often assumed that each agent has an equal right
to whatever quantity of good is available and that none can claim special treatment
on account of his/her individual characteristics. In this setup, it seems natural to adopt

6 This condition is called "unrestricted individual utility profile" in d'Aspremont (1985) and "relative
attainability" in d'Aspremont and Mongin (1997).
7 Reference to the word "preference" is often omitted in the above expression. This practice is
appropriate for the Arrovian social welfare function context, but it can be misleading in our SWFL
context.
468 C. d'Aspremont and L. Geuers

a specific single preference profile, such that every agent is both self-oriented and
insatiable. If we further assume that any n-tuple of utility representations consistent
with the preference profile is relevant for our family of problems, we are facing an
N.
example of a Pollak domain, and 7t(X, D) = E9 Yet, in the absence of risk and
uncertainty, one can argue that measuring utility by consumption is the only appropriate
convention, since it is the only one about which hard evidence can be adduced.
Moreover, we attach importance to the units of measurement of consumption. If this
argument is accepted, only one profile of identical evaluations is considered valid and
H(X, D) = 9iV, where D is a singleton, so that the SWFL is degenerate. The problem
of ranking income distributions from a social viewpoint has elicited a whole body of
literature, which is surveyed by Dutta (2002).
Among others, we shall be interested in SWFLs treating individuals according to
the rank they occupy in the hierarchy of individual evaluation levels, a hierarchy
which depends on both the profile at hand and the alternative being appraised.
If 7-(X, D) = 9 N, it will prove convenient to restrict some properties to the subset
of well-ordered individual evaluation vectors. For future reference, we register it here
as N = < W2 < ... < W l < W,}.
9N I Wt {W

2.2. Examples

Prior to elaborating further on the significance of the concepts used in the above
definition, we proceed by describing some useful examples of SWFLs.

2.2.1. Imposed SWFLs

One can simply deny that the individual evaluation profile has any relevance for
establishing a social ranking. The rest of the information concerning the alternatives,
which is subsumed under the description of X, however minutely detailed it happens
to be, is deemed sufficient to determine the social ranking. The latter is thus constant
once X is given and the SWFL is said imposed. If our reader is persuaded by this
extreme opinion, it is unlikely that he or she will be enthusiastic about the rest of this
chapter, where profiles play a leading role.

2.2.2. Dictatorship

Another hardly more appealing family of SWFLs eschews any interpersonal compar-
ison among the columns of any given profile: each one is based on an exogenously
determined hierarchy among individuals, and the social ranking always reflects the
relative individual evaluation of the agent enjoying the highest hierarchical rank,
unless he or she values equally the alternatives under consideration. If this occurs,
the relative social evaluation endorses the relative individual evaluation of the agent
enjoying the next position, and so forth, until the subset is exhausted. Eventually,
social indifference between two alternatives occurs only if they are equivalent from
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 469

every agent's viewpoint. To each permutation of the set of agents, there corresponds
an exogenous hierarchy among them and an SWFL defined as above. Each of these
n! SWFLs is called lexicographically dictatorial. More generally, we shall say that
an SWFL displays weak dictatorship if there exists an individual such that the social
ranking always mimics his or her strict preference. This privilege cannot be based
on talent or merit as it would become endogenous. This SWFL is characterized in
Section 4.1.
Anyone favoring impartial decision procedures would consider such an SWFL
as grossly unfair. This negative judgement does not necessarily extend to its social
consequences, for instance if D consists of various representations of a single
preference profile and the dictator's individual preference relation is sufficiently
altruistic or equity-minded. If absolute power does not corrupt the ethical stand of
the dictator, he or she could even endorse the drastic form of egalitarianism displayed
in our next example.

2.2.3. Maximin and leximin

A way to introduce egalitarian consideration is to focus on the least favoured individual

[see Kolm (1966) and Rawls (1971)] and, between two alternatives x,y E X, to
prefer strictly x to y whenever mini U(x, i) > mini U(y, i). Based on this property,
two SWFLs can be defined, corresponding to two ways to get a complete ordering
for every profile U. One is the maximin principle requiring that, for any two
alternatives x,y E X, x be declared at least as good socially as y if and only if
mini U(x, i) > mini U(y, i). The other is the leximin rule (an expression coined
by Sen). To define it, we associate with every given profile another one that is
well ordered for each alternative: for each x X, we permute the elements of the
row Ux and obtain (U(x, i(l)), U(x, i(2)),..., U(x, i(n))), where i(k) is the individual
who happens to be the kth-worst off at x (in formal terms, i(.) is a permutation such that
U(x, i(1l)) < U(x, i(2)) < ... < U(x, i(n)). Then xRuy according to the leximin SWFL
if and only if the lexicographically dictatorial SWFL, applied in the natural order
(1, 2..., n) to the associated well-ordered profile, stipulates that x is at least as good
as y socially. Characterizations of leximin are studied in Subsections 4.2.3 and 5.2.
More extreme egalitarian principles could be thought of: they would justify a sheer
reduction in individual utility without benefit to anyone else on the ground that this
pruning operation leads to smaller interpersonal utility differences. Although this
cannot happen if the maximin or leximin rules are adopted, both principles have been
criticized because they recommend to give absolute priority to the slightest gain of
the least advantaged agent over potentially very significant losses incurred by all other
agents.
As a concluding observation, we note that the leximin SWFL has meaning only
if individual evaluation levels can be compared interpersonally. This is in complete
contrast with our previous examples, and also with our next SWFL, which relies
470 C. d 'Aspremont and L. Geuers

perhaps on the most popular social evaluation formula since the days when Bentham
(1789) started propagating it.

2.2.4. Pure utilitarianism,weighted utilitarianismand relative utilitarianism

The SWFL is said to be purely utilitarianif and only if Vx,y E X, VU D),
M n
xRuy 4C U(x, i) > E U(y, i).
i- i=-

Thus, xRuy if the sum of algebraic gains in individual evaluation is non-negative as

society moves from y to x. We can also express the latter condition as requiring the
total evaluation gain of the gainers to be at least equal to the total loss of the losers.
Summing individual evaluation indicators or their first differences makes sense only
if their units of measurement can be meaningfully compared across persons.
The family of rules called weighted utilitarianismalso requires that the units of
measurement of individual evaluation indicators be comparable with each other; as its
qualifier suggests, it is parameterised by means of a vector of n individual weights,
, e 9 \0}, and is such that, Vx,y E X, VU D,
n n

xRuy X E i U(x, i) > i U(y, i).

i=l i=l

It is natural to assume all weights to be positive. If they are all equal, we are back
to pure utilitarianism. Related characterizations can be found under Subsections 4.1,
4.2 and 4.3. Readers familiar with the theory of individual decision under risk and
uncertainty will have noticed the close relation between expected utility and utilitarian
principles. We elaborate on this topic in Subsection 5.1, where we also discuss relative
utilitarianism, to which we turn next.
For this purpose, we restrict the SWFL domain D so that every individual evaluation
function U always displays both a maximum and a minimum, denoted hereafter i
and iii, respectively. We also let every individual evaluation function in every profile
in the SWFL domain undergo a positive affine transformation which is called in the
sequel a Kaplan normalisation8; in other words, to every U D, we associate a
normalised profile Ku defined on the set of stations as follows:

VU E D, V(x,i) E X x N, Ku (x,i)=
We are now in position to define relative utilitarianism:VU C D, Vx,y X,
n n

xRuy X ZKu(x, i) > Ku(y, i).

i-I i=1

We have described in effect a two-stage procedure. As a result of the first-stage

normalisation, the range of each Ku(, i) extends from 0 to . At stage two, the

8 Arrow (1963, p. 32) mentions that this normalisation was suggested to him by Kaplan.
Ch. 10: Social Welfare Functionals and InterpersonalComparability 471

utilitarian summation formula is applied to the first-stage output. Whether the units
of the individual evaluation indicators belonging to any given profile are comparable
or not is immaterial, because the normalisation process implies a new endogenous
calibration.
Arrow criticizes relative utilitarianism for the fact that adding an alternative which
everyone considers worst may play havoc at the top of the social ranking. This criticism
seems to lose its cogency if the SWFL domain is based on a single preference profile
and if there exists such a thing as the worst imaginable outcome from the viewpoint of
every individual, which is an element of X even though it may involve an unfeasible
exchange of individual characteristics.
More generally, as we hinted in Section 1, the utilitarian formula is consistent with
a very wide array of ethical attitudes towards inequality that may be inherent to the
description of alternatives in X, for instance when X consists of various allocations
of private goods to a set of otherwise identical individuals. In this case, the degree
of social inequality aversion is dictated to utilitarians by the degree of concavity of
each individual evaluation indicator, i.e. by the decrease of successive first differences
of each U with respect to income. The selection of a particular profile as ethically
relevant in a given social conflict is thus a delicate matter, even if one is persuaded
by relative utilitarianism. On the other hand, utilitarians have been aptly criticized
for being insensitive to individual evaluation levels and their distribution, since only
first differences with respect to this metric matter in their formulation. See e.g.,
Theorem 1.2 in Sen (1973, p. 20). We elaborate on this in subsection 5.1. The example
we discuss next may be immune from the latter bias.

2.2.5. Weighted rank utilitarianismand the generalised Ginifamily

The family of SWFLs that goes by the name of weighted rank utilitarianism puts all
individuals on the same footing, like the maximin principle and pure utilitarianism,
both of which it encompasses. To parameterise it, we rely on a vector of n non-
negative weights; each weight is associated with a particular rank k and this is
reflected in our generic notation k. Hence, for weighted rank utilitarianism, there
exists A E 91f\{0} such that Vx,y c X, VU D,
n n

RUY YE Sk U(x, i(k)) > E k U(y, i(k)).

k=l k=l

In the above formula, i(k) denotes again the name of the individual whose evaluation
indicator is the kth-smallest for the social decision under study. Equity requires to
treat any lower rank at least as well as any higher rank, by allocating to the former a
weight at least equal to the weight allocated to the latter. By adding the clause Al > 0,
Vk C N\{n}, Ak > k+l > 0, we satisfy this requirement and we define the generalised
Gini family. The importance of this family of SWFLs is illustrated in our discussion
of social decisions under risk and uncertainty (Subsection 5.1).
472 C. d 'Aspremont and L. Gevers

2.2.6. Nash s bargainingsolution

Some aggregation rules exhibit a multiplicative form. An example is the symmetric

Nash bargainingsolution [following Nash (1950)], which is defined relative to some
constant status quo point xo in X and is such that, Vx,y E X, VU E D having Vi e N,
U(x, i) > U(xo, i) and U(y, i) > U(xo, i),
ni- i-I

xRuy X J [U(x, i) - U(xo, i)] > 1 [U(y, i) - U(Xo, i)].

We remark that our definition says nothing about the social ranking of alternatives
weakly less preferred than x by one or more individuals. The social ranking
corresponding to the weighted version is not complete either; to wit: for some
) E 9t\{0} and x E X such that Vx,y X, VU C D having Vi N,
U(x, i) > U(xo, i) and U(y, i) > U(xo, i),
n n

xRuy X ] [U(x, i) - U(xo, i)i > J [U(y, i) - U(xo, i)'i

i- i=l

A characterization is offered in Subsection 4.3.

2,2.7. Borda method of voting

Various methods of voting may be relied upon to define SWFLs. We give two
examples. We begin with the Borda method, which cannot be used unless X is finite.
To define it, we suppose first that D is such that every individual evaluation indicator
represents a strict ordering on X. Thus, D C {U E U ii N, Vx,y X,
U(x, i) U(y, i)}. In this case, we let B(x, i) denote the number of alternatives of X
which are less preferred than x by i for the given profile U cE . Formally,

Vx E X, VU E , vi E N, B(x, i) = #{y XI U(x, i) > U(y, i)}

In other words, B(x, i) registers the number of victories of x when it is pitted

successively against every other alternative. As we proceed to the social ranking, we
shall maintain in the same spirit and put all individuals on the same footing. We shall
say that x is ranked socially higher than y if and only if the total number of victories
scored by x is greater than the corresponding number for y:
n n

Vx,y EX, U E D, xRuy a ZB(x, i) > B(y,i).

i=l i=

This voting method extends easily if the domain of the SWFL allows for individual
indifference among alternatives. For any xo E X, if there is no alternative indifferent
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 473

with x, the definition of B(xo, i) is the same as above. If the indifference curve
through xo consists of, say, k distinct alternatives x, xl, ... , Xkl scoring by definition
the same number g of victories, then we define for every h, 0 < h < k - 1,
B(Xh, i) = g +
We move next to the family of generalised Borda methods. Any member can be
obtained by selecting an increasing transformation and by applying it to every B(., i)
for every individual and for every profile in D. As before, a representation of the social
ranking is obtained by summing the transformed numbers. All these voting rules share
two features: (1) any two distinct evaluation functions representing the same ordering
get mapped into the same individual representation, and (2) if any two alternatives are
adjacent in any i's evaluation, the corresponding representation difference seems to be
the result of a mechanical process which is foreign to equity considerations. Indeed, the
social ranking is fully determined by the positions occupied by the alternatives in the
individual rankings. The following example is tailored to criticize the Borda method,
but it could have been fitted to criticize any other member of the generalised family.
Suppose N consists of three selfish people who have an equal title to a unit cake. The
set of alternatives is defined as some finite subset of the set of all non-negative triples
such that the first individual gets less than one third, whereas the other two people
share the balance evenly:

X C {(Xi,x2,X3) G , <X2 =x3 (l -X 1)}.

Following the Borda method, this is enough to conclude that xPuy if and only if
xl < yl, i.e. the poor become poorer and the rich richer, because they all have self-
centered preferences. Although it may appear a decent aggregation procedure from
the political viewpoint, the Borda method is a nonstarter in the social evaluation
competition. Relative utilitarianism, which is not mechanically linked to intermediate
positions in individual rankings, seems to fare better in this respect, because the Kaplan
normalisation it implies leaves unchanged the concavity of the original individual
evaluation functions.

2.2.8. Majority voting

According to the familiar method of majority voting, an alternative is socially

ranked above another one if the number of individual evaluations ranking the former
strictly above the latter exceeds the number of strictly opposite evaluations. Formally,
Vx,y c X, VU e D, xRuy if and only if

#{i E N I U(x, i) > U(y, i)} > #{i C N I U(y, i) > U(x, i)}.

Ever since Condorcet's days it has been known that profiles must be selected with care
if we want to be sure that the corresponding social rankings are transitive. As a rule,
they are not the same as the social rankings obtained by the Borda method for the same
474 C. d Aspreniont and L. Gevers

profiles, unless IX I = 2. However, the two social rankings coincide also in the situation
we conjured up to criticize the implications of the Borda method. In conclusion, even
though the method of majority voting is often considered superior to its Borda rival as
a voting procedure when it results in a transitive social ranking, it does not fare better
as a tool for judging whether social outcomes are equitable.

2.3. Domain interpretation

The set of stations and the domain of definition of any SWFL are taken as a data set
provided without deeper formal justification. Although this information is not without
structure, it could prove very unwieldy. As we already indicated, the literature suggests
several axioms meant to select, from an ethical point of view, the information which
may matter and to delete unimportant details. They are partly motivated by the
cost of gathering and processing information, and in particular, by the degree of
precision deemed acceptable in the ethical observer's attempts to perform interpersonal
evaluation comparisons. However, this selection process cannot be appraised without
reference to the ethical interpretation of the data set: it depends on the ethical intuitions
one may entertain about what can be morally relevant for the problem at hand, and
this may be linked with views concerning the legitimate objectives of society and
its domain of intervention, on the one hand, and each individual's private sphere of
responsibility, on the other. General discussions of this topic may be found in Sen and
Williams (1982) and in Sen (1997).
From one polar standpoint, individual evaluation functions are just representing
individual preference relations over X, whatever they happen to be, self-oriented,
altruistic or anti-social. This assumption is in agreement with the bulk of the modern
theory of positive economics. Under ideal conditions, individual preferences can
be estimated or even observed. One might wish to base an ethically agreeable
social ranking on such a parsimonious information, that leaves simply no room for
interpersonal comparisons of either welfare levels or welfare gains. As Arrow (1963,
p. 112) points out, empirically distinguishable phenomena can be equated by our
value judgments, whereas empirically undistinguishable states cannot be differentiated.
Indeed, the pure individual preference interpretation of every profile U in D is implicit
in Arrow's (1951) definition of the social welfare function. It shares with his other
axioms the responsibility of his impossibility result.
Sen's (1970) definition of the SWFL is designed to circumvent this conclusion, by
allowing greater flexibility in the selection of usable information, while maintaining
a set of fully abstract alternatives. It does register interpersonal welfare comparisons,
and the latter have at least potential ethical significance. Information of this nature
may be indirectly based on other people's behavior, but Sen (1979) recognizes that one
cannot rely on commonly accepted inference procedures. On the other hand, Arrow's
views about empiricism are not so restrictive as the above judgment might lead some
to suppose. Indeed, he adds that empirical experiments may be of an idealized type. In
our context, he calls extended sympathy this source of evidence. In operational form,
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 475

the most basic version of the corresponding judgment reads as follows: "it is better (in
my judgment) to be myself in state x than to be you in state y" [Arrow (1963, p. 115)].
If this thought experiment is interpreted as an interpersonal exchange of characteristics
which does not alter the identity of individuals, it requires a parallel extension of the
set of possible alternatives. To be systematic and develop intuition about this kind of
statement, it may be helpful to figure oneself behind some veil of ignorance such as
those proposed by Vickrey (1945), Harsanyi (1955) and Rawls (1971), i.e. to pretend
one does not know about one's personal traits and circumstances in actual life, while
some more or less inscrutable impartial chance mechanism is about to allocate them
to society's members.
Hammond (1991) requires the ethical observer to assume an even more elevated
stance, as though he or she were capable of choosing not only who is to become
a member of society, but also what endowment of individual characteristics must
be allocated to any one member. This viewpoint involves a complete change of
perspective. In Hammond's (1998) own words, "... comparisons of the utilities of
different people to the chooser, rather than comparisons of different people's own
utilities ... ", are the ones which matter from his viewpoint. As common economic
usage seems to have converged towards the latter interpretation of the word utility, we
shall rather use the less specific word evaluation to fit either perspective.
When the description of the alternatives involves the allocation of idiosyncratic
characteristics to every individual, the idealized experiment we have been discussing is
of a highly subjective nature, and one must raise the question of the bound of this soul-
searching method, assuming that the process of augmenting X reaches its own limit.
Harsanyi (1955), Kolm (1972) and Hammond (1991) entertain the view of a unique
limiting preference relation. After postulating a deterministic theory of individual
preference formation, the first author went on to suggest that some kind of fundamental
preference would indeed emerge as a common bedrock for all human beings, once
they are stripped from their personal characteristics. Whether the postulated existence
of such a Holy Grail can prevent a severe lack of unanimity among subjective exper-
imentators, we leave our readers to decide after consulting the relevant literature 9 .
Independently of the conclusions reached about interpersonal comparability, we
may be persuaded that each Ui is a representation of i's individual preferences. In
this case, we say that the SWFL is utility-based. Nevertheless, profiles could be
prevented from having a dominant influence on the social ranking. In this direction,
the polar SWFLs are the imposed ones of example 2.2.1. In the opposite polar case,
one can let profiles occupy center stage and eliminate completely the influence of
whatever characteristics of the alternatives are not accounted for by the profile at hand.
Hicks (1959) criticized 0 this viewpoint which he considered to be dominant among

9 See Arrow (1977), Kaneko (1984), Broome (1993), Roberts (1995), Kolm (1996a) and Suzumura
(1996).
10 We are indebted to S.C. Kolm for pointing this out to us.
476 C. d'Aspreniont and L. Gevers

his fellow economists, and which he traced to the influence of A.C. Pigou's (1920)
Economics of Welfare. Hicks created the word welfarism to designate this doctrine.
Although he deals more explicitly with a multiprofile approach, Sen (1979) is faithful
to the same spirit, when he writes, "welfarism asserts that the goodness of a state of
affairs depends ultimately only on the personal utilities in the respective states". He
also stresses the limitations of welfarism, with reference to the SWFL context. In the
Tanner Lecture delivered in 1979, Sen (1980) seems to have lost the hope to discover
a minimally satisfactory welfarist SWFL, and he suggests a nonwelfarist approach
that builds on some arguments put forth earlier by Rawls (1971). According to the
latter, society should not be concerned with the distribution of individual utility or
welfare levels, as this would amount to meddling with basically private responsibilities.
Instead, society should be concerned with the distribution of what he calls primary
goods: basic liberties and "things that every rational man is presumed to want". Sen
(1980) comes up with an alternative proposal. He agrees that people are responsible
for their individual preferences; the latter are legitimately concerned with functionings
which can be achieved by consuming goods and by taking advantage of various social
opportunities. Sen aptly remarks that people are not equally proficient at transforming
goods and opportunities into functionings: some can be very gifted and some others
may be handicapped without being accountable for this state of affairs. Sen concludes
that justice requires society to be primarily interested in the distribution of individual
capabilities, i.e. sets of functioning opportunities.
We remark that anyone persuaded by these arguments, or for that matter, by
Hammond's (1991) interpretation of individual evaluation, may still be interested in
the formal SWFL approach. There does not seem to exist any good reason preventing
us from reinterpreting every U(x, i) as actually ascribed by the ethical observer to i if
the outcome is x in consideration of what is desirable for society: for instance, it could
be an index of the availability of primary goods, or an index measuring capabilities.
This is why we call Ui an individual evaluation function instead of a utility function.

2.4. Some related concepts

SWFLs make up the focus of the present chapter. According to Sen's original
definition, the social ranking Ru recommended by an SWFL is liable to change
whenever there is some variation of the profile U given a priori as relevant for social
evaluation. This is the only explicit independent variable. However, the SWFL setup
can be extended in several directions that can be combined:
(1) The set of individuals N may vary. This extension is mandatory if we want to
evaluate social decisions influencing the size of society. We refer the reader to
Blackorby et al. (2002).
(2) The feasible set X may vary. In this case, the ambition of the ethical observer may
stop short of uncovering a full social ranking, since he or she might simply look
for a solution, i.e. the subset of alternatives among which society ought to choose
in consideration of any given pair (X, U) of some well-defined domain. The study
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 477

of solutions is the main topic of axiomatic allocation theoryI . In some cases, as

in revealed consumer preference theory, the solution set coincides systematically
with the set of alternatives which are best for some social ranking, and it may
be possible to link properties of solutions to the properties of the social rankings
associated with them. See for instance Young (1987) or Sprumont (1996).
(3) Both X and some singled out alternative x° E X may be included as independent
variables together with U. Given N, we have here the basic elements of cooperative
bargaining theory. In Nash's (1950) tradition, X is interpreted as the set of
outcomes corresponding to as many feasible social agreements, whereas x© stands
for the outcome resulting from the lack of unanimous agreement; moreover, an
extended version of welfarism is adopted, so that bargaining problems are entirely
described by the respective images of X and x° in individual utility space. This
kind of structure may be further expanded to describe the fallback position of every
subset of N, and it can be found relevant for social evaluation exercises. See for
instance Yaari (1978). The limitations of welfarism in the bargaining context are
stressed by Roemer (1986, 1990).

3. Axioms and their use

3.1. Preliminary

It is one of our main tasks to define various features of SWFLs which may be
considered socially desirable. Formally, these value judgments are stated as axioms
and they are used singly or in combination to reject uninteresting SWFLs. It seems
natural to distinguish them following a two-way classification.
A first criterion refers to the axiom content. Some axioms are concerned with
separating formally superfluous details from potentially paramount information,
whereas others are reflections of ethical intuitions concerning equity or justice, and
it is useful to subdistinguish whether they may apply to conflicting situations or not.
The other criterion has to do with the number of profiles taken in consideration
to state the axiom. Interprofile axioms are distinct from intraprofile statements. The
latter actually define an incomplete binary relation over X on the basis of features that
any U could display. They would also have meaning out of the SWFL context. For
instance, when x dominates y in every individual evaluation, we may want the social
evaluation of x versus y to reflect the unanimous dominance relation. This axiom may
well register an ethical requirement, but it cannot handle conflicting situations.
When a single SWFL meets a number of axioms, we say that it is characterised
by them. Instead of discussing exclusively the merits and demerits of SWFLs with the

tl This is treated in this Handbook by Moulin (2002), Thomson (2002) and Fleurbaey and Maniquet
(2002).
478 C d 'Aspremont and L. Gevers

help of examples, we shall interpret them, in the next section, as compromises between
axiomatic properties and compare various characterisations.

3.2. Invariance axioms

Two seemingly distinct individual evaluation profiles may turn out to be equivalent
from the SWFL viewpoint, once what is deemed irrelevant has been left out. Although
the two profiles do not look the same in mathematical terms, they may be considered
equivalent because after all they contain the same usable information from the ethical
observer's standpoint. If this kind of analysis is pursued systematically, we obtain a
binary relation on D that is reflexive, symmetric and transitive: this is defined as an
equivalence relation; as such, it generates a partition of D consisting of various subsets
called equivalence classes, and any two profiles belonging to the same class have
the same image according to F. In other words, the social ranking associated with
some profile is required to remain invariant if the original profile is replaced with an
equivalent one.
Several types of reasoned value judgments, including equity judgments, may account
for dropping some profile features as negligible details. However, in the SWFL
literature, the name invariance axiom is usually applied to a principle meant to restrict
the measurability and comparabilityproperties of individual evaluation profiles. These
properties make up what Sen has called the SWFL informational basis. We shall not
attempt to define the italicized expressions we have just used. If the exact magnitude
of every number listed in a given profile were to matter for the definition of the
corresponding social ranking, the ethical observer would face a decision problem very
much like the ones described in economic consumption theory, where the agent is to
choose between bundles of goods measured in exact quantities. It is usually recognized
that individual evaluation counts do not involve such a degree of precision. We shall
discuss some useful distinctions in this respect under the measurability heading. As
it is often believed that a lesser degree of precision is achieved in interpersonal
evaluation comparisons than intrapersonally, further distinctions are called for under
the comparability heading.
Once the superfluous information carried by a profile has been pruned, what remains
is a set of meaningful statements, a notion formally developed in the measurement
literature, [e.g., Krantz et al. (1971)], and brought to bear to our debate by Bossert
(1991). From the invariance viewpoint, any two profiles are equivalent if they convey
the same set of meaningful statements. Thus, together with the description of X, the
latter set potentially encapsulates every bit of information about individual evaluations
that is necessary and sufficient to generate an appropriate social ranking of X. We have
emphasized the word potentially because some other axioms introduced at a later stage
may concur to eliminating excessive information.
As our ethical intuitions concerning equity and justice are hardly helpful for
appraising directly the merits and demerits of invariance axioms, we shall attempt to
do it indirectly, by describing the social rankings implied jointly by invariance and
Ch. 10: Social Welfare Functionals and InterpersonalComparability 479

other axioms taken together. In the same respect, axiom incompatibilities and trade-
offs may prove quite useful. For instance, we may be justified in rejecting the original
Arrovian informational basis because it cannot accommodate the anonymity axiom.
We shall also find it convenient to interpret the equivalence relation between any two
profiles as the outcome of a (possibly multiple) functional composition operation. In
other words, moving from any profile to another that is equivalent may be considered
as operating an invariance transformation.As will become clear, all the relevant ones
imply an n-tuple of strictly increasingtransformations, because the sets of meaningful
statements associated with two profiles are never declared equivalent whenever there
is an individual whose evaluation functions do not represent the same ordering. The
systematic study of such transformations antedates Bossert's approach; it is best
explained by Roberts (1980a). Other important contributions [Sen (1977, 1986b),
Moulin (1983), Basu (1983)] pertain also to these conceptual issues. We shall elaborate
on some of them at the end of this section.
Measurability and comparability issues are also discussed in other contexts such
as bargaining theory [see Shapley (1969), Shapley and Shubik (1975) and Shubik
(1982, pp. 92-98)] and axiomatic allocation theory [see Sprumont (1996) and
Fleurbaey and Maniquet (1996)].
To conclude these remarks, we introduce a new piece of notation. For any two
stations (x, i), (y, j) X x N, and any U D, we shall denote the first difference
U(x, i) - U(y, j) = Au(x, i;y,j).

3.2.1. Comparisons of evaluation levels

We begin with the presentation of the most restrictive invariance axiom, which is
implicit in the original Arrovian formulation. Each individual evaluation function is
assumed to be only a representation of the individual's preference relation. In the
behaviorist approach that is consistent with this interpretation, an individual preference
relation in combination with the maximizing assumption is simply a handy analytical
device capable of describing succinctly behavior under a variety of constraints. The
map associating the latter behavior with each constraint set is as much a primitive as
the analytical structure meant to facilitate its description. In the same spirit, Ui must be
thought of as a kind of shorthand presentation of the underlying preference relation.
For any two alternatives x,y X, the only kind of meaningful statement contained
in Ui is the relation "at least as great as" between U(x, i) and U(y, i) and vice versa.
If another individual utility function V: X -4 9i implies the same set of relations,
it has the same implications as the original Ui from the standpoint of positive theory.
Alternatively, we can require the sign of any first difference Au(x, i;y, i) to be the same
as the sign of Av(x, i;y, i).
Suppose this is the case; then, we can tabulate side by side the values taken by
both Ui and Vi for each alternative in X and look at the figures in one column as
functionally related to the corresponding numbers in the adjacent column. Prior to
elaborating formally on this idea, we denote Ui(X) (resp. Vi(X)) the image set of X
480 C. d'Asprenmont and L. Geuers

by U (resp. V). In other words, we let Ui(X) = r E N9 t 3x X such that

U(x, i) = r} and similarly for Vi(X). We are now ready to define the transformation
qi: Ui(X) -- Vi(X) such that Vx X, V(x, i) = pi(U(x, i)). In effect, we have so
defined a composition operation we can also write V = i o Ui, and one quickly
realises that qpi must be (strictly) increasing if Ui and Vi are required to be equivalent
from the positive viewpoint.
In the reasoning we just went through, we started with a pair of equivalent utility
functions and our conclusion pertained to a transformation. This order can obviously
be reversed: if we are given any Ui and a set of increasing transformations defined
on Ui(X), we can generate a set of equivalent utility functions. Thus, according to
modern positive theory, individual behavior is consistent with the maximization of
any member of a set of utility functions which can be obtained from each other
through increasing transformations: the latter invariance is known as an ordinality and
noncomparabilityproperty (or, in terms of meaningful statements, intrapersonallevel
comparability). This is our first invariance axiom:

Invariance with respect to individual increasing transformations (Inv(cpi(Ui)):

VU, V C D, for every n-tuple of increasing functions (i) CN,
Ru = Rv if Vi E N, qPi is defined on Ui(X) and Vx e X, V(x, i) = vi(U(x, i)).

It says that the social ranking is invariant if individual evaluation functions undergo
possibly distinct increasing transformations. The set of meaningful statements con-
tained in any profile consists only of n-tuples of individual statements of the form
described above in the particular case of individual i. To convince our reader that
our last axiom implies, as it should, an equivalence relation on D, we define a new
set as follows: consists of all n-tuples of increasing functions (i)i c N that can
be defined on XiENUi(X) with range Xic;NVi(X), and such that Vi E N, x X,
V(x,i) = (pi(U(x,i)). Thus, is generated by testing successively all pairs of
profiles U, V in 2D. We are now in a position to adduce three arguments: (i) If
the conditions defined in the axiom are met, there exists for each i an increasing
transformation qpI 1 that is defined on Vi(X) and such that Ui = cpf 1 (Vi), so that the
relation between U and V is symmetric; (ii) For every triple U, V, W E D, if Vi E N,
ci (resp. i) is an increasing transformation defined on Ui(X) with range Vi (X) (resp.
on Vi(X) with range Wi(X)) and satisfying the last condition stated in the axiom, then
the joint composition operation (i o i)i N results in a third increasing transformation
which is by definition an element of 'P, so that the relation on D is also transitive;
(iii) Finally, reflexivity is implied by the other two properties. Thus we have an
equivalence relation on D.
Let us move next to less restrictive principles, by introducing comparability
among individual evaluations. For this purpose, we shall interpret any profile as an
expression of extended sympathy. In the simplest case, we content ourselves with level
comparability,as the ethical observer supplies the missing link between the individual
evaluation orderings in order to obtain no less and no more than a single ordering
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 481

over X x N. The corresponding set of meaningful statements may thus be limited to

the following: for any two stations (x, i), (y, j) C X x N, the individual evaluation
of the former is at least as great (as small) as the individual evaluation of the latter.
Alternatively, we can register for every ordered pair of stations the sign of the first
difference between their individual evaluation scores. We observe that we have so
defined a superset of the set of meaningful statements that would be implied by our
previous axiom Inv(pi(Ui)).
Suppose now that any two profiles U and V happen to deliver the same set of
meaningful statements according to our new principle, so that they are equivalent.
We can then tabulate side by side the values taken by either profile for each station
in X x N. In effect, we are facing the graph of a transformation with domain
U(X,N) = {r 9 I 3(x, i) E X x N such that U(x, i) = r} and similarly defined
range V(X, N); we observe that it must be increasing. Since the reciprocal is also true,
we can conclude that we are measuring individual evaluations on a common ordinal
scale (leading to what has been called a property of co-ordinality, or, in terms of
meaningful statements, of interpersonal level comparability). We will use it as an
axiom of
Invariance with respect to common increasing transformations (Inv( p(Ui))):
VU, V E D, for every increasing function defined on U(X,N),
Ru = Rv if V(x, i) e X x N, V(x, i) = p(U(x, i)).
As we compare this axiom with the previous one, we remark at once that any
two profiles declared equivalent by Inv(p(Ui)) must also be declared equivalent by
Inv(i(Ui)). Hence, the partition of D in equivalence classes generated by the former
must be finer than the one generated by the latter. Given any profile U E D, the
subset of profiles considered equivalent in the former case must be contained in the
subset corresponding to Inv(qi(Ui)), notwithstanding the seemingly opposite assertion
we just made about sets of meaningful statements. To get an idea of the magnitudes
involved in the refinement process, let us consider an example: suppose IXI = s, a finite
number. Suppose also that D consists of all the profiles in the universal domain made
out of ns distinct numbers, so that indifferences never occur. Then, the finer partition
consists of (ns)! cells, whereas the coarser one consists of (s!)n subsets; these numbers
are equal if n = 1, but the ratio of the former to the latter increases by a factor greater
than (n + 1)s each time we add an individual to an n-member society. An analogy
between a profile and a set of n geographic maps may also help comparing Inv(T(Ui))
with Inv((pi(Ui)). If the latter axiom is adhered to, any profile looks like n disconnected
contour maps with ascending directions indicated, whereas the n contour maps would
be fully connected in the former case.

3.2.2. Comparisons of evaluation differences

For usual geographic purposes, information about ascending directions would have
to be supplemented at least by information about the gradient steepness. Similarly,
482 C. d 'Aspremont and L. Geuers

having profiles measured on an ordinal scale might be deemed insufficient, and

information about the sign of first differences in individual evaluation provided by
our first two invariance axioms would have to be completed in a variety of manners
to which we shall turn next. All of them involve first differences in evaluation counts.
In the four following axioms, we find it simpler to assume the same degree of
comparability between individual evaluation functions as within them, a feature shared
with Inv(q((Ui)).
For instance, the set of meaningful statements could be an exhaustive list of
sentences of the following form: the net individual evaluation gain obtained by moving
from (y, j) to (x, i) is at least as great (as small) as the net gain obtained by moving
from (z,l) to (w, k). In this case, each comparison involves two ordered pairs of
stations, possibly all distinct, and the full list of Au(x, i; y, j)'s makes up the range
of an ordering representation over (X x N)2 , which in turn includes the definition of
a unique ordering on (X x N). This partition of D was introduced by Bossert (1991)
under the name of strong interpersonaldifference comparability.
If any two profiles U and V in D are equivalent in this sense, we may again observe
that there exists a transformation Vp: U(X, N) -- V(X, N) such that V(x, i) = p(U(x, i))
for every station (x, i) C X x N, and where in Basu's (1983) terminology, p is
first-difference preserving: Vsl,s 2 ,s 3 ,s 4 U(X,N), sl - s2 > s3 - s4 if and only if
1p(sl)- 1(s2) > VP(S3) - 1P(s4). Moreover, we may note that the reciprocal statement is
also true.
Alternatively, we may rephrase the equivalence condition between U and V in D
as follows: there exists a domain AU = {r C 9i1 3I(x,i), (y, j) E X x N such that
Au(x, i; y,j) = r}, and an increasing function : AU - 9 such that for any two
stations (x, i), (y, j) X x N, Av(x, i; y, j) = cp(Au(x, i; y, j)).
This possibility is almost trivial to prove: By assumption, we are given U and V
in D and they are equivalent. Suppose first that there exists Vi: U(X,N) -- V(X,N)
which is first-difference preserving; then for any two stations (x, i), (y, j) X x N,
we let q((Au(x,i; y,j)) = Vp(U(x,i)) - (U(y,j)) = Av(x,i; y,j) and we
observe that is a well-defined increasing function. Suppose next that there
exists qp: AU - 91 as in the reciprocal statement to be proved; then we define
ip: U(X, N) - V(X, N) such that V(x, i) = p(U(x, i)) for every station (x, i) E X x N,
and we observe immediately that for any two stations (x, i), (y, j) X x N,
Av(x, i; y, j) = (Au(x, i; y, j)) = p)(U(x, i)) - ip(U(y, j)). Since p is increasing,
p must be first-difference preserving.
We can now proceed by introducing formally the axiom, meant to capture strong
interpersonal difference comparability.

Invariance with respect to common first-difference preserving transformation

(Inv(( (AU))):
VU, V E D, for every increasing function (p defined on AU,
Ru = Rv if V(x, i), (y, j) C X x N, V(x, i)- V(y, j) = (U(x, i)- U(y, j)).
Ch. 10: Social Welfare Functionals and InterpersonalComparability 483

Much more discriminating would be a set of meaningful statements delivered for

any two stations as follows: Au(x, i; y, j) = c, for c E 91 and (x, i), (y, j) e X x N. In
other words, the full list of ordered pairs of stations and the exact numerical magnitude
of the first differences in evaluation counts would be of potential interest for the ethical
observer. Thus, if any U, V C D are equivalent according to this principle, we must
have V(x,i),(y,j) C X x N, Au(x,i; y,j) = Av(x,i; y,j). Obviously, the latter
equality is not only necessary, but it is also sufficient to make sure that U, V D
are equivalent.
As Bossert (1991) remarks, the same equality condition is in turn satisfied if and
only if there exists some real number a such that V(x, i) e X x N, V(x, i) = a + U(x, i).
To prove necessity, we fix some (y, j) C X x N, and let V(y, j) - U(y, j) = a. We
observe that, by assumption, V(x, i) C X x N, V(x, i) - V(y, j) = U(x, i) - U(y, j),
so that V(x, i) - U(x, i) = V(y, j) - U(y, j) = a, as was to be proved. In conclusion,
V is equivalent to U if and only if they are the same up to a change of origin, a
property also known as translation-scale measurability with full comparability. This
is the same as having individual evaluation counts measured in common natural units,
while their common origin is arbitrary. The corresponding axiom can be stated formally
as follows:

Invariance with respect to common changes of origin (Inv(a + U)):

VU, V C D, a C 9, Ru = Rv if V(x, i) X x N, V(x, i) = a + U(x, i).

We turn next to an informational basis that lies in between the two last ones. Accord-
ing to this intermediate principle, the set of meaningful statements consists of sentences
formulated as follows: given any four stations (x, i), (y, j), (w, k), (z, 1) E X x N, such
that Au(w, k; z, ) # 0, the ratio of Au(x, i; y, j) to Au(w, k; z, 1) amounts exactly to c,
for some c E 91. Any two profiles U, V E D are thus equivalent for this principle if
and only if V(x, i), (y, j), (w, k), (z, ) X x N, the ratio of their corresponding first
differences is equal, assuming that sgn{Au(w, k; z, l)} = sgn{Av(w, k; z, l)} # 0.
When the last assumption is satisfied for two pairs of stations, say (w, k)
and (z, 1), this condition is in turn satisfied, if and only if V is a positive affine
transformation of U, or, in other words, there exist two real numbers a, b with b > 0,
such that V(x, i) E X x N, V(x, i) = a + bU(x, i). To prove necessity, we avail
ourselves of Au(w, k; z, I) and Av(w, k;z, 1) and we let b stand for the ratio of the
latter to the former. By assumption, b > 0. Moreover, we fix some (y, j) X x N,
and let V(y, j) - bU(y, j) = a. We observe that, by assumption, V(x, i) C X x N,
(V(x, i) - V(y,j))/(Av(w, k;z, I)) = (U(x, i) - U(y,j))/(Au(w,k;z, 1)), so that
V(x, i) - V(y, j) = b(U(x, i) - U(y, j)) and V(x, i) = bU(x, i) +(V(y, j) - bU(y, j))=
a + bU(x, i), as was to be proved.
Under these assumptions, profiles U and V are said to be measured on a cardinal
scale with full comparability. The origin and the scale of individual evaluation counts
484 C. d'Aspremont and L. Gevers

are both common and arbitrary. The axiom (corresponding to a property called co-
cardinality) can be stated formally as follows:

Invariance with respect to common positive qffine transformations(Inv(a + bUi)):

VU, V C D, Va, b C , with b > 0,
Ru = Rv if V(x, i) E X x N, V(x, i) = a + bU(x, i).

It is easy to see that any positive affine transformation is first-difference preserving (so
that Inv((p(Au)) implies Inv(a + bU)) whereas the reciprocal statement is not always
true. Think for instance of U1(X) = {1,2,4} and Vi(X) = {1,2,5}. Basu (1983)
provides an interesting domain condition that is sufficient for equivalence, viz. that
Ui(X) be dense on a nontrivial connected subset of Bi. This condition is met, for
example, if the ethical observer is concerned with a set of alternatives and a domain
of profiles consistent with the definition of a family of Arrow-Debreu economies,
but it cannot be satisfied if the SWFL is defined on an unrestricted domain [see also
Fishburn, Marcus-Roberts and Roberts (1988) and Fishburn and Roberts (1989)]. The
following table recapitulates the five invariance properties described so far. Arrows
indicate inclusion relations between sets of meaningful statements. To list the latter in
summary form, we have used the language of first differences.

Transformation Meaningful statement

a+U Au(x, i;y,j)

a + bUi Au(x, i;y,j)/Au(w, k; z, I)

p(A U) sgn{Au(x, i; y, j) -Au(w, k; z, l)}

cp(U/) sgn{Au(x, i; y, j)}

qi(U,) (sgn{Au(x, i; y, i)})i c N

Except for our very first invariance axiom, we have relied so far on interpersonal
evaluation comparability. This feature may be hardly desirable for anyone wishing to
remain as close as possible to hard data and commonly accepted inference procedures.
Therefore, we shall sample next some weakenings of interpersonal comparability. For
this purpose, let us reconsider how Inv(cpi(Ui)) relates to Inv(qp(U)): instead of treating
a profile as a single function defined on X x N, we look at it as a n-tuple of individual
evaluation functions defined on X; instead of comparing the respective evaluations
of any two stations, meaningful statements are restricted to comparing them from
the viewpoint of one single individual at a time. Similarly, the invariance property
is no longer based on a single transformation defined on a typical image set U(X, N),
but it calls for a n-tuple of transformations, each one being applied to an individual
image set Ui(X).
Ch. 10: Social Welfare Functionals and InterpersonalComparability 485

It is straightforward to remove formally interpersonal comparability from Inv(a + Ui)

by relying on the same pattern of reasoning. In this case, profiles are said to be
translation-scalemeasurable, without additional qualifier, and we obtain
Invariance with respect to individual changes of origin
(Inv(ai + U/)):
VU, V D, V(ai) G fiN, Ru = Rv if V(x, i) E X x N, V(x, i) = ai + U(x, i).
In this case, the unit of measurement cannot be changed, and a natural interpretation
would hold it interpersonally comparable.
In contrast, as we turn to cardinality with full comparability, we notice that the
latter feature may be erased thoroughly, but the process need not be so radical. We
have pointed out in due time that the set of meaningful statements corresponding to
Inv(a + bUi) pertains to ratios of first differences in individual evaluations involving
up to four distinct individuals. If we decide to throw out of the list any ratio involving
more than one individual at a time, the end result is a new axiom prescribing
Invariance with respect to individual positive affine transformations
(Inv(ai + bi Ui)):
VU, V G D, V(ai)E N, (bi) E 9+N+,
Ru = Rv if V(x, i) E X x N, V(x, i) = ai + biU(x, i).
Indeed, we observe that the origin and the scale of each individual evaluation
function are arbitrary. In other words, each evaluation is measured on a cardinal
scale without interpersonal comparability. If the SWFL is utility-based, this can
perhaps be defended when uncertainty is made explicit and individuals are assumed
to display von Neumann-Morgenstern rationality. Moreover, as we reconsider the role
of individual preference profiles in the positive theory of rational social interactions,
we remember that one of its commonly accepted tools is the Nash noncooperative
equilibrium concept based on mixed strategies, whose conclusions are invariant under
independent positive affine transformations of each individual utility function in the
profile on which equilibrium is based. The same invariance property was invoked by
Nash (1950) to justify his bargaining solution, and it could be used as an interesting
SWFL informational restriction.
Alternatively, we may content ourselves with a less severe shearing operation and
consider significant any ratio involving up to two individuals, provided each first
difference refers to one single person at a time. We get
Invariance with respect to common rescaling and individual change of origin
(Inv(ai + bUi)):
VU, V E D, V(ai) E MN, Vb eG
9++,
Ru = Rv if V(x, i) G Xx N, V(x, i) = ai + bU(x, i).
In this case, each individual evaluation is measured on a cardinalscale; there is unit
comparabilityamong individuals, but level comparability is excluded. This axiom may
486 C. d'Aspremont and L. Gevers

look less directly intuitive than either Inv(a + bUi) or Inv(ai + biUi); it can appeal to
someone who wishes to concede as little as possible to interpersonal comparability.

3.2.3. Comparisons of evaluation indicatorratios

Among the invariance axioms we have been discussing, none does restrict a priori
the sign of individual evaluation counts. However, there may be interesting contexts
where they are all positive for each profile in the SWFL domain, while they refer to
a worst (possibly unattainable) alternative which is used as a natural origin. Or they
might all be of negative sign by reference to a best (possibly unattainable) alternative
used in similar fashion; in this case, referring to cost figures seems to be closer to
usual conventions. For instance, a firm manager may judge the relative effort level of
two salesmen by looking at their sales ratio; the same manager may also rely on the
cost ratio to evaluate the relative performance of two workers who deliver the same
quantity of output. Suppose now we want to compare two profit centers, one of which
is in the red whereas the other one earns a positive amount. In this case, it would
seem odd to compute the ratio of two figures of opposite sign. We present next a set
of meaningful statements that is hardly intuitive on domains which mix up positive and
negative signs. A typical domain that meets the requirement of being homogeneous
with respect to sign will be denoted D*.
The new set of meaningful statements consists of propositions revealing the
exact magnitude of every well-defined ratio of individual evaluation counts. Thus,
U and V are equivalent for this principle if and only if we have (V(x, i))/(V(y, j))
= (U(x, i))I(U(y, j)) for any two stations (x, i) and (y, j), with sgn(U(y, j)) =
sgn(V(y, j)) X 0, and the latter condition holds true if and only if there exists b e 9++
such that for every (x, i) C X x N, V(x, i) = bU(x, i). We omit the easy proof, which
is analogous to the previous ones. In this case, individual evaluations are said to be
measured on a ratio-scale with full comparability. Tsui and Weymark (1997) note
that percentage changes in individual evaluations might be substituted for ratios in the
above developments without altering the conclusion. The latter is captured formally in
our axiom of

Invariance with respect to positive similarity transformation (Inv(bUi)):

VU, V E D*, Vb E 9i,,, Ru =Rv if V(x, i) E X x N, V(x, i) = bU(x, i).

It should be noted that anyone wishing to rely exclusively on the transformation

approach to invariance may be interested in similarity transformations over an extended
domain and reject the sign restriction inherent in D*.
We may consider also ratio-scalemeasurability without comparability. The relevant
set of meaningful statements discloses the exact magnitude of every well-defined
ratio (U(x, i))/(U(y, i)) for each individual separately. Therefore, any two profiles
U and V in D* are equivalent for this principle if and only if Vi N, we
have (V(x, i))/(V(y, i)) = (U(x, i))/(U(y, i)) for any two alternatives x and y, with
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 487

sgn(U(y, i)) = sgn(V(y, i)) # 0, and the latter condition holds true if and only if there
exists bi E f++such that V(x, i) = bi U(x, i) for every x e X The corresponding axiom
is
Invariance with respect to individual positive similarity transformations
(Inv(bi Ui)):
VU, V D*, (bi) C 9N+i, Ru = Rv if V(x, i) X x N, V(x, i) = biU(x, i).
It should be stressed that our new axiom retains implicitly a piece of information that
is interpersonally comparable. Indeed, the origin cannot be changed arbitrarily, and
a zero count can be attributed to the evaluation of the worst (or perhaps the best)
alternative by every individual, an alternative which is not necessarily the same for
everybody. As we noted for Inv(bUi), the sign restriction of D* might be rejected by
someone who does not care about intuitive meaningful statements.

3.2.4. Other invariance axioms

To illustrate a mixed-invariance principle, we combine the restrictions imposed by both
Inv(ai + bUi) and Inv(q(Ui)), to wit
Mixed invariance (Inv((ai + bUi)&q(Ui))):
VU, V D, V(a,) cE N, Vb E i++,
for every increasing function defined on U(X,N),
Ru = Rv if V(x, i) E X x N, V(x, i) = (p(U(x, i)) = ai + bU(x, i).
Interpersonal comparisons of levels and gains are allowed, but at most two individ-
uals are involved in any comparison of utility gains, in contrast with Inv(a + b Ui). The
corresponding set of meaningful statements is obtained by operating the union of the
two sets implied by the parent axioms. 12
More exotic invariance axioms could be added to our list. The following example has
little interest by itself: it is included only to illustrate some boundaries of the invariance
transformation approach. At first glance, the opening statement looks symmetrical to
Inv(ai +bUi).
Invariance with respect to individual rescaling and common change of origin
(Inv(a + biUi)):
VU, V D, Va 9i, V(b,) C 9f+,

Ru = R if V(x, i) C X x N, V(x, i) = a + biU(x, i).

Dixit (1980) points out rightly that the inverse transformation is not necessarily
increasing, so that the relation between U and V implied by the transformation fails to

12 This is to be distinguished from the requirement that the SWFL should satisfy more than one
invariance axiom.
488 C. d'Aspreniont and L. Gev;es

bUi-a + bU, |a + Us

cp(AU) a, +bUi

I / ( [ 0ai + Ui

_TarXl
biUi- qi(Ui) ai + bUi
j

ai + bi Ui Fig. 1. Invariance transformations.

be symmetric. It follows that profiles can hardly be considered equivalent, unless he

equivalence criterion is redefined in a way that is more roundabout than in the previous
cases. Thus, U and V may be declared directly equivalent if either the condition stated
in the axiom holds or if, instead, the last clause is replaced with U(x, i) = a + bi V(x, i).
Furthermore, they may be declared equivalent if they are either directly equivalent, or
there exists in D a sequence U, U', U",..., V such that any two successive profiles
in the sequence are directly equivalent. If this stricture is accepted, we obtain a well-
defined partition of D, which is consistent with Inv(a + bi Ui), but we failed to derive
its implications in terms of meaningful statements.
On occasion, we shall rely on weakened versions of some invariance axioms; we
introduce them when we need. At this stage, we feel that we have been already taxing
too much our reader's imagination. Let us recapitulate their implication relations. As
we compared Inv(q(Ui)) with Inv(i(Ui)), we already pointed out that the latter axiom
is stronger than the former, whereas the set of meaningful statements delivered by any
profile is always richer in the former case than in the latter; hence, the ethical observer
has a more acute discriminating power in the former case. On the other hand, he or she
is more susceptible to being misled by potentially slippery information. This inclusion
relation between sets of meaningful statements associated with pairs of invariance
axioms is displayed in Figure 1 by means of arrows pointing towards the smaller set: it
is not a complete relation, but nevertheless it proves quite useful. We make at once the
following remarks: the Arrovian axiom Inv(pi(Ui)) is dominated by every single one
of our remaining ten axioms, and there are two undominated and thus non-comparable
axioms, viz. Inv(a + Ui) and Inv(bUi). One can conjure up an axiom of independence
with respect to identity transformations which would dominate both of them, and which
is implicitly built in the SWFL definition; but this would hardly help. Anyone interested
in characterizing a specific SWFL may be tempted to choose an invariance axiom that
lies as close as possible to Inv(qi(U,)), whereas someone interested in uncovering the
implications of a richer information structure will attempt to maintain unchanged a set
of basic axioms while combining them successively with weaker and weaker invariance
axioms. By moving closer and closer either to Inv(a + U) or to Inv(bU/), one is bound
to admit a larger and larger set of SWFLs consistent with the axioms already listed.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 489

Moreover, it proves illuminating to define ordered subsets or types of axioms and to

study the tradeoffs between weaker axioms of one type and stronger axioms of another
type.

3.3. Other information-filteringaxioms andformal welfarism

3.3.1. Pareto indifference, Independence and Neutrality

If information gathering and processing is costly, one is tempted to focus attention

exclusively on what seems most important. Independently of invariance principles,
our next sequence of axioms prescribes the deletion of what may appear as
unnecessary details in the formulation of social evaluation judgments. This is what
the axioms called Pareto indifference, Binary independence, Neutrality, Continuity and
Separability are meant to capture. We introduce them in the present section, together
with some neighboring axioms. We introduce also an important property of SWFLs
we call formal welfarism; furthermore, we are in position to characterise the set of
SWFLs satisfying this property.
We begin by restricting the analyst's attention exclusively to the content of the
relevant individual evaluation profile. This can be expressed as a single-profile
property, known as the Pareto indifference principle; it says that any two alternatives
are socially indifferent if they are represented by the same image in the evaluation
space:

Pareto indifference (PI).

VU e D, Vx,y E X, xIuy if U = U.

Thus, as long as Ux = Uy,, the particular consequences involved in the description of

the alternatives cannot affect the social ranking.
We introduce next another method for eliminating superfluous details when dealing
with a subproblem. Suppose indeed that the social ranking we are interested in is
restricted to a particular pair of alternatives; does this judgment require any information
regarding other alternatives which could be feasible? A negative answer is best
expressed as an interprofile property. Two profiles U, V and two alternatives x,y are
involved; we further assume that their respective images in the evaluation space are
the same; in other words, Ux = Vx, U, = Vy. Then, we require the social rankings Ru
and Rv to be the same with respect to the pair (x, y), independently of the discrepancies
existing between U and V over the other alternatives.

Binary independence (BIN).

VVE D, Vx,y X, xRvy if U E D such that Vx = U, y = Uy and xRuy.

In other words, if the restrictions of two profiles to a given pair of alternatives cannot
be distinguished, the relative social ranking of the two alternatives under focus must
490 C. d'Asprenmont and L. Gevers

be unique. Without some similar principle, the exact definition of X or the exact
application of the definition would always be problematic. Directly adapted from the
Arrovian axiom of Independence of irrelevant alternatives, it has become a workhorse
of the bulk of the SWFL literature. Even though it does not imply any invariance
axiom, unlike its Arrovian model, it has been criticized as being too constraining.
For instance, it makes the SWFL less suited for the discussion of rights; games in
various forms provide more appropriate tools of analysis. See, for instance, Gaertner
et al. (1992) and Suzumura (2002). In contexts closer to the SWFL, several authors
recently managed to derive social rankings despite the fact that they weaken Binary
independence: we shall report briefly about their characterisation results at the end of
this chapter.
We turn next to the discussion of welfarism and its relation with the two properties
we just described. Drawing our inspiration from Blackorby et al. (1990), we consider
first the image of any profile U E D in 9 N interpreted as the evaluation space, and
we denote it U(X). Formally:

U(X) = u 9N 3x X, u = Us}.

Letting Ru denote the set of orderings on U(X), with typical element R*., we shall
say that F displays profile-dependent welfarism if and only if VU D, R7 E R*r
such that

Vu, v C U(X), Vx,y X, (u = U and v = U) (uRov X xRuy).

The interpretation seems clear: the association between Ru and R; is one to one; to
study the restriction of F to U, one can equivalently rely on Ru or on R*, since each
version may be translated into the other without ambiguity. In particular, Ru is then
said to be fully recoverable from Ru.
We proceed with the analysis of Pareto indifference, which has some remarkable
implications by itself. We stressed already one of them: viz., for any given profile,
neither the intrinsic characteristics of an alternative nor its name can have any bearing
on its social ranking: all that matters is its individual evaluation vector; this idea may
also be re-expressed formally by a property called

Intraprofile neutrality (IAN):

VU E D, Vx,y,x', y' C X such that U = U, and U, = Uy,, xRuy iff x'Ruy'.

Intraprofile neutrality implies Pareto indifference, because we may choose x' =y and
y' = x in the last statement. On the other hand, it is also implied by Pareto indifference,
due to the transitivity of any social ranking Ru. Thus, we are led to the following
Theorem 3.1. For all D C U, Pareto indifference is equivalent to Intraprofile
neutrality.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 491

To clarify the relation between Pareto indifference and profile-dependent welfarism,

it is convenient and always possible to define on U(X) a binary relation hR as follows:
Vu, v C U(X), uRUv iff 3x,y E X such that u = U, v = Uy and xRuy.

We observe at once that R lacks one of the characteristics of the relation R; used to
define profile-dependent welfarism. Indeed, xPuy can hold true despite the fact that
u = Ux, v = Uy and uIuv; in other words, recoverability is not guaranteed by R. Let
us elaborate on this. The relation R u inherits reflexivity from Ru, Suppose we reject
Pareto indifference; then, there must exist some x and y such that xPuy and Ux = Uy;
in this case, by reflexivity, we register UVJ Ux, whereas xPuy, so that Ru cannot be
fully recoverable from Ru. Partial recoverability is however warranted in the following
sense: uPbv, Ux = u and Uy = v together imply xPuy. Indeed, any conclusion to the
contrary would immediately contradict uPbv. Failing Pareto indifference, the bluntness
of R: proves unfortunately infectious:
Lemma 3.2. For any U G D, Vx,y,z E X, such that xPuyPuz, Ux = U implies
UXIU UY U

Proof: By assumption and by definition, we obtain UXR^R* U,, whereas UXI, U,

by reflexivity, so that Uxlb Uy. By a similar argument, Uyl1 Uz. D[
Having shown that Pareto indifference is necessary for profile dependent welfarism,
we can prove it to be also sufficient:
Theorem 3.3. For all D C U, Pareto indifference is equivalent to profile-dependent
welfarism.
Proof: Suppose we have PI and uRv for some u, v E U(X); we want to show that
xRuy if U = u and Uy = v. By definition, uu v only if there exist x',y' E X
with x'Ruy', Ux, = u = U and Uy, = v = U,. There remains to apply the previous
theorem. D1
Of course, the ordering RU depends in general on U, the profile selected in )D.
For instance, R can agree with utilitarianism for some profiles in D, whereas it
reflects the leximin principle for all other profiles in D. If the SWFL is utility-based,
profile-dependent welfarism seems to agree with what Hicks (1959) probably meant
when he coined the word welfarism. By dropping profile dependence, Sen developed
a more useful but demanding concept. As he was dealing exclusively with utility-
based SWFLs, he relied on the same word as Hicks. We shall take the liberty to
twist Sen's definition of welfarism because we believe the scope of the SWFL tool
is wider: we shall say that an SWFL is welfarist in a formal sense if the relative
goodness of two states of affairs can be entirely judged by comparing the goodness
of their respective individual evaluation vectors, independently from the other aspects
of the profile at hand. In this case, each social ranking Ru on X can be construed
as emanating from a unique ordering defined on the relevant subset of the individual
492 C. d'Aspremnont and L. Gevers

evaluation space. The various rankings on X remind us of spokes, whereas the unique
ordering to which they are linked is analogous to a hub. This property will prove
invaluable in the sequel because it is easier to characterise an ordering or to study
its representation than to deal directly with a multiprofile structure. What axioms are
necessary and sufficient to obtain formal welfarism? This is the question which we
address next.
As a preamble, we define a Social welfare ordering (SWO); for lack of a less
ambiguous word, this is simply how we designate an ordering on the evaluation
space (9 v, AN or 9lN+ as the case may be). We shall further say that F displays
formal welfarism if and only if we can associate with F an SWO R* with the following
property:

'F(X,D), Vx,y G X, VU
Vu, v GE D,
(u = Ux and v = U) z= (uR*v xRuy).

In this case, we say that R* is the formally welfarist associate of F We remark

that, whatever the profile U considered, Ru is fully recoverable from R*; under these
circumstances, one also has xIuy if uI*v and xPuy if uP*v: there is thus a one-to-one
link between the SWFL and its SWO associate under formal welfarism. For example,
the purely utilitarian SWO (i.e. the formally welfarist associate of the purely utilitarian
SWFL) is defined by
1 1

VU, E 9i, UR*u ui Vi.

i=l i-1

Similarly, the formally welfarist associate of the leximin principle (resp. of its
inequitable mirror-image called leximax) is defined by:

Vu, v C 9tN, uP*v iff 3k E N, such that Ui(k) > i(k), and
Vh E N, h < k (resp. h > k), Ui(h) = Vi(h),

where ui(.) (resp. i(.)) denotes the non-decreasing re-ordering of the vector u (resp. v)
according to the permutation i(.) such that i(k) is the kth-worst off individual:
Ui(l) ui(2) < '< i(n) (resp. i(l) < i(2) < ''' < i(n)).
Suppose we are given some SWFL F and we want to find out whether it displays
formal welfarism. As a first step, we associate with F a reflexive binary relation R*
defined on 7t(X, D) as follows 13: Vu, v E 7-(X, D),

uR*v ( U E D, x,y X such that u = U, v = Uy and xRuy).

13 Alternatively, uR*v 3EU E D such that u,v C U(X) and uRuv.

Ch. 10: Social Welfare Functionals and InterpersonalComparability 493

We have to make sure that every Ru can be retrieved from R*. For this purpose, it is
convenient to define another SWFL property, which is often used as an axiom in its
own right. In the literature, it is referred to as
Strong Neutrality (SN):
VU E D, Vx,y EX,
xRuy if there are x',y' E X and V e D such that Vx, = Ux, Vy, = Uy, x'Rvy'.
We can state at once an obvious remark, which does not require any domain condition
pertaining to F:
Theorem 3.4. Formal welfarism implies Strong neutrality.
However, for the characterisation coming next, a rich domain condition (e.g., the
domain is Pollak) is maintained (see Section 2.1):
Theorem 3.5: Formal welfarism. Suppose F satisfies Domain universality or
attainability. Then it satisfies Strong neutrality if and only if it displays formal
welfarism, whereas R* is an ordering over '7-(X, D) = 9 N
Proof: One direction follows from the preceding theorem. For the other direction, we
first prove full recoverability. Take any u, v e 7-(X, D), any U D, any x,y X,
such that u = U, v = Uy and uv. We want to show that SN implies xRuy. Now,
by definition of R*, there exists V D, x',y' E X such that u = Vx,, v = Vy, and
x'Rvy'. Applying SN, we obtain xRuy. We have indicated earlier that R* is reflexive
over 7-(X,D). To establish the completeness and transitivity of k*, we select any
u, , w E 9AN. As F satisfies UD or AD, there exists a profile U L U, and x,y,z X
such that u = U, v = U, w = Uz. Because Ru is an ordering, xRuy or yRuz so
that uR*v or vR*u. We may w.l.o.g. assume xRuy and yRuz, and obtain xRuz, so that
uR*v, vR*w and uR*w by applying the definition of R*. D
Whatever the domain of definition of the SWFL, Strong neutrality implies
Intraprofile neutrality and thus, Pareto indifference; to see this, simply add to
the statement of Strong neutrality the following clause: U = V. Another useful
implication of Strong neutrality may be obtained by adding the orthogonal requirement
x = x',y = y' to its statement; without surprise, we obtain the axiom of Binary
independence:
Theorem 3.6. For all D C U, Strong neutrality implies both Pareto indifference and
Binary independence.
If the domain of F is universal, a reciprocal statement can also be proved. It has
been called welfarism theorem by Sen (1977), who derived it by strengthening a result
of d'Aspremont and Gevers (1977).
Theorem 3.7. If D = U, Pareto indifference and Binary independence in conjunction
imply Strong neutrality, and hence, formal welfarism.
494 C. d'Aspreinont and L. Gevers

Proof: Suppose we are given two profiles U, V , two pairs (x,y) and (x',y')
of alternatives in X, and two vectors u,v C 9N such that Ux = Vx, = u,
Uy = Vy = v, and xRuy. We have to show that x'Rvy'. Since we always assume
that XI > 3, we may choose r C X\{y,y'}, with r a third alternative if the two
pairs coincide, and construct profiles U l , U2 and U3 such that U = U,j = u,
UI = v, U~2 , U =U 2 I= , ,U = U = and U,!, = v. Then letting ""
mean "implies by BIN", whereas ">" means "implies by PI and transitivity", we get:
xRuy =: xRuy rRuiy 4 rRu2y => rRu2y' rRu3y' = x'Ru3y' 4 x'Rvy'. We
have thus SN. Formal welfarism follows from Theorem (3.5). C

3.3.2. Formal welfarism and invarianceproperties

All axioms we have been studying in this and the previous subsection have in common
the goal of filtering information. Whenever some of them taken together imply formal
welfarism, the associated SWO properties are inherited from axioms pertaining to the
parent SWFL. In most cases, there is a one-to-one link between the parent axiom
and the corresponding SWO property. To designate the latter, we shall use an asterisk
flanking their parent's name, be it in full or in shortened version. In particular, each
invariance axiom stated so far for SWFLs translates (in Section 3.2) as a relation
between pairs of points in 9 N. As the proof of their equivalence is almost trivial, we
shall content ourselves to quote the ten main axioms, relying only on their short names.
Only one of them requires some care. This is invariance with respect to common first-
difference preserving transformation, the domain of definition of which matters a lot, as
Basu (1983) points out. Towards defining it properly, we first pick any point u E 9N
and let T(u) = t C 1 I 3i C N such that u = t}. For any pair u,v E C 9 N, the
appropriate domain is defined as Y(u, v) = T(u) U T(v).
* Inv*(qp(Au)): Vu, v C 91 N, and any first-difference preserving function ip: Y(u, v)
9, i.e., VS1,s 2 ,s 3 ,s 4 C Y(u, ), (sl) - I(s 2) •< (s3)- (s4) S1 s2 < s3 -4,

uR*v X (V(ul), .. , (u,))R* (tvi). , P(u,)).

We remark that we could in particular choose 3 = 4 Y(u, v) in the above

statement, and observe that ip must be increasing. The next item on our list is the
axiom of invariance with respect to common increasing transformation:
* Inv*(q(ui)): V u, v c 91N, and for any increasing function oq: Y(u, v) --+91,

uR*v X (p(ul) ... v(u))R* ((v3) ... , (v,,)).

We observe that our explicit reference to Y(u, ) in the last statement is in fact
superfluous; for simplicity, 91 could have been used as domain of definition of p.
The next invariance axiom lend itself to a similar remark; to economise on notation,
we state the following version:
Ch. 10: Social Welfare Functionals and InterpersonalComparability 495

* Inv*(qi(ui)): for any n-tuple of increasing functions (i)iGN defined on 9i, and
Vu, C 91N,

uR*v ((pl(ul),. .. , ,(u,))R* (1(Vl), .. , (n(un)).

The remaining invariance axioms can be simply stated as follows:

* Inv*(a + u): Va 91, and V u,v E 91N,

uR*v > (a+ ul,..., a + u,)R* (a +vl,. .a + ).

* Inv*(a + bui): Va, b E 91, with b > 0, and Vu,v E 91 N,

uR*v <> (a+bu,.. . , a + bu)R* (a+bvl,... ,a+bvu).

* Inv*(ai + ui): V(ai) E 91N, and V u, v 91N,

uR*v * (al +ul,..., a, +u,)R* (al + ... a, +vn).

* Inv*(ai + biui): V(ai) C 91N, V(bi) e 9N+, and Vu, v C N,

uR*v 4: (al +b ul,... ,an + b,u,)R* (al +bvl ..., a, + bv).

* Inv*((ai + bui)&4p(ui)): u, v, u', ' C

91N, (ai) E 91N, Vb E 91++, and for every
increasing function o defined on 91 such that, Vi N, u = ai + bui = qo(uj) and
V = ai + bvi = (p(vi),

u'R*U' > uR*u.

* Inv*(bui): Vb E 91, and Vu, v CE N+,

uR*u : buR*bv.

* Inv*(biui): V(bi) E 91N and Vu, v C 9+N+,

uR*v 4: (blul,.., bnu,)R* (blvl ... , bv,).

Notice that the last two axioms have been, for simplicity, defined on a subset of 91 N
with no change of sign, i.e. 1N+.

3.3.3. Separability, Continuity and Convexity properties

We turn next to another instance of parsimonious attitude towards information: what

do we do when some individuals are completely unconcerned by the issues at stake,
so that their evaluation function remains constant over X? Can the social ranking be
496 C. d'Aspremont and L. Gevers

affected by the level of their constant evaluation count? If we answer in the negative,
we express it as an interprofile statement known as Separability axiom:
Separability (SE). VU, V E D,
Ru =Rv if M C N such that i M, U = V
whereas, Vj C N\M, Vx,y X, U(x,j) = U(y, j) and V(x,j) = V(y,j).
If we accept this idea, we need not worry too much about the precise definition of N,
as long as all interested individuals are included. We notice here some analogy with
the axiom of Binary independence which allows the decision-maker not to worry about
the precise definition of X. We remark also that Inv(ai + uj) or any stronger invariance
axiom implies Separability.
The SWO translation, directly implied from Separability and formal welfarism, is
as follows:
Separability* (SE*): Vu, v, u', v' E 9V,
uR*v 4 u'R*v', if EM c N such that Vi M, ui = vi and u' = vi
whereas, Vj e N\M, u = uj and vj = v.
We introduce next the SWO version of a weakening of Separability which proves
interesting in the sequel. It restricts this property to the set of well-ordered vectors gN
defined at the end of Section 2.1:
9N-Separability (N-SE*): Vu, V,u L C N,
uR*v X u'R*v', if 3M C N such that Vi c M, ui = vi and u' = v'
whereas, Vj C N\M, u = ui and v = .
At this stage, we turn to an axiom relieving potential anxiety about slight
measurement errors in evaluation profiles: we shall require such errors to have a limited
bearing on the corresponding social rankings. This is explicitly an interprofile concern,
which requires defining a distance between numerical functions over a given domain.
We state the following as an example:
Continuity (ofF) (C):
Vx,y E X, VU ° E D and any sequence (U')i C D converging pointwise to UO,
if VI > 1, xRuiy, then xRuoy.
In the literature, continuity seems to have always been used in a formally welfarist
framework. Hence, it is defined directly as a condition on the social welfare
ordering R*:
Continuity* (of R*) (C*):
1
Vv E 9lN, the sets {u e 9NluRv} and {u e 9iNvR*u} are closed in t9.
We have the following:
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 497

Lemma 3.8. Continuity: Suppose an SWFL F satisfies Domain universality, Binary

independence, Pareto indifference and Continuity, then F has a continuous formally
welfarist associate R*.
Proof: Suppose R* does not satisfy C*. Then, for some v E fN, say, the set
{u iN: uR*v} is not closed, and it is possible to find a sequence (ul)'l in 9iN,
converging to some u, such that ulR*v for > 1, but vP*u° . Now, we may choose
x,y E X and, by UD, construct a sequence (U')~1 such that Uxl = ul and U = v, for
I > 1, and converging pointwise to some profile U0 such that U ° = u° and Uy° = v.
Then, by the last theorem, Vl > 1, xRUiy, but yPuox, that is, F does not satisfy C. E
In this result, we ensure formal welfarism through Domain universality. If we wanted
to use attainability instead, then the continuity axiom on F would have to be modified,
and some topological structure put on X, in order to derive the continuity of R*.
We conclude this section with another regularity axiom. According to Convexity,
society is never hurt if it moves toward a better alternative, wherever it may have to
stop along the way on feasibility grounds. In the SWO context, the formal version is
a straightforward application of the general mathematical definition:
Convexity'*.
Va Ef++, a < 1, u, v, w E N,
(w = au + (1 - a) v and uR*v) = wR*v.
The SWFL version is hardly less transparent:
Convexity:
VU E D, Vx,y,z X, Va c ++, a < 1,
(xRuy and Uz = aUx +(1 - a)Uy) X zRuy).

3.3.4. Alternative approaches to formal welfarism

The importance of formal welfarism makes it worthwhile to study the relationships of
Strong neutrality with seemingly weaker axioms.
What if, instead of permuting names among individuals, we permute labels among
alternatives? One can surmise that this arbitrary operation should not affect the social
ranking of the alternatives once they have been properly relabelled. The corresponding
axiom is obtained by weakening Strong neutrality or by strengthening Intraprofile
neutrality.
Interprofile neutrality (IRN):
Consider any permutation r: X -- X and any U, V e D,
such that Vx,y E X, Vy = Ux whenever y = (x);
then, Vx,y e X, xRuy if (x)Rv ((y).
Our readers are invited to check again the examples provided in Section 2.2: the
imposed SWFLs are the only ones violating Interprofile neutrality, as they violate
498 C. d 'Aspremont and L. Gevers

both Pareto indifference and Intraprofile neutrality. Although Interprofile neutrality is

necessary to obtain formal welfarism, it fails to be sufficient.
The Borda method (Subsection 2.2.7) illustrates this point: it satisfies Interprofile
neutrality, but it violates Binary independence and, hence, Strong neutrality. Suppose
X = {x,y,z}, N = {i, j} and U(x,i) = U(y,j) -= 1, whereas U(y,i) = U(x,j) = 2.
Then the Borda social ranking of x and y and the relation it induces between
U- = (1,2) = u and Uy = (2, 1) = v are highly influenced by U~, the evaluation
vector of the third alternative. Indeed, xluy if both x and y either Pareto-dominate z
or are Pareto-dominated by z, or if U is such that there is no Pareto domination at all.
If z is Pareto-dominated by x (resp. y) alone we get xPuy (resp. yPux). This pattern
gets reversed if Pareto domination goes the other way.
We can easily build on the above example and define a Borda SWFL with domain
D = {U, V}, for which uP*:v coexists with vP'u, consistently with profile-dependent
welfarism, but in opposition to the Binary independence requirement. Let us inquire
about the consequences of this state of affairs for R* as defined in Subsection 3.3.1: we
immediately realise that u*v by definition of R*. In this case, neither Ru nor Rv could
be recovered from a*, be it directly or not, because R* lacks discriminating power.
The following hypothetical example suggests that Interprofile neutrality might be
strengthened without going all the way to Strong neutrality. Suppose X is a large
set and we are asked to check whether for some pair of profiles (U, V), for some
permutation a of X and for every pair of alternatives (x,y), with y = o(x), V,, = U,.
In other words, we wish to find out whether the first sentence in the statement of
Interprofile neutrality holds true. Imagine now that V, = U for all pairs but one.
We surmise that the SWFL would seem abnormal if it were to prescribe both a
utilitarian Ru and a leximin Ry. As a possible strengthening of Interprofile neutrality,
the following axiom could be relied on for some integer m, where 2 < m < IX :

m-ary Neutrality:
Consider any two subsets A,B C X, such that A = BI = m,
any bijection a: A - B,
and any U, V D, such that x c A, y E B, y = U whenever y = (x);
then, x,y c A, xRuy if (x)Rv (y).

Which number m should we select? If m < m2, m -ary Neutrality obviously implies
m2-ary Neutrality, whatever the domain of the SWFL may be. On the other hand,
Strong neutrality and Interprofile neutrality correspond respectively to m = 2 and to
m = IXI, two values we chose to exclude a priori to prevent ambiguity.
Theorem 3.9. For all D C U, for all integers ml,m2 such that 2 < m < m2 < IX[,
Strong neutrality implies ml-ary Neutrality, which implies m2-ary Neutrality, which
implies Interprofile neutrality, which implies Intraprofile neutrality.
If the domain of the SWFL is universal, a partial reciprocal statement is also true:
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 499

Theorem 3.10. If D = U, if 2 < ml < m2 < IXI, m2 -ary Neutrality implies ml-ary
Neutrality, which implies Strong neutrality.
As Strong neutrality implies Binary independence, m-ary Neutrality implies m-ary
Independence, a property we introduce next:

m-ary Independence.
Consider any subset A C X, such that AI = m,
and any U, V C D, such that Vx E A, V = Ux;
then, Vx,y E A, xRuy if xRvy.

It is interesting to note that the last two theorems have an analogue dealing with
m-ary Independence, the method of proof being analogous. See d'Aspremont and
Gevers (1977), who draw their inspiration from Blau (1971).
Theorem 3.11. For all D c U, for all integers ml, m 2 such that 2 < m < m2 < IXI,
Binary independence implies ml-ary Independence, which implies m2-ary Indepen-
dence.
As before, a partial reciprocal statement holds true if the SWFL domain is
tmiversal:
Theorem 3.12. If D = 1l, if2 < ml < m2 < IXI, m2 -ary Independence implies ml-ary
Independence, which implies Binary Independence.

3.4. Pareto dominance principles and weak welfarism

At this stage, it is useful to bring in the familiar Pareto dominance principles. Their
popularity among economists and the criticisms recently levied against them warrant
some discussion. We start with a formal introduction, before we move on to interpretive
comments.
The least controversial axiom is known as Weak Pareto; it grants a higher social
rank to an alternative beating another one on every evaluation count:

Weak Pareto (WP):

VU E D, Vx,y EX, xPUy if U > Uy.

Most economists seem to agree with a stronger dominance principle, called hereafter
Strong Pareto:

Strong Pareto (SP):

VU E D, Vx,y EX, xRuy if U > Uy and xPuy if U > Uy.

The latter property implies Pareto indifference. We retained this formulation for the
sake of simplicity, and not on grounds of logical necessity. On the other hand, Pareto
500 C. d'Asprenont and L. Gevers

indifference is also implied by Weak Pareto, Continuity and Universal domain taken
together:
Theorem 3.13. If an SWFL satisfies WeakPareto, Continuity and Domain universality,
it satisfies also Pareto indifference.
Proof: Consider any U0 E , any x,y E X, such that U° = UJ°; by Universal domain,
we can form a sequence of profiles (Ul) I C U converging pointwise to U0, and such
that VI > 1, UL > U° , whereas U = Ut°. By Weak Pareto we obtain Vl > 1, xRuiy. By
Continuity, xRuoy follows. Since y can be interchanged with x in the above argument,
we observe that xluoy. D
If we adopt the formally welfarist structure of the previous section, our weak version
translates as follows:

Weak Pareto* (WP*):

Vu, v E 9N, if u > v then uP*v.

On the other hand, as Pareto indifference is subsumed by welfarism, the strong axiom
becomes

Strict Pareto* (SP*):

Vu, v E 9 iNv, if u > v then uP*v.

To discuss this set of axioms, we shall have to refer to the SWFL purpose and
to the meaning imparted to individual evaluation vectors. In what follows, we shall
maintain the value judgment interpretation of the Pareto principles, which tell us how
influential individual evaluations must be when they are in agreement about the relative
merits of any two alternatives; indeed, under these circumstances, the description of the
alternatives cannot affect the social ranking, and the ethical observer's own influence
is banished. As it transpires, we move quite far from the spirit of imposed SWFLs.
In the sequel, we have to distinguish whether the SWFL we analyse is utility-based
or not. If the SWFL is utility-based, we need distinguish again the constitutional
design problem from the social evaluation process. We realise that we can accept every
unanimous individual preference profile at face value in the former case, whereas we
cannot in the latter.
If the SWFL is meant to be used as a utility-based tool for social evaluation, the
ethical observer should make sure that all the individuals potentially concerned with
the decisions are represented in the unanimous appraisal process; they should also
be reasonably cognizant of the issues at stake. Moreover, all individual preferences
had best be self-centered; for instance, one cannot consider as socially acceptable the
unequal division of the fruits of a common property if all the owners have a valid
title to an equal share and one of them has either altruistic or anti-social preferences,
so that an equal split happens to be Pareto-dominated from the positive viewpoint.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 501

From this simple example, we remark that, although social choice theory may operate
formally from given individual preferences, it is sometimes felt appropriate to prune
or launder them informally prior to aggregating them into a social preference. Thus, a
utility-based interpretation of the domain of definition of an SWFL can be maintained
at the cost of reducing its scope.
We turn next to the alternative interpretation of our informational basis: what
happens if an ethical observer adopts a non-welfarist theory of the good and the
corresponding individual evaluation vector U, dominates some competitor Uy? It
seems quite sensible to apply a Pareto-like dominance principle, and to declare
that xPuy. However, one could perhaps make an exception when every individual
preference relation (possibly after laundering) points in the opposite direction. In this
conflicting case, one can surmise that the ethical observer should give in, out of respect
for the exercise of the free will at least when there is no conflict among individuals.
In effect, this could be achieved at the informal preparation stage of the aggregation
exercise by reinterpreting the set of possible alternatives X as the preference-based
Pareto set, so that the SWFL would be defined for a fixed preference profile at least if it
is to be utility-based. In another approach, suggested by the work of Kelsey (1987), the
SWFL should be enlarged to accommodate two criteria per individual instead of one:
both a non-welfarist criterion and some preference representation would be formally
taken into account by the ethical observer.
Be this as it may, one can also imagine to obtain a complete social ranking on the
basis of a sequential reasoning of lexicographic nature, whether the informational setup
is utility-based or not. For instance, an ethical observer having to choose between any
two alternatives x and y could begin to reason as any utilitarian: he or she would
add up individual evaluations for each alternative. If one of them dominates, it is
selected, and the process stops. If utilitarian sums turn out to be equal, then one
can invoke another set of arguments, possibly based on individual evaluation counts
too, but also possibly on other variables. If individual evaluation vectors happen to
coincide fully, one can also apply the same idea and rely on other variables to justify
a strict social ranking. In a remarkable paper, Roberts (1980a) managed to articulate
fruitfully such a sequential reasoning: his approach is based on an SWO R*, defined
N
over NH(X, D) = 9 N, 91+ or +N+
as in the previous section, so that his construct
is related to formal welfarism; however, it does not require Pareto indifference, so
that we have to content ourselves with weak recoverability. As a substitute for Pareto
indifference, the weak Pareto principle is an essential input in this theory; and either
minimal invariance axiom Inv(a + ui) or Inv(bui) is also invoked. We shall say that
F displays weak-welfarism (short for formal weak-welfarism) if and only if we can
associate with F an SWO R* with the following property:

Vu, v E '(X, D), Vx,y E X, VU E D, (u = Ux and v = Uy) (uPv = xPuy).

As the last statement implies, xPuy is not consistent with vP*u, whereas nothing can
be inferred from u*v, and we say that Ru is only weakly recoverable from R*. In
502 C. d'Aspremont and L. Geuers

keeping with our definition of weak welfarism, the trivial SWO, according to which
all individual evaluation vectors are socially indifferent, can be associated with every
SWFL. By itself, the weakly welfarist structure seems thus at first unpromising. In
particular, starred SWO properties can no longer be mechanically associated with
parent SWFL axioms as under formal welfarism.
Yet, as the following theorem suggests, the prospects for weak welfarism are better
than what one might think. Its key proposition is essentially based on Theorem I in
Roberts (1980a):
Theorem 3.14: Weak welfarism. IfF has 'H(X, D) = s91 (resp. N(X, D) = 9i +)
and if it satisfies Binary independence, Weak Pareto and Inv(a + Ui) (resp. Inv(bUi)),
there exists a unique SWO R* satisfying Continuity*, Weak Pareto*, Inv*(a + ui) (resp.
Inv*(bui)) and weak-welfarism associated with E
Two remarks are in order:
(1) Originally, weak welfarism was derived by introducing a technical condition of
"weak-continuity" [see Roberts (1980a), and the correction by Hammond (1999)].
Since these proposed weak-continuity conditions are implied by each of the more
transparent invariance conditions we require F to satisfy alternatively, we have
simply articulated them in the above presentation.
(2) Under the assumptions of the theorem, weak welfarism captures only the first
and foremost step in the lexicographic sequential reasoning we sketched earlier in
this section: however, the loss is limited due to Weak Pareto*. Uniqueness further
dissipates ambiguity. Moreover, as the ones mentioned in the theorem, all our
invariance axioms can be translated without being weakened.
Proof: Omitted, except for the uniqueness and invariance statements. To establish
uniqueness, we reduce ad absurdum the claim that, under the assumptions of the
theorem, F is consistent with two distinct SWOs, say R* and R*, satisfying C*, WP*
and weak-welfarism. We observe first that we cannot have both uP*v and vP2*u, by
definition. Suppose we have both uP*v and v*u. Then we can choose some w close
enough to u and such that w << u and vP*w, by Weak Pareto* and transitivity, whereas
wPju, by Continuity*. Thus, we obtain the announced contradiction.
We show next that R* also satisfies Inv*(a + ui) in three steps:
Step 1. It does satisfy a property weaker than Inv*(a + ui), viz. for any u, v E 9lN and
any a E 9, uP*v implies (a + ul,..., a + u,)R* (a + vl,..., a + v,). This is
almost immediate by weak recoverability and because H(X, D) = 9 NV since,
whenever uP*v for any u,v E 9N, we may find V in D and w,z E X
such that u = V, v = V, and wPvz. Then, by Inv(a + U): VV' E D,
Va C 9l, Rv, = Rv if V(x,i) X x N, V'(x,i) = a + V(x,i), so that
(a + ul,...., a +u,)R* (a + v, ... , a + ,,), because F displays weak welfarism.
Step 2. We show that uI*v X (a + ul,... ,a + u)I*(a + ul,... ,a + v,) for every
a E 91, and any u, v E W1N.Let ul*v and consider the dominating sequence
(uk)I=l C 9N, k, Uk > u (resp. uk >> v), converging pointwise to-
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 503

wards u (resp. v). By Weak Pareto* and transitivity together with Step 1, we ob-
tain: Vk, ukP*u, ukP*v and Va E 9i, (a+ u,... ,a+un)R*(a + l,.. , a + ),
(resp. (a + u,..., a + un)R* (a +ul,...,a + un)). By Continuity*, we obtain
(a + ul,...,a + u)R*(a + vl,...,a + v), (resp. (a + vl,...,a + v,)R*
(a + ul, ... , a + u,). The result follows.
Step 3. Since every invariance transformation is invertible, the result of step two can
be reversed: (a + ul,..., a + u,)I* (a + ul,... ,a + v,) = uI*v; this is incon-
sistent with uP*v. Thus, uP*v = (a + ul,... ,a + u,)P* (a + vl,..., a + u,)
and Inv*(a + ui) is established. The proof pertaining to Inv*(bui) is almost the
same. D
For instance, the maximin SWO, viz. u, v E N, uR*v mini ui > mini vi, is
the unique SWO associated with the maximin SWFL satisfyingformal welfarism, and
it can fruitfully be associated with both the maximin SWFL and the leximin SWFL
under weak welfarism, whereas the leximin SWO is the unique formally welfarist (with
full recoverability) associate of the leximin SWFL. The maximin SWO satisfies Weak
Pareto* and Continuity*, whereas the leximin SWO satisfies Strict Pareto* but fails to
be continuous.
Let us consider next Separability*. Even though the assumptions of the last theorem
are satisfied, they do not suffice to obtain it, unless it is implied by another property,
such as an invariance axiom, and it must be replaced by a weaker version which proves
rather unwieldy. 14 For instance, the maximin SWO does not satisfy Separability* even
though the leximin SWO passes the test.
In conclusion, let us recapitulate some remarks about the properties of the Pareto
axioms. Even though their direct influence is very limited since it is felt only in the
absence of social conflict, the very incomplete social ranking they induce in isolation
does display a great deal of informational parsimony. This has two aspects: once
established, the Pareto ranking of any two alternatives is fully independent of the
individual evaluations of the remaining alternatives, and moreover, the social ranking is
simply based on the rankings underlying the individual evaluations: it is consistent with
the strongest invariance principle, i.e. Inv(qi(Ui)). As the analysis of (weak) welfarism
shows, they have remarkably profound implications when they are associated with
appropriate information filtering axioms.

3.5. Equity axioms

In a spirit of impartiality or equity, the Pareto-inspired principles may be directly
extended to the cases where individual names have been permuted in one of the vectors
U,, Uy. This is known as the Suppes (1966) Grading principle 15. We proceed by stating

14 By an argument analogous to the one used in Step 1 of the proof of the last theorem, the following
SWO property is implied if SE and weak welfarism hold: Vu, v, u', v' E 9N, uP*v > u'R*v', if 3M c N
such that, Vi C M, ui = vi and u = v whereas, Vj N\M, uj = u and ivrj. -
15 See also Suppes (1957).
504 C. d'Aspremont and L. Gevers

its weak form. To simplify, we shall represent a permutation of the players by a nxn
permutation matrix r where each element is either a 0 or a 1, and each line and each
column contains a single 1.

Weak Suppes dominance (WS).

For all permutations Jg, VU G D, Vx,y E X, xPuy if Ux > JTUy.

Much as the equity content of Weak Pareto may be reinforced to generate Weak Suppes
dominance, we can reinforce Pareto indifference to obtain

Suppes indifference (SI):

For all permutations r, VU c D, Vx,y E X, xluy if U = LrUv.

The former axiom which implies Weak Pareto, induces a less incomplete social
ordering than the latter since it can solve some social conflicts. Both incomplete social
rankings satisfy Binary independence equally well, but Weak Suppes dominance proves
much more demanding informationally. On the other hand, Suppes indifference which
implies Pareto indifference, retains some bite within the welfarist structure; it is also
much more demanding than the latter from the invariance viewpoint, at least when
Weak Pareto is also imposed. These observations are made precise in the following
Theorem 3.15. Suppose F has universal domain D = U. Then, if it satisfies Weak
Suppes dominance, it contradicts both Inv(ai + biUi) and Inv(cip(Uj)). The same
conclusion is valid if IXI > 3 and both Suppes Indifference and Weak Pareto hold.
Proof: We prove first the statement pertaining to WS. Actually, we can rely on positive
evaluation counts. Suppose N = l, g,k}. Consider any U e U, any x,y E X
such that U(y,l) > U(x,g) > U(y,k) > U(x,k) > U(y,g) > U(x,l) > 0.
By WS, we get yPux. We proceed by constructing a profile V, which is infor-
mationally equivalent to U by Inv(ai + biUi), and yet which is chosen so that
V(x,g) > V(y,l) > V(x,l) > V(y,k) = U(y,k) > V(x,k) = U(x,k) > V(y,g) > 0.
First of all, we let V(x,l) = U(x,l) + U(y,k), V(y,l) = U(y, 1) + U(y,k)
and V(x,g) = bU(x,g), where b > (U(y,l) + U(y,k))/U(x,g). By construction,
V(x, 1) > U(y, k), V(y, 1) > V(x, 1) and V(x,g) > V(y, 1), and these three inequalities
hold for any arbitrary small pair U(y, g), U(x, 1), with U(y, g) > U(x, 1), in particular
if U(y,g) < U(x,k)/b. Indeed, if the latter inequality holds, U(x,k) > V(y,g)=
bU(y,g). By WS, we obtain xPvy despite yPux, contradicting both Inv(ai + biUi)
and Inv((pi(Ui)). If there are other individuals in N, we simply clone individual k. If
N is a pair, we eliminate k.
To prove the statement involving both SI and WP, we go back to the case pertaining
to three individuals, with the pair of profiles U and V constructed as above. Moreover,
by UD, we can choose z E X so that U(y, 1) > U(z, 1) = U(x, g) > U(y, k) > U(z, k) =
U(x, k) > U(y,g) > U(z,g) = U(x, ). By SI, we have zlux, whereas we have yPuz
by WP, so that yPux. We can pick V and a fourth element in X, apply UD, SI and WP
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 505

again to conclude xPvy. Eventually, we invoke Inv(aj + bi Ui) and we have the same
contradiction again. D
Other axioms are interprofile statements, and they have no meaning out of the SWFL
context since they relate the respective images of two distinct profiles; they tell us
whether the social ranking ought to change and, if yes, in what direction whenever
some profile undergoes a specific change. This may be exemplified with help of the
Anonymity axiom: if two profiles are the same once the individual evaluation functions
have been permuted in one of them, impartiality would recommend that they both be
assigned an identical social ranking by the SWFL.
Anonymity (A):
For all permutations r, VU, V E D, Rv = Ru if Ux = Vx, Vx C X
Interested readers will easily check our earlier examples: they will find out that most
of them satisfy Anonymity. In association with any Pareto axiom, Anonymity may
thus be consistent with the strongest invariance axiom. This is in contrast with its
welfarist version; indeed, both Anonymity and Suppes indifference have the same
welfarist translation (among others, a proof can be found in d'Aspremont and Gevers
(1977)):
Anonymity* (A*):
For all permutations if, Vu, v E 9 N, uI*v if v = Jru.
Requiring an SWO to satisfy both Anonymity* and Inv*(ai +biui) leads to the most
unpalatable consequences. This is a direct implication of a theorem by Krause (1995)
summarized in the next section. We provide here a simple direct proof:
Theorem 3.16. Suppose R*, an SWO defined over N , satisfies both Anonymity*
and Inv*(ai + biui) (resp. Inv*(pi(ui))). Then all individual evaluation vectors must be
socially indifferent.
Proof: Let us pick any u, v eC We also avail ourselves of w, s, r, a, b E 9N, such
N
that w1 < minful,u 2 ,... , u,,vIu,v 2 ...,,) maxful,u 2 ,..., ,,,Uvl,v 2,...,vUn <
W2 < W3 < · · < W, and s = (w, wl, w2,... wnl), whereas Vi N, ri = ai + biwi and
ui = ai + bisi. Furthermore, a is such that r = w; i.e. a = w - bw. By substitution, we
obtain u = w + b(s - w), so that, by definition of s, we can write: ul = wl + bl(w, - wl),
u2 = W2 + b2 (wi - w2), u3 = W3 + b3 (w 2 - W3 ), etc. By A*, wI*s. By Inv*(ai + biui),
uI*w, provided Vi E N, bi > 0. The latter condition can always be met if ul > wl,
u2 < W2 , u3 < W3, etc. Notice that we have defined w so that the same reasoning may
be applied to prove that vI*w. By transitivity, we conclude that uI*v. D
Another intraprofile property, which is implied by Suppes indifference, simply
requires that there be a very minimal symmetry in the treatment of any two
individuals 16 . It sets a limit on the influence any individual can exert on the social

16 This condition was introduced in d'Aspremont (1985), where it is called Weak Anonymity.
506 C. d'Aspremont and L. Gevers

ranking when he/she has a single opponent. As such, it does not rely on any
interpersonal comparison.

Minimal individual symmetry (MIS):

For any two individuals i,j C N,
3U E D, 3x,y CX such that
U(x, i) > U(y, i), U(x, j) < U(y,j), U(x, I) = U(y, 1), V1 {i,j},
i and xlj.

This requirement is satisfied by both the Borda method and the method of majority
voting. If formal welfarism holds, we obtain the following version, which is expectedly
much weaker than Anonymity*:

Minimal individual symmetry* (MIS*):

For any two individuals i,j E N,
3u, v C 9 jN such that ui > vi, uj < j, ul = ul, Il {i,j}, and uI*u.

We turn next to three more demanding equity conditions. The SWO form of the
well-known Pigou-Dalton principle comes first.

Pigou-Dalton*principle.
V i,j E N, Vu, v E 9N, VE E 9,,,
vR*u if (1) vj > Vi, v = U - E, i = ui + E and (2) V1 8 {i,j}, vl = ul.

A transfer from a relatively better-off individual to a relatively worse-off, without

reversing their ranking, is weakly improving socially. A strict version of this axiom
can also be found in the literature. If the SWO R* is represented by a social evaluation
function W from 9 iN to 9i (i.e., uR*v X W(u) > W(v)), then our version of the Pigou-
Dalton principle is equivalent to Schur-concavity: viz. for any n x n doubly stochastic
matrix a, for any u, v C 9 N, vR*u if v = ou, whereas Convexity* of R*, the last
axiom of Subsection 3.3.3, is equivalent to quasi-concavity of its representation, and
this implies Schur-concavity if R* satisfies also both Weak Pareto* and Anonymity*.
See Moulin (1988).
Our last two equity axioms must be compared to the Pigou-Dalton* principle, as
their opening statement is identical. The first one is known as Hammond's equity axiom
[see Hammond (1976a, 1979)]:

Hammond's equity* principle.

Vi,j C N, Vu, v c 9, Ve, lE 9i-+,
vR*u if (1) vj > vi, vj = uj - , vi = ui + E and (2) VI ~ {i,j}, v1 = ul.

We remark that the triggering condition implies u > > vi > ui, a slight change
from the previous axiom; yet, it is of much wider scope in the new axiom, since there
Ch. 10: Social Welfare Functionals and InterpersonalComparability 507

is no requirement such as /E = 1. A priori, the latter ratio is not bounded above

(e can be arbitrarily small), and this new feature is controversial, despite the fact that
the axiom conclusion is phrased only as a weak social preference relation. Our last
axiom is introduced by Blackorby et al. (2002) under the name

Incremental equity*.
Vi,j E N, Vu, v E 9iN, E 9i,
vI* u if (1) vj = u - , v i = ui + E and (2) V1 d {i,j}, vl = ul.

Here also, we observe a widening of the triggering condition, in comparison with

the Pigou-Dalton* principle; indeed, we have lost any reference to the relative welfare
levels of i and j, and the size of the welfare transfer is unrestricted. On the other
hand, the conclusion vI*u is more demanding than the version we ascribe to Pigou
and Dalton. In view of these remarks, we doubt that the above axiom will appeal to
many as an equity property.

4. Independence and invariance-based characterisations

In order to uncover some ethical consequences of the axioms we have already

introduced, it seems best to find out which social rankings they lead to. Our main
purpose, in this survey, is to reveal the implications of declaring irrelevant various
kinds of interpersonal comparisons. Ideally, we should investigate every invariance
axiom we have defined. Our intention is not to be exhaustive but simply to illustrate
the existing results. On occasion, we shall introduce new ones. For convenience, we
shall cut this section in three parts based on as many clusters of axioms. Moreover, our
theorems are often phrased in SWO language. Except for some special cases which are
mentioned explicitly, formal welfarism is assumed to hold and the SWO pertains to the
full domain 9l N. In the next section, we also display some characterisations dispensing
with invariance and/or independence axioms, at the cost of introducing other kinds of
restrictions.

4.1. Restricting interpersonal level comparability

We begin this subsection with four invariance axioms, which we list in order of relative
logical weakness: Inv*(pi(ui)), Inv*(ai + biui), Inv*(ai + bui) and Inv*(ai + ui). Each
of the first two excludes interpersonal comparisons and leads to Arrow's negative
conclusion, viz. weak dictatorship. From either of the last two, allowing for some
interpersonal comparisons, we will derive something more palatable, the family of
weak weighted utilitarian rules (clearly compatible with both of them). All three rules
are weak in the sense that they do not define completely an SWO. Moreover, as we
allow some weights to be nil in the weak utilitarian rule, weak dictatorship is a special
case (where all weights are nil except one). We will indicate how these variations of
508 C. d'Asprenmont and L. Gevers

classical results can be easily completed, in particular how pure utilitarianism can be
obtained. We shall also innovate (1) by introducing a slight weakening of Inv*(ai + ui),
(2) by providing a new characterisation of the "rank-weighted" utilitarian SWOs,
among which the generalised Gini SWOs make up the most attractive subfamily; we
do this with help of a new invariance axiom, obtained by restricting Inv*(ai + bui) to
the subset of well-ordered evaluation vectors 9N.
As it turns out, to derive weak weighted-utilitarianism, we adapt the proof of
Blackwell and Girshick (1954), Theorem 4.3.1, based on a supporting hyperplane
argument. Instead of Inv*(ai + ui) (their condition L3), we use the weakened invariance
condition

Weak Invariance with respect to individual changes of origin* (Wlnv*(ai + ui)):

Va E 9iN, Vu, v E 91N,
uP*v >(al + ul,..., a, + u,)R* (a + a + v,)).
al,..,

Moreover, we replace their dominance condition L2 (i.e., u > v > uR*v) by

Weak Pareto* (Section 3.4). The result is strengthened if we add Minimal Individual
Symmetry* or Anonymity* (Section 3.5).
Theorem 4.1: Weak weighted utilitarianism. If an SWO R* satisfies Weak Pareto*
and WInv*(ai +ui) (resp. Inv*(ai + ui) or Inv*(ai + bui)), then there exists A C 91\{o}
such that Vu, v G 91N

E ,iui > E Si uP*v.

iEN icN

Moreover, if we add Minimal individual symmetry* (resp. Anonymity*), we must have

every component of A strictly positive (resp. strictly positive and equal).
Proof: Let P _ p 91N: p > 0}, S _ {s 9N: sR*e}, with e = (0,...,0),
and Q = {q E 9 N: q = s +p, s S, p P}. By WP*, for any s S and
p P, (s +p)P* sR* e, so that P C Q c S. Since Q = UsEs(P+ s), it is open
in 91N with e as a boundary point. Thus, if we can show that Q is convex, there is
[see, e.g., Theorem 2.2.1 in Blackwell and Girshick (1954)] a supporting hyperplane
to Q through e: E N - {e} such that, for all q Q, i=Xiqi > 0. Moreover,
Ai > 0 for each i, since P C Q. Finally, for any u,v E 9N, uR*v implies (by
WP*) that (u + Op)P* v, for any p P and 0 C (0, 1), and (by WInv*(ai + ui))
that [(u - v) + Op]R* e so that [(u - v + Op) + (1 - O)p] E Q. Therefore, uR*v implies
Zi= ,Ai(ui- vi +pi) > 0, for any p P, and hence Z= 1 Ai(ui - vi) > O. This is the
result. To end the proof, it just remains to show that Q is convex.
For s, s' S, p,p' E P, we have (s +p), (s'+p') Q, implying (s +p)P*e and, by
WInv*(ai + ui), (s +p +s') c S (since (s +p + s')R* s' R* e) so that (s +p + s' +p') e Q.
Therefore Q is closed under addition, and to show that it is convex it is enough to show
that ugq e Q whenever q · Q and p > 0.
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 509

We first show that, for any s E S, p E P, for all positive integers k,m,
and any 0 (0, 1), [(k/m)(s + p)] E S. Suppose not. Then, for some positive
integers k,m and some 0' (0, 1), eP* [(k/m)(s + p)] for all positive 0 < 0'
(using WP*). By WInv*(ai + ui), we get [-(k/m)(s + Op)] R* e, for all positive 0 < 0'
and then: [-(k/m)(s + O'p)]P* [(k/m)(s + 0'p)], implying eR* [2(k/m)(s + O'p)] and
eP* [2(k/m)(s + O"p)] for 0 < 0" < 0'; also [-(k/m)(s + O"p)]P* [2(k/m)(s + O"p)],
implying eR* [3(k/m)(s + O"p)] and eP* [3(k/m)(s + 0"'p)] for 0 < 0"' < 0"; and so
on, until we get eP* [m(k/m)(s + Op)], for some positive 0 (and all 0 E (0, 0), by WP*).
However, for any 0° E (0, 0), (s + 0 °p) P* e, a contradiction if k = 1. If k = 2, we may
use WInv*(ai + ui) to get eR* [-(s + 0°p)], which implies (s + 00p) P* [-(s + 0°p)] and
(using WInv*(ai + ui) and WP*) [2(s + d 1p)] P* e for 01 E (0 ° , 0), also a contradiction.
If k = 3, since eR* [-(s + Olp)], we get [2(s + Olp)]P * [-(s + 0 1p)], and (using
WInv*(ai + ui) and WP*) [3(s + 02p)]P * e for 2 (1, ), again a contradiction.
And so on, for any value of k. Therefore [(k/m)(s + Op)] E S for all positive integers
k and m, and any 0 E (0,1).
Now, choose any q = s +p (with s E S, p E P) and any p > 0. We can find c a pos-
itive rational close to p, and some p' e P, such that y(s +p) = t(s + Op + (1 - O)p) =
K(s + Op) +p', with K(s + Op) S, by the above. Hence, y(s +p) C Q, so that Q is
convex. The last clause of the theorem proceeds from a reductio ad absurdum. E
This is a derivation of weak weighted utilitarianism, which does not characterise
completely an SWO, because it says nothing about the social ranking of any two
alternatives whose numerical utilitarian evaluation is the same. In order to obtain a
complete characterisation of weighted utilitarianism as an SWO, (1) we use Continuity*
(see Subsection 3.3.3) to get indifference hyperplanes, and (2) we replace Weak Pareto*
by Strict Pareto* in order to ensure that all weights be positive; (3) alternatively, to
obtain the same result, we maintain Weak Pareto* and we add Minimal individual
symmetry* or we can rely on the latter two axioms and restrict ourselves to the
more demanding invariance axiom, viz. Inv*(ai + bui)}, Weak Pareto* and Minimal
individual symmetry* while we drop Continuity*.
Theorem 4.2: Weighted utilitarianism.
(1) Suppose an SWO R* satisfies Continuity*, Weak Pareto* and any element of the
triple {WInv*(ai + ui),Inv*(a + ui),Inv*(ai + bui)}. Then, R* is a member of the
weighted utilitarianfamily: there exists 9+f\{0}, and Vu, v C RN ,
uR*v a E niui > E hivi.
ieN iGN

(2) Moreover, if either Strict Pareto* is substituted or Minimal individual symmetry*

is added, we must have every component of X.strictly positive.
(3) The latter result holds also if R* satisfies only Weak Pareto*, Inv*(ai + bui) and
Minimal individual symmetry*.
Proof: (1) By the above theorem, we know that there exists E 9\{0},
and Vu,v C 9 N, EENAiUi
i > iNAiUvi implies uP*v. We first prove that
510 C. d'Asprenmont and L. Gevers

Ei CN iUi = EiN,,aiUi implies u*u, whenever C* holds. Suppose, instead,

Ei N iui = i iN ivi and uP*u for some u, u C 9 N . Since the set {v' GE N I u'P*u}
is open by C*, there is some v' E N in a neighbourhood of u such that u'P*u and
Ei CN iui > i c N ivy, a contradiction.
(2) 2 > results from SP* or MIS*, as in Theorem 4.1.
(3) Using Inv*(ai + bui) and MIS* we may show now that, for any u, v E 9l ' , uI*v if
and only if ZieNliui = Ei CN i vi. We know that, for all u, v eCR9', uR*v implies
i N iUi > i C N ii, and, hence, uI*v implies Ei E N U= iui EiE N2li The
converse of this last implication can be proved by recurrence. Assume that, for an
integer m, 1 < m < n, and for any u,v C 9 N satisfying ui = vi for all i > m,
Ei EN AiUii = Li v Xivi X uJ*v (of course this holds trivially for m = 1, assuming
w.l.o.g. that 2u1> 0). We need to show that the same implication holds if we take
any u',v' E 9 N, with (whenever m+ I < n) u = v' for all i > m+ 1. So,
suppose CiN
c iwi' = 0, for w' -_ u' - •4 e. We want w'I*e. Observe that,
by Inv*(ai + bui), the set w E 9JN 1 w*e} is convex' 7. Then, using MIS* for
appropriate pairs of individuals and taking convex combinations, we can construct
u", v" E 9 N such that u" • v" and u"I*v", or, by Inv*(ai + bui), w" (u" - v")I* e,
and hence i N iwi' = 0, and such that sgn(w",+) = sgn(w',,) and w = 0 for all
i > m + 1 (whenever m + 1 < n). If we let b > 0 be such that bw+ = w',,, then
Ei c N 2i(w' - bw"7) = 0, and wt - bw7" = 0 for all i > m, implying (w' - bw") 1* e, or,
using Inv*(ai + bui) twice, w'I*bw"I*be = e. Therefore, for any u, v EC9", uI*v if
and only if w (u - v)I* e if and only if Li ENniui = Li CN ,ivi. Again, > 0
results immediately from MIS*. [
By taking advantage of weak welfarism, we can also derive from Theorem 4.1 the
SWFL version 18 of Arrow's (1963) General Possibility Theorem.
Theorem 4.3: Weak dictatorship. If an SWFL F satisfies Domain universality, Binary
independence, Weak Pareto and Inv(i(Ui)) (resp. Inv(ai + biUi)), then it is weakly
dictatorial:i.e., there exists i e N, such that, Vx,y G X, VU E U,

U(x, i) > U(y, i) xPuy.

Proof: By weak welfarism, there is an SWO R* satisfying WP* and Inv*(qpi(ui)) (resp.
Inv*(ai +biui)), and hence WInv*(ai + ui). Then, by Theorem 4.1, R* is weak weighted
utilitarianism for some non-negative vector of weights A 0.

17 By Inv*(ai + bui), for wl*e and w'le, and for 0 E [0,1], OwJIe and (1 - )w'JIe, so that
(Ow +(1 - 0) w') (1 - 0) w'Ile.
18 This is slightly less general than Arrow's original results in terms of social welfare functions, defined
on profiles of individual preference orderings, since not all preference orderings are representable by
utility functions. Arrow's Independence of irrelevant alternatives can be seen as the conjunction of BIN
(see Subsection 3.3.1) and Inv(qi(Ui)).
Ch. 10: Social Welfare Functionals and InterpersonalComparability 511

Now, take x,y X, and U E U such that U(x,i) > U(y,i), for some i,
U(y, j) > U(x,j), for all j i, and EjcN AXj(U(x,j) - U(y, j)) > 0. Hence, letting
u Ux and v - U,, we get uP*v and, by Inv*(qi(ui)) (resp. Inv*(ai + biui)):
(UI,PU2 , ... , fu,) P*(vl,13U2, ... , /fv,), hence Aij(ui - vi) > /3Ej i.j(vj -uj) > 0, for
any / > 0. This is impossible, unless 2; = 0, Vj • i. Weak dictatorship follows. []
As suggested by Luce and Raiffa (1957 p. 344), such a result (and the argument
of Theorem 4.1), can be used repeatedly to get weak lexicographic dictatorship, i.e.,
there exist a subset of N and a permutation of its members such that the social ranking
always mimics the strict preference of the first one, and in cases he or she is indifferent,
moves on to mimic the strict preference of the following member, and so forth until the
subset is exhausted. By dropping 19 Weak Pareto (while assuming Pareto indifference),
Krause (1995) provides a neat generalisation of Theorem 4.3: the set N is exogenously
partitioned in three subsets, one of which consists of dummies who are denied any
influence on the social ranking, so that society is indifferent if and only if all the
remaining individuals are indifferent. The strict preference of the latter is taken into
account in lexicographic fashion, the order of priority being given exogenously. Yet, the
nature of the influence exerted by each member of the two remaining subsets depends
on the one they belong to. Society mimics the strict preference of members of one of
them, whereas it reverses the strict preference of members of the other one. Adding
Anonymity as another requirement in the same context precipitates social triviality, as
we already showed in Theorem 3.16.
In contrast, availing ourselves of Anonymity*, Weak Pareto* and Inv*(ai + ui), we
can characterise pure utilitarianism with help of a beautifully simple proof argument
due to Milnor (1954)20, to recall
Theorem 4.4: Pure utilitarianism. An SWO R* is pure utilitarianism if and only if
it satisfies Weak Pareto*, Anonymity* and Inv*(ai + bui) (resp. Inv*(ai + ui)).
Proof: Necessity is clear. To prove sufficiency, suppose first that two vectors u and v in
91N add to the same amount. By A*, we can permute their components to get two new
vectors in increasing order, but mutually ranked as u and v. Considering successively,
in these new vectors, each pair of corresponding components and subtracting from
each the minimal one, we get again two new vectors which by Inv*(ai + ui) (or
Inv*(ai + bui)) are again mutually ranked as u and v. Repeating these two operations
at most n times, we finally get two vectors with all components equal to zero, but
still mutually ranked as u and v. Therefore u and v should be indifferent. Second, to
prove that ZiENui > uP*v, we define Vi
=iENvi N, wi = ui - 6, where
6 = (l/n)(Zi EN Ui - Ei CN vi), and, to get uP*v, we simply combine the preceding
argument to obtain wI*v with WP* to obtain uP*w. D

19 A seminal paper along this line is Wilson (1972).

20 As shown in d'Aspremont and Gevers (1977).
512 C. d 'Asprenmont and L. Gevers

Denico16 (1999) points out that the above result and its proof remain valid if
we assume R* to be a reflexive binary relation defined on 91N that need not be
transitive. The underlying structure must be more general than the SWFL concept,
since its image set does not consist only of orderings of X. What amounts to a
slight strengthening of A* within this more general context is however required, to
wit: For every permutation Jr of N, Vt, u, v 9 jN such that v = :ru, we must have
uR*t X vR*t.
Alternative characterisations of pure utilitarianism can also be obtained straightfor-
wardly from Theorem 4.2.
Corollary 4.5. An SWO R* is purely utilitarian if and only if it satisfies Weak Pareto*,
Anonymity*, Wlnv* (ai + ui) and Continuity*.
Other implications of Anonymity can be derived with help of an argument inspired
by the proof of Theorem 4.1, provided we restrict invariance to the well-ordered
space 9N (defined in Section 2.1) in imitation of Weymark (1981)21. This is

5N-Invariance* with respect to common resealing

and ordered individual changes of origin (N - Inv*(ai + bui)):
Vb > O, Vu, , a N, uR*v 4, (a + bu)R* (a + b).

We then get the SWO which is the formally welfarist associate of the rule defined in
Subsection 2.2.5.
Theorem 4.6: Weighted rank utilitarianism. If an SWO R* satisfies Continuity*,
9N - Inv*(a, + bui), Weak Pareto* and Anonymity*, then, there exists E 9i\{0}O
such that Vu, v 91N,

: Akui(k) '>E kVi(k) uR*v.

kEN kiCN

Proof: Using the notation of Theorem 4.1, we define the sets p _ (p E 91' p > 0,
Pl < P2 < ' < p,}, S(u) - s E N I sR*u}, and Q(u) - q E GN q = s +p,
s E S(u), p CE P}, for any u C 9N. By WP*, (s + p)P*sR*u, so that
Q(u) c S(u). We remark that S(u) is convex by N - Inv*(ai + bui); indeed, for
sR*u and s'R*u, and for 0 [0,1], OsR"Ou and (1 - 0)s'R*(l -)u, so that
(Os + (1 - 0) s') R* (Ou + (1 - 0) s'), and (Ou + (1 - 0) s') R* (Ou + (1 - O)u). Therefore,
(Os + (1 - 0)s')R* u. Clearly u is a boundary point of Q(u), and, since P is open
in 9 N and convex trivially, Q(u) is open in 9N and convex in view of the last

21 Weymark (1981) characterizes the generalized Gini absolute inequality indices. See also Bossert
(1990). Notice that our restricted axiom is too weak to be helpful in Theorem 4.4: Milnor's argument
would not go through.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 513

remark. Therefore, there is a supporting hyperplane to Q(u) through u: there exists

u
E 9 N\{0 }, and Vq E Q(u), Ei ,GN 'qi > Ei N ui. Also, Ai > 0, Vi E N, since
otherwise taking q E Q(u), with qi large enough, would violate the inequality. Now,
by contraposition, Vv cE vN,if Zi EN ~yUVi < Ei N Ui, then v Q(u). Moreover,
we cannot have R*u. Otherwise, we could find p E P such that (v +p) E Q(u) and
Ei EN i(vi +Pi) < i e N XAui, a contradiction.
Assuming w.l.o.g. that Ei NX = 1, for any u GE N, we may show that, for
u = (,..., u) and any v E N,

(eN I iEN ) I

Indeed, by C*, we cannot have vP*u, since then, for p > 0 small enough, we
would have (v -p)R*u implying that v E Q(u), in contradiction with ieN,2 IVi =
yEi N X'Ui. Moreover, if uP*v, then, for p > 0 small enough, we would have
(u -p)R*v implying that u e Q(v), and hence

u = YE ui > E )L:vi = E iUVi = E XUi = ,

iGN icN ieN iEN

again a contradiction. Therefore: vI*u.

Finally, to show that all lus are equal, it is enough to take u = (ii, ... , )
and v = (D, ... , D) in N and show that their associated weights, ,"u and U
respectively, are equal. Suppose 2 u" ),I and assume w.l.o.g. that both ii and are
positive (otherwise, by N - Inv*(ai + bui), we would translate the two hyperplanes
HU = {w e N : Ei cN X'Wi = a} and H ° = w E N : i CN iuWi = D} by adding
a > 0 to both fi and D,while preserving the social welfare ordering between all pairs
of points in Hu U HV). Define the hyperplane Hb = {w E N : Ei EN i2Wi = b},
with b > 0 in such that i = b. By 9N - Inv*(ai + bui) again, for any w, wI E H',
wR*w' bwR*bw' and bw, bw' Hb y. Then, since X;' • 2O, there is some
°
w° e H u \ Hb° and some w E HbU such that, say, EiEN uW < i EENZiWi
I °
and w > w . By WP*, this is in contradiction with w0I*ul*wl. D
Substituting SP* to WP*, all weights become positive. As we shall see later, we
can characterise the generalised Gini family of SWOs by adding the Pigou-Dalton*
principle to our list of axioms.

4.2. Full comparability:from cardinal to ordinal measurability

We proceed by studying the implications of invariance axioms allowing at least to
compare evaluation levels interpersonally, and not weaker than the one based on
common positive affine transformation. In other words, we deal not only with the polar
cases Inv(((Ui)) and Inv(a + bUi), but also with their intermediary links Inv(p(AU))
514 C. d'Aspremont and L. Gevers

and Inv({ai +bUi }&(Ui)). We adopt again the language of social welfare orderings to
simplify. Application to SWFLs proves straightforward, when formal welfarism holds.
Under weak welfarism, care may be required at least when Separability holds. The
extent to which Separability may be imposed may vary: this provides a criterion for
dividing our presentation.

4.2.1. No Separability

We start with Inv*(a + bui) and report first about its implications when the SWO
satisfies Weak Pareto* and Continuity*. In order to gain some intuition, we assume
initially that N reduces to the pair {i, j}. The family of SWOs we are led to describe
is known as the min-of-means family: it encompasses the Generalised Gini family of
orderings. We first split 29 by means of the main diagonal to obtain two rank-ordered
subsets, viz. {u E 92 I ui > uj} and {u c 93I2 ui < uj}: for each of them, we are
given a pair of non-negative individual weights and we apply weighted utilitarianism
separately; social indifference half lines meet on the 45-degree line. This geometric
sketch translates formally as follows:
Theorem 4.7. If N = {i,j} and R* satisfies Inv*(a + bui), Weak Pareto* and
Continuity*, there exist two orderedpairs (, 2.j), (i, ,uj) c 912 summing to one and
such that, either Vu, v e 912, uR*v (+ min{Au, ,uu} > min{kv, ,uv}, or Vu, v E 92,
uR*v X max{lu, lu} > max{v, !uv}.
A proof may be found in Bossert and Weymark (2000), Theorem 5 or in Deschamps
and Gevers (1977). This straightforward result proved difficult to generalise: a beautiful
solution is provided by Gilboa and Schmeidler (1989) and adapted to our context by
Ben-Porath et al. (1997). Notice that the expression min{Au, u} (resp. max{au, u})
used in Theorem 4.7 can equivalently be replaced by min c [o,l]{au + (1 - a)tu}
(resp. max, E[0,]{a2u + (1 - a)1tu}), and this seemingly more complicated presenta-
tion involving a line segment is in effect the nondegenerate two-dimensional version of
the compact and convex set which is the key element of the SWO representation they
provide. Their theorem characterises the min-of-means family of SWOs. Each member
of the family may be defined as follows:

Min-of-means SWO:
An SWO belongs to the min-of-means family if and only if there exists some
compact and convex subset of weights A of the (n - l)-dimensional simplex,
(i.e. A C {2i 91E I i = 1),
such that Vu,v 9EN, uR*v X min{ju} > min{.Av}.

If the SWO is continuous and satisfies Convexity*, they establish that its relation
with A is one-to-one. Continuity* is necessary, as our discussion of weak welfarism
makes clear. Convexity* is also required, in view of Theorem 4.7, where the same
Ch. 10: Social Welfare Functionals and Interpersonal Comparability 515

set A is relied on in two contrasting representations, one function being concave and
the other convex.
If the SWO satisfies Anonymity*, its representation is based on a correspondence
defined on the set consisting of the rank-ordered subsets of the SWO domain and
linking each of them to one or more weighting schemes, i.e. points of the (n - 1)-
dimensional simplex, located in the same rank-ordered subset of the latter, so that their
union makes up a symmetric A. In case this statement requires clarification, we remind
the reader that each dimension of the simplex is associated with a fixed individual,
just as it holds true for the full SWO domain. Instances in three dimensions may be
illuminating. Under Anonymity*, A is symmetric with respect to the three 450 lines.
For example, the maximin SWO is paired with the full simplex of dimension two; it is
a member of the generalised Gini family, which as a rule calls for an hexagonal figure
within the interior of the simplex. If the SWO reflects weighted utilitarianism, it is
paired with a mere single point in the simplex. All the above comments are intended
to convey some intuition about the following 2 2:
Theorem 4.8: Min-of-means SWO. A SWO R* defined on 9 lN satisfies Inv*(a + bui),
Weak Pareto*, Continuity*, and Convexity* if and only if it is a member of the min-
of-means family.
The implications of this representation theorem 23 for the SWFL context are obtained
at once, provided formal welfarism holds:
Theorem 4.9: Min-of-means SWFL. Suppose F has universal domain; then, the
two following statements are equivalent: (1) F satisfies Binary independence, Weak
Pareto, Continuity, Inv(a + bUi), and Convexity, and (2) there exists a compact and
convex set: A C { E 9I, i2ii = 1} such that VU c U, Vx,y E X,

min{AUx}
;,EA
> min,
AEA
Uy} xR uy.

In view of Roberts's theorem on weak welfarism (our Theorem 3.14), we can drop
Continuity from the above list and exhibit another characterisation result based on
Theorem 4.8, provided Convexity ofF implies Convexity* of its associated SWO under
weak welfarism. To check this, we consider any u, v E 1N
9 such that uR*v, and any
a C 91++, a < 1. Then, we can find U E D)and x,y, z X such that u = U,, v = Uy and

22 In their 1997 account of this theorem (p. 199), the three authors add unnecessarily the following
clause to Weak Pareto*: Vu, v E N, u > v = uR*v; indeed, this is implied by Weak Pareto* and
Continuity* taken together. On the other hand, our Convexity' axiom is stronger than the related axiom
they rely on.
23 Each representation theorem specifies as narrowly as possible the utility representations of each
element of a set of orderings displaying common characteristics. In particular, representation theorems
prove very useful in formally welfarist contexts
516 C. d 'Aspremont and L Gevers

au + ( - a) v = U,, and we observe that UR*v by Convexity of F; indeed, VU E U,

Vx,y C X, xRuy implies UxR*Uy; hence, we can apply Roberts's theorem:
Theorem 4.10: Weak Min-of-means SWFL. Suppose F satisfies Domain
universality, Binary independence, Weak Pareto, Inv(a + bUi), and Convexity,- then,
there exists a compact and convex set A C {1 C R9 + I Zji = } such that VU C D,
Vx,y E X,

min{AUx} > min{,l U} > xPuy.

However interesting these results may be, they uncover an embarrassment of riches
and call for the addition of new axioms.

4.2.2. A modicum of Separability

Our next theorems introduce more or less demanding versions of Separability

and/or more restrictive invariance axioms. One of them, proved by Gevers (1979)
and further strengthened by Ebert (1987), is based on the mixed-invariance axiom
Inv*({a, + bui}&p(ui)), which is consistent with a very limited amount of separabil-
ity.
Theorem 4.11. Suppose an SWO R* satisfies Inv*({ai + bui}&ql(ui)), Strict Pareto*
and Anonymity*; then, there exists a subset M C {1,2,.., n}, such that Vu, v E "l ¥ ,

Z ui(j) > E Ui(j) > P*.

jIM jEM

We shall omit the lengthy proof. By adding gN-Minimal individual symmetry to

our list of axioms, we eliminate some uninteresting SWOs.

G -Minimal individual symmetry*:

For any two individuals i, j E N, 3u, v E GN such that ui > vi, uj < vj,
ul = vl, Vl C {i,j}, and uI*v.

This new property is not weaker than the original statement, even though it is
obtained from it by substituting GN for 9 N. For instance, the leximin SWO satisfies
Minimal individual symmetry*, whereas it contradicts the restricted version.
Corollary 4.12. If an SWO R* satisfies Inv*({ai + bui} &q(ui)), Strict Pareto*,
Anonymity* and 9N-Minimal individual symmetry", it is weakly pure utilitarian.
Alternatively, we can add Continuity*, which turns out to have stronger marginal
implications, to wit:
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 517

Corollary 4.13. An SWO R* satisfies Inv*({ai + bui}&(p(ui)), Strict Pareto*,

Anonymity* and Continuity* if and only if it is purely utilitarian.
It is also instructive to go back to the weaker and more natural invariance axiom
Inv* (a +bui), and to study the family of anonymous SWOs satisfying gN-Separability*,
a property defined in Subsection 3.3.3. Indeed, by restricting the unconditional property
to well-ordered evaluation vectors under these circumstances, we obtain a slightly
modified version of an important result due to Ebert (1988b):
Theorem 4.14. If an SWO R* satisfies Inv*(a + bui), Strict Pareto*, Anonymity*,
Continuity* and gN-Separability*, then, there exists (j)j=1 E 9iN such that,
Vu,vU9C N ,
n In

uR*v > jUi(j) > Z JUi(j).

j-1 j-1

Remark 1. As he applies Debreu's (1960) classical theorem on the representation of

orderings involving at least three independent factors, Ebert requires n > 3. This is
unnecessary in view of Theorem 4.7.

Remark 2. This family of SWOs intersects the generalised Gini family described in
example 2.2.5.

4.2.3. Full Separability

We proceed with a result due to Deschamps and Gevers (1977, 1978), which is based
on the full strength one can impart to both Separability and Pareto dominance. It
relies neither on Continuity, nor on Roberts's theorem, as the latter seems little helpful
under Separability; yet the theorem provides a joint derivation of weak weighted
utilitarian rules, together with leximin, and its inequitable mirror image, leximax. This
result (the proof of which is too long even to be sketched here), leads to corollaries
characterizing very antagonistic social welfare orderings: viz. weighted utilitarianism
and pure utilitarianism on one hand, leximin on the other.
Theorem 4.15. For n > 2, an SWO R* satisfying Strict Pareto*, Minimal Individual
Symmetry*, Inv*(a + bui), (resp. Inv*({ai + bui}&p(ui)) and Separability* is either
leximin, leximax or weak weighted utilitarianism (with all weights positive).
Proof: This follows directly from the theorem of Deschamps and Gevers (1977), which
is proved for Inv*(a + bui). We can also observe that the original result goes through if
we rely on the stronger mixed-invariance axiom Inv*({ai + bui}&(ui)) since none of
the conclusions are contradicted. The really new feature is the use of MIS*. Indeed, as
stated without MIS*, the original version of the theorem only requires a non-empty set
of at least three undominated individuals. An individual i is undominated if and only
518 C. d'Asprenont and L. Gevers

if for all j A i, there is some u, v E 9' such that Uh = h for all h, i h • j, uj > j
and vR*u (with SP* one should have v > ui). This clause is required because the
Separability axiom loses its bite and reduces to Strict Pareto* if n = 2, so that we are
back to Theorem 4.2. Clearly, MIS* implies that all individuals are undominated. O
A first characterization that one can deduce from Theorem 4.15 is that of leximin. To
get rid of utilitarianism one can further strengthen the mixed-invariance requirement by
considering cardinal interpersonal comparisons as irrelevant, and rely on Inv*((q(ui)),
but one can also use another axiom lying between the latter and Inv*(a + bui), viz.
Inv*((Au)). To eliminate leximax, any innocuous equity condition can be introduced,
such as

Minimal equity* (ME):

For some i,j e N, there exist u, v C 9 jN such that
Uh = /, for all h, i h •j, v > uj > ui > vi and uR*u.

We then get
Theorem 4.16. For n > 2, an SWO R* satisfying Strict Pareto*, Minimal Individual
Symmetry*, Minimal Equity*, Inv*(q(Au)) (resp. Inv*(q(ui))) and Separability* is
leximin.
Proof: We simply prove the part relying on Inv*(qp(Au)), by going back to the previous
theorem, and by showing that the stronger invariance axiom contradicts utilitarianism.
The proof is based on two steps.
Step one: Suppose first that at least two individuals i, j have unequal utilitarian
weights denoted a and Pi, respectively, and such that a/ > 1. We consider
three evaluation vectors u0,ul,v defined as follows: k E N, i k j,
0 °
k0 ul = 0 = i == = k; U0 = -£ < 0, UO > 0, U = -E1 < 0, UI = r > 0,
where we can always choose the last four numbers so that l/e1 > a/p > l70/e( > 1.
By construction, we have a > °o and a < , so that utilitarianism implies
ulP*vP*u° , which contradicts Inv*(q(Au)).
Step two: Suppose next that all weights are equal. Let n = 3 and consider the three
following evaluation vectors: u° = (1,2,8.5), u = (1,2,8.5 + E), where > 1, and
°
v = (4,4,4); we notice again that utilitarianism implies ul'PP**u , which contradicts
Inv"*((Au)). If there are more individuals, we simply add indifferent people with
constant evaluation count 4. []
Our proof is obtained by adapting an argument used by Bossert (1991) in the context
of a weaker theorem.
Let us turn next to characterisations of utilitarianism. Weak weighted utilitarianism
immediately follows from adding N-Minimal individual symmetry* to the set of
axioms used in Theorem 4.15. Continuity also contradicts both leximin and leximax
[see, e.g., Maskin (1978), Moulin (1988)], so that the following consequence can be
easily established (see the argument in Theorem 4.2).
Ch. 10: Social Welfare Functionals and InterpersonalComparability 519

Corollary 4.17. For n > 2, the SWO R* satisfies Strict Pareto*, Minimal individual
symmetry*, Inv*(a +bui) (resp. Inv*({ai + bui}&T(ui))), Separability* and Continuity*
if and only if it is weighted utilitarianism(with all weights positive).
Another way to eliminate both leximin and leximax in Theorem 4.15 is to strengthen
the invariance axiom Inv*(a + bui) to Inv*(ai + bui), which implies SE*. Only weak
weighted utilitarianism (with positive weights) remains. This result can be obtained
using still other arguments [Maskin (1978), d'Aspremont (1985)].

4.3. Homothetic s. translatable social welfare functionals

The last group of results will be concerned with ratio-scale invariance. As we have seen
for SWFLs, statements that are compatible with such invariance axioms are meaningful
if the domain is restricted to be D*, a domain which is homogeneous in sign. To
illustrate such axioms in the welfarist framework, we shall restrict ourselves to the
domain of positive utilities 91N. This will have the additional advantage of simplifying
the proofs. For the first result we may even use Blackwell and Girshick (1954) again.
Theorem 4.18: Weighted utilitarianism vs. Nash. Suppose an SWO R* satisfies
Continuity* and Strict Pareto*. If Inv*(ai + ui) holds, then it is weighted utilitarianism;
if Inv*(biui) holds and H(X, D) = 9, then we get the Nash bargaining solution
with status-quo point normalised to zero, i.e. there exists ()>=l E C9 such that
Vu, v Ec 9+,

n n

uR*v X I ui V
v,.
i-1 i=l

With Anonymity* in addition, we get respectivelypure utilitarianismand the symmetric

Nash solution.
Proof: From Theorem 4.2 we know that, with Inv*(ai + ui), C* and SP*, we get a
weighted utilitarian SWO R*. Using this fact, another SWO R* can now be defined on
9+N+ by putting, Vu, v E 9l N,

(eM ,. ... ,e")k * (e°,..., e °"

i) X uR*v.

Clearly R* also satisfies C* and SP*, and Inv*(ai + ui) for R* translates into Inv*(biui)
for R* (taking bi = ei). Finally, for any u, v E 9j+N, uR*v, being equivalent to
(In ul,..., In u,) R* (n vl,..., In v,), holds if and only if Ei E NAi In ui > Ei C N i In vi
or, equivalently, In= u'i > i=1 vi. With A*, all the weights should be equal. D
If we allow for interpersonal comparability, we enlarge considerably the admissible
class of social welfare orderings. For example, with C*, SP*, A* and Inv*(bui) (resp.
Inv*(a + u)), the class of admissible SWOs are all those representable by a social
520 C. d 'Aspremnont and L. Gevers

evaluation function defined on 9')+ which is continuous, strictly increasing in each

argument, symmetric and homothetic (resp. translatable). However, the addition of
separability makes it possible to get functional-form characterisations, as is well known
from the work of Blackorby and Donaldson (1980, 1982). An interesting intermediate
case is given by imposing the weaker separability condition gN-Separability* defined
in Subsection 3.3.3. In fact, having a large class of SWOs gives the possibility of
introducing additional equity conditions, more demanding than simple anonymity. In
the final result that we state here to illustrate this possibility, combining results in Ebert
(1988a,b), we use the Pigou-Dalton* principle defined in Section 3.5:
Theorem 4.19. Suppose n > 2, and the SWO R* satisfies Continuity*,
gN-Separability*, Anonymity*, Strict Pareto* and the Pigou-Dalton*principle.
(1) If Inv*(a + u) holds, then, for some ,B G (0, 1) and some ()j= 1 E 91'+with
; > j+I

Vu,v E 9v, uR*v 1n>l13

ln 3je'/S) I> i3n Siije(lnfl)vd,
1 n
or, VuE G 9++, uRUvu Ei5jui(j) > ,E5
jVi(j).
j=l j=l

(2) If Inv*(bui) holds, then, for some E (0, 1) and some (j)= 1 E fN+ with
n > j+l,
n A 1/f) ( 1/f3
Vu, v E9i++, uR*v > j
j1 ; , j

or, Vu, v E 9+l, uR*v HUi() > :V(j.

j=-1 j=

(3) If Inv*(a + bui) holds, then we get (only) the generalised Gini social evaluation
function. for some (jl)>;= E 9+N with )j > Aj+ ,1
n n

VU, E 9i, uR% X Aju(j) jUi(j).

j=- j=1

This is one characterization of the generalised Gini social evaluation function. Oth-
ers follow readily from Theorems 4.6 and 4.14, provided the Pigou-Dalton* principle
is added.

5. Discarding neutrality or invariance

Some interesting SWFL characterisations do not rely on Invariance, and some others do
away with Strong neutrality. We have assembled below some of those we are aware of.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 521

Harsanyi (1955, 1977) and Hammond (1976a) pioneered this approach. Their results,
which are not based on invariance properties, deal respectively with utilitarianism and
the leximin principle. For easy comparison, we present them in the SWFL framework,
even though neither contribution was using it originally. As we shall indicate briefly,
some very recent work has developed in either case the consequences of leaving aside
Strong neutrality, while reintroducing at least implicitly the invariance properties which
are familiar in noncooperative game theory, viz. Inv((qP(Ui)) or Inv(ai + biUi). Our
presentation falls into two natural subsections: in the first, we deal with utilitarianism
as a tool for social evaluation under risk and uncertainty and we comment extensively
on the latter context; in Subsection 5.2 we deal with the leximin principle, in both
abstract and more structured economic environments. In the latter case, the limitations
of formal welfarism are put in evidence.

5.1. Uncertainty and risk: from Harsanyi to Relative Utilitarianism

As we stand on the threshold of this subsection, we mention the invariance-free
approach to utilitarianism which is developed by Blackorby et al. (2002). In contrast
with Harsanyi, it does not rely on risk and uncertainty, but it makes essential use of
the Incremental equity axiom (defined at the end of Section 3.5), the cogency of which
is not self-evident.
Theorem 5.1: Pure utilitarianism. An SWO satisfies both Strict Pareto* and the
Incremental equity* axiom if and only if it is purely utilitarian.
Requiring the SWO to be purely utilitarian in each case where there are only two
nonindifferent individuals is likely to appear objectionable if this is proposed as an
independent axiom. Yet, as d'Aspremont (1985, lemma 3.3.1) shows, it may replace
both Incremental equity* and Strict Pareto* to characterise pure utilitarianism. Even
though this result cannot be more convincing than the axiom it is based on, it may prove
useful as an intermediate lemma if it can be derived from more palatable axioms.
As we wish to relate some of Harsanyi's classical contributions and SWFL theory, we
have to introduce uncertainty and risk. Indeed, explicit consideration of these features
raises specific difficulties when evaluating social decisions. We shall be inexorably
brief on technical matters and concentrate on interpretation; we shall emphasize, in
particular, an example originally proposed by Diamond (1967) and further developed
by Ben-Porath et al. (1997).
In order to add the dimension of state contingency to our aggregation problem, we
shall define a set S of states of the world, the realization of any of which is regarded as
a possibility, whether the evaluation of social decisions is to be made by an individual
or by the social observer or by both. The states are assumed mutually exclusive and
such that only one of them will eventually obtain.
The evaluation takes place or the social decision is selected before the true state is
observed, and it can take into account the relative likelihood of the states. Furthermore,
people's beliefs about the latter may legitimately diverge in the context of uncertainty;
522 C. d'Aspremont and L. Gevers

in this case, it is assumed that an n-tuple of not necessarily distinct probability

distributions (p )i, can be elicited from the individuals, while the ethical observer's
beliefs may be captured by yet another probability distribution (ps)s Cs. In the context
of risk, we assume that everyone entertains the same beliefs, which can be captured
by the same probability distribution 24 .
Under uncertainty or risk, the consequence of any social decision may vary
according to the state. As we define it, a consequence involves a full description of all
aspects which are expected to matter in the eyes of any individual or in those of the
ethical observer, except for the evaluation process itself. In the sequel, we shall denote
by C the set of all relevant consequences, whereas c will designate the particular
consequence contingent on the realisation of s if a particular social decision x X is
chosen.
In this setup, the ethical observer, required to establish a social ranking over X, is
supposed to be informed of as many individual evaluation profiles as there are states
in S. The generic element of the s-conditional profile is given by V(c', i), where
x X and i N. Having s as a direct argument would prove redundant if we
define consequences in a comprehensive manner as we do. The observer's goal is to
aggregate with respect to both individuals and states. Before we embark on this double
exercise, let us point out that we shall single out expected utility maximization as the
main individual decision criterion under uncertainty and risk2 5. Even though it may
lack empirical support, it is considered by many authors, including Harsanyi, as the
hallmark of rationality. Its mathematical closeness to utilitarianism is interpreted by
Mirrlees (1982) in a natural way: under uncertainty and risk, the individual decision-
maker imagines by anticipation the experience of as many possible selves as there are
states. It is therefore tempting to combine both aggregation principles and to assume
that inequality aversion and risk aversion are two faces of the same attitude, as Harsanyi
suggests.
Before we embark on the technical presentation, we would like to convey some
intuition about the problem at stake and discuss Diamond's (1967) example, as
developed by Ben-Porath et al. (1997). It combines a simple pure distribution problem
with the possibility to toss a fair coin, the sides of which are respectively denoted
h and t. To simplify, we omit any reference to C. Society consists of two perfect
twins I and m, with the same risk aversion and the same level of initial income which
we use as origin for individual utilities. The ethical observer is to rank three social
decisions y, z and w. Decision y consists of allocating an extra unit of utility to 1,
without transferring anything to m, independently of the toss outcome. Decision z
involves such a dependence since it consists of a fair lottery: 1 gets the extra unit

24 For more on this issue, see Mongin and d'Aspremont (1998).

25 It has invaluable advantages when decisions must be taken sequentially while information evolves
between decisions, a situation best described by a decision tree. See for instance Hammond (1988, 1998)
and Sarin and Wakker (1994).
Ch. 10: Social Welfare Functionals and Interpersonal Comparability 523

of utility and m receives nothing if h shows up, whereas the reverse takes place if
t shows up. Decision w is such that one unit of utility is allocated to both I and m
if t obtains, whereas both receive nothing otherwise. This description is summarised
in the following table, where xis means "decision x, given state s", and the numbers
in the various cells indicate the relevant values of V(cx, i):

I m 1 m

ylh 1 0 ylt 1 0
zlh 1 0 zlt 0 1
wlh 1 1 wit 0 0

We shall define yet another reference decision when we need it. How are we to rank
socially y, z and w? It may depend on the order of aggregation. We distinguish two
methods:
1. In the ex post approach, aggregation over individuals is done first, conditionally
on each state in turn. This is a problem SWFL theory is meant to help solving. If
the ISI conditional social rankings can be represented by as many social evaluation
functions, the intermediate output is a SI-tuple of evaluation indicators of each
decision through its consequences associated with the states. When both IXI and ISI are
finite the typical element of the corresponding XI x ISI matrix may be denoted W(x, s).
For instance, in case the maximin principle is applied at the first round of aggregation,
we let W(x,s) = mini EN{V(c, i)}, whereas we define W(x,s) = Zi CN V(c, i) if
the aggregation principle is purely utilitarian. The ethical observer's own probability
distribution is relevant in the next step of this approach, when the social ranking of
decisions is obtained by aggregating over the set of states.
In our example, decision z, given h, is the same as ylh, with respect to both income
and individual utilities, so that we declare them socially indifferent, given h. Assuming
that the relevant social orderings are anonymous, we observe that zt must be also
socially indifferent with ylt. As we proceed to the next aggregation stage, we may
conclude that y and z are socially indifferent for lack of a reason pointing one way or
another.
When comparing zlh with wlh, the latter dominates, whereas the opposite is true if
t obtains. Suppose we sum utility numbers at the first level of aggregation to evaluate
social decisions conditionally on states and we rely on expected social utilities at the
second stage; then z and w must be socially indifferent. This seems to conform to
Hammond's (1982) version of utilitarianism which relies on the ex post approach;
individual utilities are initially summed, and their degree of concavity is chosen to
reflect the ethical observer's aversion for income inequality, independently of the
individuals' attitudes towards risk. In this particular instance, however, many people
prefer w to z because equality is achieved in the former case both ex ante and ex post;
this ranking is implied if the slightest degree of social aversion towards inequality in
524 C. d'Aspremont and L Gevers

utilities is applied at the first stage of aggregation, as is consistent with the generalised
Gini SWFLs among others, even though expected social welfare is maximised at the
ensuing stage.
2. In the ex ante approach, aggregation over states is done first for each i N in
turn. It relies on the individual's own method and beliefs, as the latter are captured
by (p')s c s, at least when the individual's learning process is deemed reasonable. If
the latter condition fails, it is natural to rely on the ethical observer's own method
and/or beliefs. The intermediate output is a list of n individual evaluation functions
defined over the set of social decisions. The typical element of the corresponding
IXl x n matrix may be denoted U(x, i). For instance, in case an individual's
risk aversion is extreme, as perhaps Rawls would think appropriate in the original
position, the maximin principle is applied at the first level of aggregation, and
we let U(x,i) = minSEs{V(c ,i)}, whereas we define U(x,i) = ,SESp'V(c',i)
if aggregation over states relies on the expected utility principle, as Harsanyi
recommends. This approach is relevant in contexts where the SWFL is utility-based
and the ethical observer disregards what is to take place after the state of the world is
disclosed, in particular, as Kolm (1998) explains, when individuals are held responsible
for the risk they incur. The intermediate output is used as an input for the ensuing
interpersonal aggregation stage, at which SWFL theory can be invoked. Let us turn
next to our example and rearrange our data in more appropriate fashion; here, xi
means that the corresponding row displays the relevant values of V(c", i) for each
state in turn.

h t h t

yll 1 1 ylm 0 0
zl 1 0 zlm 0 1
wl 1 0 wim 1 0

At the first round of aggregation, we must have U(z, ) = U(w, ) as well as

U(z,m) = U(w,m), because the coin is fair. As both individuals are indifferent
between w and z, society may also be declared indifferent between w and z at the
next aggregation stage. On the other hand, it is natural to assume that U(y,l) = 1
and U(y, m) = 0. If we denote by the common value of U(z, I) = U(z, m), we must
have 0 < t < 1. For instance, if individuals are expected utility maximizers, = ,
whereas = 0 if they endorse the maximin principle. To make more transparent the
choice between y and z as we move to interpersonal aggregation, we may assume that
there exists some v E X of which both I and m derive a sure utility level E and which
is socially indifferent to y, so that comparing socially y with z amounts to Pareto-
comparing Ewith q, i.e. social acceptance of inequality in sure utilities with private
acceptance of utility risk. Harsanyi argues that the same expected utility ought to be
applied both in ordinary life and in the original position, so that E = r/= This makes
Ch. 10: Social Welfare Functionals and InterpersonalComparability 525

for simpler decision criteria since there is no need to distinguish the ex ante approach
from the ex post method. Yet, as Diamond pointed out, many people's ethical intuitions
indicate a strict preference for z over y because their aversion to utility inequality
among individuals is greater than the individuals' aversion to utility risk. The same
conclusion is reached by an interesting alternative theory developed by Epstein and
Segal (1992), which is also based on the ex ante approach; it assumes expected utility
at the private level and implies quadratic aggregation of individual expected utilities,
so that < = . The generalised Gini SWFLs could be used to the same effect
instead of the quadratic, as Ben-Porath et al. (1997) suggest.
Unless one is persuaded by Harsanyi's arguments, the ex ante approach may thus
clash with the ex post method. One way of solving the clash would be to add a time
dimension, allowing for the fact that each individual can only entertain expectations
at the first period, whereas he or she experiences the true state of the world at the
following period. This would generally warrant yet another round of aggregation,
even though social evaluation takes place before the true state of the world is known.
As Ben-Porath et al. indicate, an averaging process would seem reasonable and it
would have the distinct advantage to be consistent with expected utility or welfare
maximisation at both the individual level (in the ex ante approach) and the social level
(in the ex post method), while it concludes that w is ranked socially above z, and the
latter is ranked socially above y, an ordering which fits many people's ethical intuition
and which no single aggregation method seems capable of delivering in isolation.
We introduce next the formal apparatus we rely upon to link Harsanyi's classical
contributions to SWFL theory. For this purpose 26 , we shall restrict ourselves to risky
situations and treat any decision x e X as a lottery with prizes in C. Thus, we want to
associate any c C C with its probability denoted pX, given x. Let us collect all c x E C
such that c = c as well as the subset of states involved; if this set is empty, we define
px = 0; if it is not empty, we define pX as the sum of all probabilities ps associated with
the relevant states. In conclusion, we shall simply assume that the set of alternatives X
is a convex subset of some linear space.
The domain restriction we just described is not the only one we rely on. Next, we
restrict each R, be it individual or social, to be a von Neumann and Morgenstern
(VNM) preference ordering, i.e. to satisfy two additional properties:

VNM-continuity of R.
Vx,x',x" C X, the sets
{, C [0, 1] I x" R [x + (1 - A)x']} and { C [0, 1] [ilx +(1 - )x'] Rx"}
are closed in [0, 1].

26 The following developments are based on d'Aspremont and Mongin (1997). See also Mongin (1994).
526 C d'Asprenont and L. Gevers

VNM-independence of R.
Vx,x',x" E X, VA ]O, 1],
xRx' [x + (1 - )x"]R [x' + (1 - )x"].

A VNM preference ordering R can always be represented by a function u defined

on X. Moreover, in this framework, every utility representation of R is either mixture-
preserving, i.e. Vx,y X, V, t [0, 1], u(Ax + (1 - )y) = Au(x) + (1 - ,) u(y), or a
monotone transformation of a mixture-preserving utility function. In comparison with
the abstract framework, in which uncertainty and risk remain implicit, we are here
narrowing down both the SWFL's domain and its image set. It is the price to be paid
for getting rid of any invariance requirement.
To insure that the full domain of the corresponding SWO (viz. 9 N) is attainable,
we shall introduce yet another domain assumption under the name of "independent
prospects axiom", as it is called by Weymark (1993) and implicitly used by Harsanyi
(1955), and which applies to a single profile of VNM preferences (R 1, ... ,Ri,..., R,,)
defined on X, which, as we have seen, is itself assumed to be a convex subset of some
linear space.

Independent prospects:
For every i N, there are x,y E X such that xPiy and xjy for allj N \ {i}.

A domain D, consisting of all VNM-utility representations of a single profile of

VNM-preference orderings satisfying independent prospects, will be called a Harsanyi
domain. This is enough to ensure Domain attainability 27.
We can now state Harsanyi's aggregation theorem (1955, 1977), which eschews
Strong neutrality, but does not dispense with Pareto indifference. We have the
following:
Theorem 5.2. Suppose that F is defined on a Harsanyi domain D and satisfies Pareto
indifference, and that, for U D, R is a VNM preference ordering. Then, there is
N
some Au E , AU O, such that

II T1

'Vx,y EX, xRuy 7 ,kU(x,i) > A UU(y,i).

-I i I

27 Independent prospects for a given profile of VNM rankings is equivalent to having affine independence
of any profile of VNM utility representations of the given rankings. Hence the range of any such utility
profile has full dimension. See Fishburn (1984) and Weymark (1993).
Ch. 10: Social Welfare Functionals and Interpersonal Comparability 527

Proof: By Theorem 3.3, we may define an ordering R* on U(X), a convex set with
nonempty interior. Now, since Ru is VNM on X, RT* is VNM on U(X). Indeed, for
VNM-independence of R*, we need to show that: Vu, v, w £ U(X), a c]O, 1],

uR v >[au + (1 - a) w] R* [av + (1 - a) w].

By definition of U(X), there are x,y and z in X such that U(x) = u, U(y) = v,
U(z) = w and, by VNM-independence of Ru,

xRuy X [ax +(1 - a) z] Ru [ay +(l- a)z].

so that

UxRuUy Uax+(l-a)zR*Uay+(la)z.

Since D is a Harsanyi domain, each Ui is mixture-preserving, hence

Uax+(l -a) = aUx + (1-a)U.,

Uay+(l-a)z = alUy + (1- a) U,
and R* is VNM-independent. To derive the VNM-continuity of R*, a similar argument
can be used. R*u being a VNM ranking of the convex set U(X) C 9l N, it has a VNM
utility representation W. This mixture-preserving function is affine28 on U(X), i.e. for
all u C U(X), W(u) = i CN iUi + y, for some vector (/3, ... 3) and some scalar 7.
The result follows.
Because in this theorem the weights depend on the chosen profile U, it should not be
taken as a characterisation of weighted utilitarianism [as remarked by Sen (1986a)],
but simply as a representation theorem [see also Blackorby et al. (1990)]. Moreover,
the weights could be negative or nil. However, strengthening Pareto indifference twice,
into strong neutrality and into strong Pareto, is enough to get weighted utilitarianism
with all weights positive, since, then, there is a well-defined SWO R* on 9lN which
is a VNM ranking and which coincides with any R* on its domain of definition.
Corollary 5.3. Weighted utilitarianism. Suppose that the SWFL F is defined on a
Harsanyidomain D, that it satisfies Strong neutrality and Strong Pareto, and that, for
every U c D, Ru is a VNM preference ordering. Then, F is Weighted utilitarianism,
with all weights positive.
Of course, in order to characterise pure utilitarianism instead of weighted utilitari-
anism, it is enough to supplement our set of axioms with an anonymity requirement.
This enables us to prove the following version of Harsanyi's aggregation theorem.

28 For the equivalence of mixture-preserving and affine functions on convex sets, see, e.g., Coulhon and
Mongin (1989).
528 C. d'Aspremont and L. Geuers

Theorem 5.4: Pure utilitarianism. Suppose that the SWFL F is defined on a Harsanyi
domain D, that it satisfies Strong neutrality, Strong Pareto and Anonymity and that,
for every U E D, Ru is a VNM preference ordering. Then, F is Pure utilitarianism.
Proof: From D being a Harsanyi domain and Strong neutrality, formal welfarism
follows. Strong Pareto and Anonymity of F implies respectively SP* and A* for the
SWO R*. Also, using the same argument as above, R* satisfies VNM independence.
This is enough to get Inv*(ai + bui), and hence Pure utilitarianism. Indeed, take any
vector a = (al,..., a,) and b > O0.Ifb < 1, we can simply put w = a/(l - b) and , = b,
then apply VNM independence. If b > 1, clearly uR*u X 2(2bu)R* 2'b(2bv), which
by VNM independence is equivalent to (2bu)R* (2bv) [letting w - 0 and A,= 1/(2b)].
To get the conclusion, let = and w = 2a, and apply VNM-independence again. D
To prevent any misinterpretation of the last two characterisation statements when
comparing them to the results presented in the previous section, we stress again that:
(1) a specific form of continuity of the social ranking is implied by the definition
of a VNM ordering; (2) no invariance axiom is relied on in the theorem statement,
although the proof is based on the equivalence of VNM independence of R * with
Inv*(ai + bU,); and (3) the definition of the Harsanyi domain rests in turn on a rich
domain of individual VNM utilities representating a single profile of individual VNM
preference relations. Yet, this has only technical significance, because we require the
SWFL to be strongly neutral. If we maintain this requirement, we do not alter the
substance of the results even though we adopt a domain of profiles involving many
alternative VNM preference relations for every individual.
We proceed with a couple of remarks meant to clarify the link between the
strongly neutral version of pure utilitarianism presented in Theorem 5.4 and relative
utilitarianism, which is to come next. Choosing interpersonally comparable preference
representations in order to evaluate adequately the alternatives in a specific social
conflict is a momentous and thorny task, but the formal Harsanyan model we have been
describing provides the ethical observer with no clue towards solving this problem, a
feature shared by the bulk of SWFL theory. Indeed, suppose two individuals have
the same VNM preference ordering and unequal scales are used to represent their
common ordering in the profile under the ethical observer's consideration; then the
associated social ranking may be strongly influenced by the specific scales chosen
a priori. In brief, the formally welfarist SWFL characterised in Theorem 5.4 does not
satisfy Inv(ai + biUi). Relative utilitarianism, introduced in Section 2.2 , is immune
from this criticism because it involves a Kaplan normalisation of VNM utilities in case
every individual VNM preference ordering has both a maximal and a minimal element.
If an anonymous aggregation method involves a process of Kaplan normalisation as
an essential intermediate stage, it must treat symmetrically any two individuals having
the same VNM preference relation. This is in contrast with Theorem 5.4.
Dhillon (1998) and Dhillon and Mertens (1999) recently extended in a natural
way the Arrovian definition of a social welfare function by incorporating as basic
datum a universal domain of individual VNM preferences and by maintaining the
Ch. 10: Social Welfare Functionals and InterpersonalComparability 529

Harsanyi requirement that the social ranking be VNM. In this framework, they propose
a set of interesting axioms and obtain two alternative characterisations of relative
utilitarianism. It is instructive, if not economical, to recast their construct in the
SWFL framework. Due to the normalisation process involved in the SWFL, the
final result is not influenced by the choice made a priori among the many distinct
VNM-utility representations of any individual VNM preference profile. In other words,
the SWFL must satisfy Inv(ai + bi Ui). Moreover, the SWFL version of Independence of
irrelevant alternatives is violated. Dhillon (1998) adopts instead a weaker independence
requirement, which proves too weak for precipitating formal welfarism even though
Pareto indifference does hold. Two new pieces of notation prove useful to introduce it.
As before, X consists of a full set of lotteries based on riskless social decisions. The
latter make up the finite subset A of degenerate lotteries and they could be interpreted
as prizes. Moreover, if A' C A, we shall denote E(A') the full set of lotteries based
on A'. Dhillon's axiom is

Independence of redundant alternatives.

VU, V E D, VA' C A, Vx,y E (A'),
xRuy X xRvy if
(1) Va E A', Ua = Va and (2) Vz G X, 3z' cE (A') such that U = Vz,.

Dhillon (1998) also relies on an interesting axiom called Extended Pareto applied
to a variable population context; the version we offer here to facilitate comparisons
is adapted for constant population and it is related to Separability, because it treats
totally indifferent individuals as though they were absent:

Extended Pareto:
Let U, V, W E D, x,y E X, and L,M C N be such that:
(1) N=LUM, LnM=0,
(2) Vi E L, UJ = Vi and Wi is trivial, and
(3) Vi E M, U = Wi and Vi is trivial;
then xRuy if both xRvy and xRwy, and xPuy if both xPvy and xPwy.

Our translation of Theorem 2 in Dhillon (1998) comes next:

Theorem 5.5: Relative utilitarianism. For a fixed andfinite set A such that AI > 4
andfor all N such that n > 3 andfor allprofiles having at least 3 linearly independent
individual utility functions, an SWFL satisfies relative utilitarianismif and only if it
satisfies the following set of axioms: Strong Pareto, Extended Pareto, Inv(ai + bi Ui),
Anonymity and Independence of redundant alternatives.
Another interesting characterisation of relative utilitarianism is provided by Dhillon
and Mertens (1999). They show in particular that a pair of new axioms can replace
530 C. d 'Asprenmont and L. Gevers

Extended Pareto in the last statement of Theorem 5.5. The first one is called
Consistency:

Consistency:
VU, V C D, x,y C X, xRuy xRrvy if there exists i E N such that
(1) V(x, i) = V(y, i),
(2) j E N\{i}, Uj = Vj, and
(3) U is trivial.

Due to the binariness involved in the last clause, Consistency is neither implied
by Inv(ai + biUi), nor by Separability. The remaining axiom has a distinctly technical
flavour; it is a complementary continuity requirement:

DM-Continuity ofF.:
Consider any sequence (U')= C D such that
Vi N\{n}, a C A\{ao},
Ul(a, i)= U*(a, i) for = 1,2,..., whereas Ul(ao, n) converges to U*(ao, n).
If the corresponding social ranking converges to limR,,,
then either Ru* is trivial or R* = limRu,.

Relative utilitarianism is worth comparing to the methods of voting based on such

scoring functions as the Borda standardisation described in example 2.2.7: the latter
also rest on a purely utilitarian formula applied to intermediate individual scores.
Young (1975) provides a remarkable characterisation (which is also based on another
version of Extended Pareto). These methods do not require a set of alternatives
endowed with a specific structure such as convexity. They satisfy the Anonymity
requirement as does relative utilitarianism. Computing individual scores is an ordinal
procedure, so that Invsi(Ui) is satisfied, while Arrovian Independence is violated and
formal welfarism does not hold. The SWFLs based on the summation of individual
scores may have merits as voting rules, but they seem even less fit for social evaluation
than relative utilitarianism, which does not require that intervals between different
levels of preferences be equalised across individuals, but requires only equalised
maxima and minima. This has some relation to, but is less demanding than, requiring a
"fundamental preference" as in Harsanyi's (1953, 1977) Impartial Observer theorem on
utilitarianism. Yet, as our discussion of Diamond's (1967) example suggests, the social
ranking implied by relative utilitarianism is not always in agreement with common
ethical intuitions.
To conclude this subsection, we mention again the elegant contribution by Epstein
and Segal (1992). After dissociating the ex post aggregation method from the
ex ante approach, they rely on the latter to separate the social aversion to inequality
from individual risk aversion, and they obtain a characterisation of a family of
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 531

SWOs involving quadratic aggregation of individual expected utilities. To address the

Diamond (1967) problem, they introduce an axiom of preference for randomization as
follows: consider any pair x,y of socially indifferent lotteries (or social decisions), such
that two individuals have conflicting strict preferences about them; then, both lotteries
are socially dominated by the two-stage lottery based on the toss of a fair coin to
decide which of x and y will be chosen eventually. This axiom does not allow for flat
line segments within social-indifference loci; it is thus inconsistent with utilitarianism.

5.2. On some egalitariansocial rankings

Instead of the three characterisations of the leximin principle presented in Section 4,

all of which invoke invariance properties, we turn next to alternative derivations. They
are obtained by dropping invariance and separability completely, and by substituting
a much stronger equity requirement. Hammond's equity axiom will be used first:
Theorem 5.6. Any SWO R* is the unique formally welfarist associate of the leximin
principle if and only if it satisfies Strict Pareto*, Anonymity*, and Hammond equity*
principle.
If we strengthen Hammond's equity* principle, we obtain the leximin principle
applied to cases where only two persons are not indifferent. This property is contagious,
and it can be extended to social conflicts involving gradually more and more
individuals: as a matter of fact, this extension is carried out when the standard strategy
of proof is followed. For details, we refer the reader to Bossert and Weymark (2000),
who also offer a nice account of the history of this result. Hammond's (1976a) role
seems essential, but he recognizes his debt to Strasnick (1976). Moreover, he did not
use an SWFL framework, and the generalised Arrovian social welfare function he
defined for his purpose hid the fact that no invariance axiom was in effect necessary.
Tungodden (1999) showed recently that any member of a set of alternative properties
can be substituted for Hammond's equity axiom in the last theorem. They have in
common the notion of a deprived group of individuals associated with every pair of
evaluation vectors. Moreover, the social ranking of this pair cannot contradict bluntly
the strict preference relation of the least favored non-indifferent individual provided
he/she belongs to the deprived group. Among the many deprivation criteria mentioned
in the specialised literature, we shall consider only a one-parameter family, which
reminds one of a popular poverty definition. It will be convenient to work with gN,
the set of well-ordered evaluation vectors, within which the identity of the k-worst
off individuals does not change. Given some number a E (0, 1], we shall say that
i E N belongs to the a-deprived group pertaining to {u, v} C N if and only if
ui < a Ej (u/n) or vi < a Ej C N(v/n).
Theorem 5.7. Suppose R* is an SWO defined on N . Then, R* is the leximin SWO
if it satisfies Strict Pareto*, Anonymity* and 3a E (0, 1] for which, Vu, v E N,
532 C d'Aspremont and L. Gevers

vR*u whenever vk > k Jbr some member k of the a-deprived group pertaining to
{u, v}, whereas vi = ui for any individual i, i < k.
The new equity axiom introduced as the last theorem condition seems to restrict to
the relevant deprived group the veto power granted to the least-favored non-indifferent
individual. Yet, as Tungodden shows, this property is contagious and can be extended to
all individuals. He also proves that more general families of definition of the deprived
group could be used in Theorem 5.6 without altering the substance of the result.
On the other hand, Tungodden remarks that the theorem does not go through if a
given percentile (e.g., the median) is substituted for the mean in the definition of the
a-deprived group.
In a highly original paper, BarberA and Jackson (1988) provide among other
things an alternative characterisation of the leximin principle. They rely on an
SWO framework, which is extended to accommodate a variable set of individuals. Our
generic notation N thus designates henceforth any subset of the set of positive integers,
the latter being interpreted as the set of potential agents, and some axioms are designed
as robustness properties of the social ranking RN when N undergoes certain alterations.
The goal of the exercise is to characterise social rankings which apply only to societies
having a constant population; we are not pursuing here an optimal population theory.
Barbera and Jackson's leximin characterisation eschews all invariance axioms. As to
equity properties, they are not so demanding as Hammond or Tungodden, since they
rely only on Anonymity* and a version of Convexity*. What is more debatable in our
context is their axiom of Independence of duplicated individuals, to wit: For any N,
for any i,j N, for any u, v C 9 N, such that u, = uj and ui = vj, uPv if and only if
u-iP\i}v-fi, where the subscripts mean that individual i is no longer a member of the
economy. As the authors write, "... the axiom constitutes a strong value judgement:
that it is not the number of individuals in a welfare group, but the level of welfare
within the groups, that should count in comparing social states". The axiom would
of course be highly appropriate for the theory of individual decision under complete
ignorance.
To conclude this section, we sample some recent work by Fleurbaey and Maniquet
(2000b) who adopt a structure much more specific than those we have dealt with so
far. Indeed, they study the set of Edgeworth boxes with a constant number I of private
goods. An economy e consists of three elements: (1) A set N of individuals defined as
in the previous paragraph; (2) a profile Uv consisting of an NI-tuple of continuous,
strictly increasing and quasi-concave self-oriented individual utility functions which
are defined over the conventional individual
individual consumption set , and (3) an l-tuple
co E 9_+designating the quantity of each good available as social endowment. By
letting these three elements vary to the largest extent consistent with their definition, we
obtain a rich domain S of economies, to which we devote our normative investigation.
We are interested in the social evaluation of a set of alternatives that we define as
follows: XN = 1[N i.e. the IN-fold cartesian product of 9I+. We want to study social
mappings which associate with every e E a social preference ordering Re over the
Ch. 10: Social Welfare Functionals and InterpersonalComparability 533

corresponding XN. To complete our notation, we let xN designate a typical element of

XN, whereas Xi,N stands for a generic component of XN.
As the number NI of individuals populating an economy e e £ may be any positive
integer, we must stretch the SWFL definition to accommodate the social mappings we
just defined. Moreover, even if we restrict our attention to some specific N, we do not
face an ordinary SWFL. Indeed, even though we want to obtain a social ranking of the
same set of alternatives, viz. XN, the social endowment co is allowed to influence Re.
Eventually, we obtain an SWFL by restricting attention to the subset of elements of £
generated by a given pair (N, o).
The family of social rankings we want to study combines a utility-based version of
Rawls's difference principle with the concept of egalitarian equivalence introduced by
Pazner and Schmeidler (1978). Therefore, we call it for short the RPS (i.e. Rawls-
Pazner-Schmeidler)family. It is convenient to proceed in two steps to define it:
(1) We associate with every individual component U- of the profile UN of any e E £ a
canonical utility representation of the preference relation underlying Ui. We denote
it Wi,e:= T - 9 and we define it implicitly by letting its image wi = Wi,e(Xi)
be i's equivalent share in the social endowment vector co, i.e. Ui(xi) = Ui(wios).
(2) For any e C E, any XN,x N E XN, XNPeX~N if mini CN{Wi} > mini N{wi}.
Our next task is to describe a set of axioms sufficient to precipitate a social ordering
belonging to the RPS family. We start with two cross-economy robustness properties,
and we introduce first the Replication invariance axiom, the technicalities of which
will not retain us: it says that for any e £, any XN,X N XN, XNReX N implies the
same social ranking of any v-fold replication of either allocation, assuming that e itself
is replicated v times. An analogous axiom was used by Debreu and Scarf (1963) in
their study of convergence of the core to a competitive equilibrium.
The e-Separability axiom is slightly more delicate: it provides a material condition
for deleting a subset M C N from the a priori given economy e E £, assuming that both
owand the restricted profile UN\M remain unaltered. We denote the reduced economy
g e £. The axiom runs as follows: for any e E E, any XN,YN E XN, XNReYN implies
XN\MRgYN\M if Vi E M, Xi,N = Yi,N. It seems worth pointing out that, despite the
binary nature of its implication, the triggering condition is much narrower than its
welfarist counterpart, because it is based on a maintained consumption level among
the members of M. An analogous remark is also valid for our next axiom, which is
meant to express an equity norm. It provides conditions under which utility transfers
do not hurt society, whenever two individuals having the same preferences are initially
not on the same indifference curve.

ConditionalHammond equity:
For any e C £, any XN,YN e XN,
XNReyN if there exist j, k E N such that
Uk = Uj and Uj(yj,N) > U(x,N) > Uj(xk,N) > Uj(yk,N), whereas
Vi E N, j i k, Xi,N = Yi,N
534 C. d'Asprenmont and L. Gevers

Another equity norm comes next; it is concerned with a highly specific allocation,
so that an exact welfarist translation does not exist.

Equal split.
For any e e £, any X, YN e XN,
XNPeyN if Vi E N, xiN = (o)/]N[ and
there exists somej · N for which Uj(Xi,N) > Uj(yj,N).

Our last theorem is a weak version of Theorem 3 by Fleurbaey and Maniquet

(2000b). Our weakening is meant to facilitate the comparison with the bulk of
SWFL theory:
Theorem 5.8: RPS family. A given member of the RPS family of social orderings
is associated with every e C whenever the social mapping is required to satisfy
the following set of axioms: Weak Pareto, Pareto Indifference, Inv(pi(Ui)), Replication
invariance, e-Separability, Equal split and ConditionalHammond equity.
We remark that no version of interpersonal welfare comparability is used: two
agents having the same self-oriented preferences are treated alike because there is
no reason for doing otherwise within the limits of our model. In principle, the
approach we just exemplified does not preclude the social ordering from taking also
into account idiosyncrasies for which individuals cannot be held responsible, such as
metabolic peculiarities; see Fleurbaey (1995) and Vol. 2 of this Handbook. Moreover,
the existence of an intermediate stage of canonical preference representation calls for
two further comments: (1) it seems to take care of the expensive taste criticism adduced
against utility-based social evaluation by Rawls among others, and (2) as Dhillon and
Mertens (1999), it is inconsistent with the Arrovian independence property and formal
welfarism does not hold.

6. Conclusion

The SWFL concept can be of help to the ethical observer aiming at an appropriate
social evaluation. It helps organize the inner debate and ask the relevant questions:
what is the set of issues, in what ways should society be concerned with individual
consequences of its decisions, how should they be adequately represented by individual
scores and how should one process this information to obtain an appropriate social
ranking? Until recently, the SWFL literature has been mostly helpful in answering the
latter query. It does indeed provide an enlightening analysis of various SWFLs as it
compares their relative merits by means of stylised axiomatic properties. If the formally
welfarist framework is accepted, a good deal of interpersonal comparability does seem
required for obtaining social rankings which satisfy minimal equity requirements.
Among formally welfarist SWFLs, the leximin principle and pure utilitarianism might
Ch. 10: Social Welfare Functionals and Interpersonal Comparability 535

seem to be the candidates displaying the most attractive set of properties. In its own
way, each one is moderately demanding in terms of interpersonal comparisons. Critics
of the leximin principle object to the absolute priority it gives to favoring the least
advantaged individual, without regard for the number of losers and the average size of
their loss. On the other hand, the examples adduced respectively by Diamond (1967)
and by Ben-Porath et al. (1997) for social evaluation under risk and uncertainty are
rather damaging for utilitarianism. Among the SWFLs which are immune from either
criticism, the other members of the generalised Gini family seem least demanding in
terms of interpersonal comparisons.
Once theoretical analysis has delivered its conclusions and an SWFL is chosen, the
most difficult task is to select an adequate profile; indeed, the rest of the selection
work becomes more or less mechanical and it can usually be fed into a computer, as
it consists of looking for the set of best social decisions associated with any feasible
subset of alternatives. If several evaluation profiles seem reasonable in consideration of
the context at hand, the ethical observer faces the embarrassment of a multiplicity of
candidate social rankings. Upon closer examination, ethical intuitions may be found in
disagreement with some of them, and the latter can be discarded. The remaining social
rankings can be intersected; this procedure is useful if the resulting incomplete relation
has enough bite, as it has proved to be the case with Lorenz dominance in empirical
analysis of the pure distribution problem. See, for instance, Shorrocks (1983). Failing
this, one could submit the social rankings to another round of aggregation. However, as
Roberts (1995) and Suzumura (1996) show, such a procedure is plagued with Arrovian
difficulties.
Choosing on a priori grounds interpersonally comparable individual evaluation
counts is probably the least consensual element in the SWFL approach, and several
authors have chosen to rely only on what seems to be the weakest possible comparabil-
ity assumption. Following their approach, some individual characteristics are singled
out as essentially mattering for equity judgments, so that it is socially desirable that
two individuals displaying the same relevant characteristics be treated similarly. This
equal-treatment principle must prevail even though the rest of the characteristics are
individually differentiated; the latter are considered as ethically negligible or secondary.
As Kolm (1996b) insists, there is an ethical drive towards equality if any reason
that might justify inequality is lacking. If individual preferences are singled out as
an essential element for elaborating equity judgments, a noncomparability invariance
axiom is always adhered to if only implicitly, as in positive economic theory. To recall,
this can be consistent with cardinality as in Nash equilibrium in mixed strategies, an
invariance axiom we have abbreviated as Inv(ai + bi Ui), or it can be only ordinally
measurable as in Nash equilibrium in pure strategies, an SWFL property we have
denoted Inv(i(Ui)). This approach is inconsistent with formal welfarism, because the
filtering of information the latter structure implies leaves out essential aspects of the
problem at hand.
Moreover, once this extreme informational parsimony is abandoned, we realise
that the abstract Arrovian framework and its natural universal domain do not seem
536 C. dAspremont and L. Geuers

conducive to any SWFL that is fit for social evaluation, even though it may be
promising as a constitutional rule, as the Borda method of voting. Nowadays, a
majority of researchers seem to have turned towards less sweeping research goals;
they prefer to deal with the conditions of justice and equity in a variety of specific
contexts, in order to take advantage of the extra structure they provide. Besides Roemer
(1986,1996) and Young (1995), we refer the reader to Moulin (2002), Thomson (2002)
and Fleurbaey and Maniquet (2002). Following this approach, the feasible set is more
narrowly specified, but on the other hand, it is allowed to vary. Together with X and U,
the set of individuals N may also be treated as an independent variable. Very interesting
solution concepts whose purpose is to recommend a set of socially best decisions for
every economy in the domain under scrutiny have been studied in the literature. The
proposed solution can always be interpreted as a social ranking consisting only of two
indifference curves, but this usually leads to violations of the Pareto principle, and it
appears to be too rough for second-best or reform problems, where the structure of
the set of alternatives is likely to be too unwieldy for an axiomatic analysis.
A more promising research strategy may be to try to associate a set of social
rankings to a set of solutions of the allocation problem. This approach was developed
successfully by Young (1987) in a taxation problem where taxable income is treated
as a fixed parameter that may vary across individuals. Young shows that two sets of
properties are equivalent; one of them defines a family of solutions, whereas the other
one pertains to social orderings whose subset of best elements always coincides with
a solution in the family. Young (1987) obtains analogous results for bankruptcy and
profit-sharing models.
As we have seen, Dhillon (1998) and Dhillon and Mertens (1999) dealt recently
with a more general context: the complete domain of lotteries one can define
over a given set of pure abstract alternatives. On the other hand, Fleurbaey and
Maniquet obtained characterisations for two classical economic domains: the set of
pure exchange economies [Fleurbaey and Maniquet (2000b)], and the set of two-
good artisan economies with linear production technologies [Fleurbaey and Maniquet
(2000a)]. In every case, a numerical profile is obtained as an intermediate product,
which is further processed as a utilitarian or a leximin practitioner would do to deliver
eventually an appropriate social ranking. One can thus speak of interim or ex post
interpersonal comparability of the intermediate evaluation profile which is induced
by the proposed procedure. Yet, to our knowledge, it has never been technically
fruitful to separate into two stages the axiomatic derivation of the social ranking,
because the intermediate profile appears as a pure by-product of the analysis. It
seems unlikely that the arguments proving useful in formally welfarist characterisations
may be of any help for normative analysis taking advantage of peculiarly structured
economic environments. Due to its discarding of Arrovian Binary independence, the
latter approach is technically very demanding and its practitioners may soon face at
least temporary feasibility limits. Whether it will eventually succeed in superseding
the abstract SWFL framework is for the future to decide. So far, empirical analysis
seems to lean in its favor, but its support comes from a specific angle, which is more
Ch. 10: Social Welfare Functionals and Interpersonal Comparability 537

positive than normative, and it will not necessarily convince the ethical observer we
have been alluding to since the beginning of this chapter.

References
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York); 1st Edition,
1951.
Arrow, K.J. (1977), "Extended sympathy and the possibility of social choice", American Economic
Review Papers and Proceedings 67:219-225.
Bar-Hillel, M., and M. Yaari (1984), "On dividing justly", Social Choice and Welfare 1:1-24.
Barbera, S., and M.O. Jackson (1988), "Maximin, leximin and the protective criterion: characterisations
and comparisons", Journal of Economic Theory 46:34-44.
Basu, K. (1983), "Cardinal utility, utilitarianism, and a class of invariance axioms in welfare analysis",
Journal of Mathematical Economics 12:193-206.
Ben-Porath, E., I. Gilboa and D. Schmeidler (1997), "On the measurement of inequality under
uncertainty", Journal of Economic Theory 75:194-204.
Bentham, J. (1789), An Introduction to the Principles of Morals and Legislation (Payne, London).
Reprinted in 1970 in: J.M. Burns and H.L.A. Hart, eds. (Athlone Press, London).
Bergson, A. (1954), "On the concept of social welfare", Quarterly Journal of Economics 68:233-252.
Blackorby, C., and D. Donaldson (1980), "A theoretical treatment of indices of absolute inequality",
International Economic Review 21:107-136.
Blackorby, C., and D. Donaldson (1982), "Ratio-scale and translation-scale full interpersonal
comparability without domain restrictions: admissible social evaluation functions", International
Economic Review 23:249-268.
Blackorby, C., D. Donaldson and J.A. Weymark (1984), "Social choice with interpersonal utility
comparisons: a diagrammatic introduction", International Economic Review 25:327-356.
Blackorby, C., D. Donaldson and J.A. Weymark (1990), "A welfarist proof of Arrow's theorem",
Recherches Economiques de Louvain 56:259-286.
Blackorby, C., W Bossert and D. Donaldson (2002), "Utilitarianism and the theory of justice", in:
K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 1
(Elsevier, Amsterdam) Chapter 11, this volume.
Blackwell, D., and M.A. Girshick (1954), Theory of Games and Statistical Decisions (Wiley, New
York).
Blau, J.H. (1971), "Arrow's theorem with weak independence", Economica 38:413-420.
Bossert, W (1990), "An axiomatization of the single-series Ginis", Journal of Economic Theory 50:
82-92.
Bossert, W (1991), "On intra- and interpersonal utility comparisons", Social Choice and Welfare
8:207-219.
Bossert, W, and J.A. Weymark (2000), "Utility in social choice", in: S. Barbera, P Hammond and
C. Seidl, eds., Handbook of Utility Theory, Vol. II, Applications and Extensions (Kluwer, Dordrecht)
forthcoming.
Broome, J. (1993), "A cause of preference is not an object of preference", Social Choice and Welfare
10:57-68.
Coulhon, T., and P. Mongin (1989), "Social choice theory in the case of von Neumann-Morgenstern
utilities", Social Choice and Welfare 6:175-187.
d'Aspremont, C. (1985), "Axioms for social welfare orderings", in: L. Hurwicz, D. Schmeidler and
H. Sonnenschein, eds., Social Goals and Social Organizations: Essays in Memory of Elisha Pazner
(Cambridge University Press, Cambridge) pp. 19-76.
d'Aspremont, C., and L. Gevers (1977), "Equity and the informational basis of collective choice",
Review of Economic Studies 44:199-209.
538 C. d'Aspremont and L. Gevers

d'Aspremont, C., and P. Mongin (1997), "A welfarist version of Harsanyi's aggregation theorem",
CORE Discussion paper; also in: M. Salles and J.A. Weymark, eds., Justice, Political Liberalism
and Utilitarianism, Volume in Honor of John Harsanyi and John Rawls (Cambridge University Press,
Cambridge) forthcoming.
Debreu, G. (1960), "Topological methods in cardinal utility theory", in: K.J. Arrow, S. Karlin and
P. Suppes, eds., Mathematical Methods in the Social Sciences (Stanford University Press, Stanford)
pp. 16-26.
Debreu, G., and H. Scarf (1963), "A limit theorem on the core of an economy", International Economic
Review 4:235-246.
Denicol6, N. (1999), "A characterization of utilitarianism without the transitivity axiom", Social Choice
and Welfare 16:273-278.
Deschamps, R., and L. Gevers (1977), "Separability, risk-bearing and social welfare judgements",
European Economic Review 10:77-94.
Deschamps, R., and L. Gevers (1978), "Leximin and utilitarian rules: a joint characterization", Journal
of Economic Theory 17:143-163.
Dhillon, A. (1998), "Extended Pareto rules and relative utilitarianism", Social Choice and Welfare
15:521-542.
Dhillon, A., and J.-F Mertens (1999), "Relative utilitarianism", Econometrica 67:417-498.
Diamond, P. (1967), "Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility:
a comment", Journal of Political Economy 75:765-766.
Dixit, A.K. (1980), "Interpersonal comparisons and social welfare functions", Mimeo (University of
Warwick).
Dutta, B. (2002), "Inequality, poverty and welfare", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds.,
Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 12, this volume.
Ebert, U. (1987), "A note on social welfare orderings", European Economic Review 31:1145-1147.
Ebert, U. (1988a), "Rawls and Bentham reconciled", Theory and Decision 24:215 223.
Ebert, U. (1988b), "Measurement of inequality: an attempt at unification and generalization", Social
Choice and Welfare 5:147-169.
Epstein, L.G., and U. Segal (1992), "Quadratic social welfare functions", Journal of Political Economy
100:691-711.
Fishburn, PC. (1984), "On Harsanyi's utilitarian cardinal welfare theorem", Theory an Decision 17:
21-28.
Fishburn, P.C., and E Roberts (1989), "Uniqueness in finite measurement", in: F Roberts, ed., Applications
of Combinatorics and Graph Theory to the Biological and Social Sciences (Springer, New York)
pp. 103-137.
Fishburn, P.C., H. Marcus-Roberts and E Roberts (1988), "Unique finite difference measurement", SIAM
Journal of Discrete Mathematics 1:334-354.
Fleurbaey, M. (1995), "Three solutions for the compensation problem", Journal of Economic Theory
65:505-521.
Fleurbaey, M., and F. Maniquet (1996), "Utilitarianism versus fairness in welfare economics", in: M. Salles
and J.A. Weymark, eds., Justice, Political Liberalism and Utilitarianism: Themes from Harsanyi and
Rawls (Cambridge University Press) forthcoming.
Fleurbacy, M., and E Maniquet (2000a), "Fair social orderings with unequal production skills", Mimeo
(Namur University, Belgium).
Fleurbaey, M., and E Maniquet (2000b), "Fair social orderings in division economies", Mimeo (Namur
University, Belgium).
Fleurbaey, M., and F Maniquet (2002), "Compensation and responsibility", in: K.J. Arrow, A.K. Sen
and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam)
forthcoming.
Gaertner, W., P.K. Pattanaik and K. Suzumura (1992), "Individual rights revisited", Economica 59:
161-77.
Ch. 10: Social Welfare Functionalsand Interpersonal Comparability 539

Gevers, L. (1979), "On interpersonal comparability and social welfare orderings", Econometrica 47:
75 89.
Gilboa, I., and D. Schmeidler (1989), "Maxmin expected utility with non-unique prior", Journal of
Mathematical Economics 18:141-153.
Hammond, PJ. (1976a), "Equity, Arrow's conditions, and Rawls' difference principle", Econometrica
44:793-804.
Hammond, P.J. (1976b), "Why ethical measures of inequality need interpersonal comparisons", Theory
and Decision 7:263-274.
Hammond, P.J. (1979), "Equity in two person situations: some consequences", Econometrica
47:1127-1135.
Hammond, P.J. (1982), "Utilitarianism, uncertainty and information", in: A.K. Sen and B. Williams,
eds., Utilitarianism and Beyond (Cambridge University Press, Cambridge) pp. 85-102.
Hammond, P.J. (1988), "Consequentialist foundations for expected utility", Theory and Decision 25:
25-78.
Hammond, PJ. (1991), "Interpersonal comparisons of utility: Why and how they are and should be
made", in: J. Elster and J.E. Roemer, eds., Interpersonal Comparisons of Well-Being (Cambridge
University Press, Cambridge) pp. 200-254.
Hammond, PJ. (1998), "Interpersonally comparable utility", in: S. BarberA, P. Hammond and C. Seidl,
eds., Handbook of Utility Theory, Vol. 2 (Kluwer, Dordrecht) forthcoming.
Hammond, PJ. (1999), "Roberts' weak welfarism theorem: a minor correction", Department of Economics
Working Paper 99-021 (Stanford University).
Harsanyi, J.C. (1953), "Cardinal utility in welfare economics and in the theory of risk-taking", Journal
of Political Economy 61:434-435.
Harsanyi, J.C. (1955), "Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility",
Journal of Political Economy 63:309 321.
Harsanyi, J.C. (1977), Rational Behavior and Bargaining Equilibrium in Games and Social Situations
(Cambridge University Press, Cambridge).
Hicks, J.R. (1959), Essays in World Economics (Clarendon Press).
Kaneko, M. (1984), "On interpersonal utility comparisons", Social Choice and Welfare 1:165-175.
Kelsey, D. (1987), "The role of information in social welfare judgements", Oxford Economic Papers
39:301-317.
Kemp, M.C., and Y-K. Ng (1976), "On the existence of social welfare functions, social orderings and
social decision functions", Economica 43:59-66.
Kolm, S.-Ch. (1966), "The optimal production of social justice", Proceedings of an IEA conference
held in Biarritz; reprinted, 1969, in: H. Guitton and J. Margolis, eds., Public Economics (Macmillan,
London) pp. 145-200.
Kolm, S.-Ch. (1972), Justice et Equite (CNRS, Paris).
Kolm, S.-Ch. (1996a), Modern Theories of Justice (MIT Press, Cambridge, Mass.)
Kolm, S.-Ch. (1996b), "Rational just social choice", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds.,
Social Choice Re-Examined, Vol. II (Macmillan Press, London) pp. 167-196.
Kolm, S.-Ch. (1998), "Chance and justice: social policies and the Harsanyi-Vickrey-Rawls problem",
European Economic Review 42:1393-1416.
Krantz, D., R.D. Luce, P Suppes and A. Tversky (1971), Foundations of Measurement, Vol. I, Additive
and Polynomial Representations (Academic Press, New York).
Krause, U. (1995), "Essentially lexicographic aggregation", Social Choice and Welfare 12:233-244.
Lockwood, B. (1984), "Social choice, interpersonal comparability and welfare economics", in: E van
der Ploeg, ed., Mathematical Methods in Economics (Wiley, New York) pp. 371-400.
Luce, R.D., and H. Raiffa (1957), Games and Decisions: Introduction and Critical Survey (Wiley, New
York).
Maskin, E. (1978), "A theorem on utilitarianism", Review of Economic Studies 45:93-96.
540 C. dAspremnont and L. Gevers

Milnor, J. (1954), "Games against nature", in: R.M. Thrall, C.H. Coombs and R.L. Davis, eds., Decision
Processes (Wiley, New York) pp. 49-59.
Mirrlees, J.A. (1982), "The economic uses of utilitarianism", in: A.K. Sen and B. Williams, eds.,
Utilitarianism and Beyond (Cambridge University Press, Cambridge) pp. 63-84.
Mongin, P. (1994), "Harsanyi's aggregation theorem: multi-profile version and unsettled questions",
Social Choice and Welfare 11:331-354.
Mongin, P., and C. d'Aspremont (1998), "Utility Theory and Ethics", in: S. BarberA, P Hammond and
C. Seidl, eds., Handbook of Utility Theory, Vol. 1, Principles (Kluwer, Dordrecht) pp. 371-481.
Moulin, H. (1983), The Strategy of Social Choice, Advanced Textbooks in Economics (North-Holland,
Amsterdam).
Moulin, H. (1988), Axioms of Cooperative Decision Making (Cambridge University Press, Cambridge).
Moulin, H. (2002), "Axiomatic cost and surplus-sharing", in: K.J. Arrow, A.K. Sen and K. Suzumura K.,
eds., Handbook of Social Choice and Welfare, Vol. 1 (Elsevier, Amsterdam) Chapter 6, this volume.
Nash, J.F (1950), "The bargaining problem", Econometrica 18:155-162.
Parks, R.P. (1976), "An impossibility theorem for fixed preferences: a dictatorial Bergson-Samuelson
social welfare function", Review of Economic Studies 43:447-450.
Pazner, E.A., and D. Schmeidler (1978), "Egalitarian-equivalent allocations: a new concept of economic
equity", Quarterly Journal of Economics 92:671-687.
Pigou, A.C. (1920), The Economics of Welfare (Macmillan, London); 4th Edition 1950.
Pollak, R.A. (1979), "Bergson-Samuelson social welfare functions and the theory of social choice", The
Quarterly Journal of Economics 93:73-90.
Rawls, J. (1971), A Theory of Justice (Harvard University Press, Cambridge, Mass.).
Roberts, K.WS. (1980a), "Interpersonal comparability and social choice theory", Review of Economic
Studies 47:421-439.
Roberts, K.WS. (1980b), "Social choice theory: the single-profile and multi-profile approaches", Review
of Economic Studies 47:441-450.
Roberts, K.W.S. (1995), "Valued opinions or opinionated values: the double aggregation problem", in:
K. Basu, P. Pattanaik and K. Suzumura, eds., Choice, Welfare, and Development (Oxford University
Press, Oxford) pp. 140-165.
Roemer, J.E. (1986), "The mismarriage of bargaining theory and distributive justice", Ethics 97:88-110.
Roemer, J.E. (1990), "Welfarism and axiomatic bargaining theory", Recherches Economiques de Louvain
56:287-301.
Roemer, J.E. (1996), Theories of Distributive Justice (Harvard, University Press, Cambridge, Mass.)
Sarin, R., and P Wakker (1994), "Folding back in decision tree analysis", Management Science 40:
625-628.
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San Francisco).
Sen, A.K. (1973), On Economic Inequality (Clarendon Press, Oxford).
Sen, A.K. (1977), "On weights and measures: informational constraints in social welfare analysis",
Econometrica 45:1539-1572.
Sen, A.K. (1979), "Utilitarianism and welfarism", Journal of Philosophy 76:463-489.
Sen, A.K. (1980), "Equality of what?", The Tanner Lectures on Human Values, Vol. 1 (Cambridge
University Press, Cambridge); reprinted in A.K. Sen, 1982, Choice, Welfare and Measurement (Basil
Blackwell, Oxford) pp. 353-369.
Sen, A.K. (1982), Choice, Welfare and Measurement (Basil Blackwell, Oxford).
Sen, A.K. (1986a), "Social choice theory", in: K.J. Arrow and M.D. Intriligator, eds., Handbook of
Mathematical Economics, Vol. III (North-Holland, Amsterdam) pp. 1073-1181.
Sen, A.K. (1986b), "Information and invariance in normative choice", in: W Heller, R. Starr and
D. Starrett, eds., Social Choice and Public Decision-Making, Essays in Honor of K.J. Arrow, Vol. 1
(Cambridge University Press, Cambridge) pp. 29-55.
Sen, A.K. (1997), "Individual preference as the basis of social choice", in: K.J. Arrow, A.K. Sen and
K. Suzumura, eds., Social Choice Re-Examined, Vol. I (Macmillan Press, London) pp. 15-38.
Ch. 10: Social Welfare Functionalsand InterpersonalComparability 541

Sen, A.K., and B. Williams (1982), "Introduction", in A.K. Sen and B. Williams, eds., Utilitarianism
and Beyond (Cambridge University Press, Cambridge) pp. 1-21.
Shapley, L.S. (1969), "Utility comparison and theory of games", in: La Decision: Agrrgation et
Dynamique des Ordres de Preferences (Editions du CNRS, Paris) pp. 251-263.
Shapley, L.S., and M. Shubik (1975), "Game theory in economics, Chapter 4, References and utility",
Rand report R-904/4-NSF (Rand Corporation, Santa Monica).
Shorrocks, A.E (1983), "Ranking income distributions", Economica 50:3-17.
Shubik, M. (1982), Game Theory in the Social Sciences (MIT Press, Cambridge, Mass.).
Sprumont, Y. (1996), "Axiomatizing ordinal welfare egalitarianism when preferences may vary", Journal
of Economic Theory 68:77-110.
Strasnick, S. (1976), "Social choice and the derivation of Rawls' difference principle", Journal of
Philosophy 73:184-194.
Suppes, P. (1957), "Two formal models for moral principles", Techn. Report No 15, Office of Naval
Research Contract 225(17) (Applied Math. & Statistics Lab., Stanford University) pp. 17-18.
Suppes, P. (1966), "Some formal models of grading principles", Synthese 6:284-306.
Suzumura, K. (1996), "Interpersonal comparisons and the possibility of social choice", in: K.J. Arrow,
A.K. Sen and K. Suzumura, eds., Social Choice Re-Examined, Vol. II (Macmillan Press, London)
pp. 202-229.
Suzumura, K. (2002), "Welfare, rights and social choice procedures", in: K.J. Arrow, A.K. Sen
and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam)
forthcoming.
Thomson, W (2002), "Fair allocations", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of
Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) forthcoming.
Tsui, K.-Y., and J.A. Weymark (1997), "Social welfare orderings for ratio-scale measurable utilities",
Economic Theory 10:241-256.
Tungodden, B. (1999), "The distribution problem and Rawlsian reasoning", Social Choice and Welfare
16:599-614.
Vickrey, WS. (1945), "Measuring marginal utility by the reactions to risk", Econometrica 13:319-333.
Weymark, J.A. (1981), "Generalized Gini inequality indices", Mathematical Social Sciences 1:409-430.
Weymark, J.A. (1993), "Harsanyi's social aggregation theorem and the weak Pareto principle", Social
Choice and Welfare 10:209-221.
Wilson, R.B. (1972), "Social choice theory without the Pareto principle", Journal of Economic Theory
5:478-486.
Yaari, M.H. (1978), "Rawls, Edgeworth, Shapley, Nash: theories of distributive justice re-examined",
Journal of Economic Theory 24:1-30.
Young, H.P (1975), "Social choice scoring functions", SIAM Journal of Applied Mathematics 28:
824-838.
Young, H.P. (1987), "On dividing an amount according to individual claims or liabilities", Mathematics
of Operations Research 12:398-414.
Young, H.P. (1995), Equity: in Theory and in Practice (Princeton University Press, Princeton, NJ).
Chapter 11

UTILITARIANISM AND THE THEORY OF JUSTICE*

CHARLES BLACKORBY l, WALTER BOSSERT 2 and DAVID DONALDSON 3

1University of British Columbia and GREQAM; 2 Universite de Montreal and C.R.D.E.;
3University of British Columbia

Contents
Abstract 544
Keywords 544
1. Introduction 545
2. Social-evaluation functionals 549
3. Welfarism 553
4. Generalized utilitarianism 560
5. Utilitarianism 566
6. Variable-population extensions 571
7. Uncertainty 581
8. Conclusion 590
References 591

* We thank Don Brown, Marc Fleurbaey, Philippe Mongin, John Weymark and a referee for comments
and suggestions. Financial support through a grant from the Social Sciences and Humanities Research
Council of Canada is gratefully acknowledged.

Handbook of Social Choice and Welfare, Volume 1, Edited by KJ Arrow, A.K. Sen and K Suzumura
© 2002 Elsevier Science B. V All rights reserved
544 C Blackorby et ai.

Abstract

This chapter provides a survey of utilitarian theories ofjustice. We review and discuss
axiomatizations of utilitarian and generalized-utilitarian social-evaluation functionals
in a welfarist framework. Section 2 introduces, along with some basic definitions,
social-evaluation functionals. Furthermore, we discuss several information-invariance
assumptions. In Section 3, we introduce the welfarism axioms unrestricted domain,
binary independence of irrelevant alternatives and Pareto indifference, and use them
to characterize welfarist social evaluation. These axioms imply that there exists a single
ordering of utility vectors that can be used to rank all alternatives for any profile of
individual utility functions. We call such an ordering a social-evaluation ordering, and
we introduce several examples of classes of such orderings. In addition, we formulate
some further basic axioms. Section 4 provides characterizations of generalized-
utilitarian social-evaluation orderings, both in a static and in an intertemporal
framework. Section 5 deals with the special case of utilitarianism. We review some
known axiomatizations and, in addition, prove a new characterization result that uses an
axiom we call incremental equity. In Section 6, we analyze generalizations of utilitarian
principles to variable-population environments. We extend the welfarism theorem
to a variable-population framework and provide a characterization of critical-level
generalized utilitarianism. Section 7 provides an extension to situations in which the
alternatives resulting from choices among feasible actions are not known with certainty.
In this setting, we discuss characterization as well as impossibility results. Section 8
concludes.

Keywords

social choice, utilitarianism, welfarism

JEL classification: D63, D71

Ch. 11: Utilitarianism and the Theory of Justice 545

1. Introduction

In A Theory of Justice, Rawls (1971) describes justice as "the first virtue of social
institutions" (p. 3) and identifies "the primary subject of justice" as "the basic structure
of society, or more exactly, the way in which the major social institutions distribute
fundamental rights and duties and determine the division of advantages from social
cooperation" (p. 7)1. The view of justice investigated in this chapter asserts that a
just society is a good society: good for the individual people that comprise it. To
implement such an approach to justice, the social good is identified and used to rank
social alternatives. Of the alternatives that are feasible, given the constraints of human
nature and history, the best is identified with justice. Even if the best alternative is
not chosen, however, better ones are considered to be more just than worse ones. If
societies are not perfectly just, therefore, social improvements can be recognized.
Social choices are not made in isolation, however. Decisions made in a particular
society affect people in other parts of the world and people who are not yet born. In
addition, both the number and identities of future people are influenced by choices
made in the present. For that reason, principles that identify the social good are
typically extended to rank complete histories of the world (or the universe if necessary)
from remote past to distant future2 . The principle that asserts that a just society is a
good society must be qualified, therefore, with a ceteris paribus clause.
In this chapter, we investigate a particular conception of the social good, one that
is based exclusively on individual good or well-being. Principles that reflect this view
are called welfarist [Sen (1979)] and they treat values such as freedom and individual
autonomy as 'instrumental' - valuable only because of their contribution to well-being.
Because of this, it is important to employ a comprehensive notion of well-being such
as that of Griffin (1986) or Sumner (1996). We therefore focus on lifetime well-being
and include enjoyment, pleasure and the absence of pain, good health, length of life,
autonomy, liberty, understanding, accomplishment and good human relationships as
aspects of it.
Welfarist principles are sometimes criticized as taking a narrow view of being a
person, seeing them as "locations of their respective utilities" only [Sen and Williams
(1982, p. 4)]. The use of comprehensive accounts of lifetime well-being, such as those
of Griffin (1986) and Sumner (1996), which attempt to take account of everything in
which individual people have an interest, is, in our view at least, sufficient to answer

See Murphy (1998) for a discussion of the relationship between the principles of social justice and
the principles that guide individual conduct.
2 Because some non-human animals are sentient capable of having experiences - their interests are
often included. Sidgwick (1966, p. 414) argues that we should "extend our concern to all the beings
capable of pleasure and pain whose feelings are affected by our conduct". Throughout the chapter,
however, we assume that only human well-being counts in social evaluation, a simplification that makes
our presentation simpler. Readers who are interested in the extension of welfarist principles to non-human
sentient creatures are referred to Blackorby and Donaldson (1992).
546 C. Blackorby et ai.

it. Welfarism is mainly a consequence of Pareto indifference which is in turn implied

by an axiom that we call minimal individual goodness. If one alternative is ranked
as better than another, the axiom requires it to be better for at least one individual
[Goodin (1991)]. Any non-welfarist principle must, therefore, run the risk of claiming
that a social change is good even though no one benefits.
Popularized by Bentham (1789), utilitarianism is a welfarist principle that can be
used to rank social alternatives according to their goodness 3 . Utility is an index
of individual lifetime well-being and, for a fixed population, utilitarianism declares
alternative x to be better than alternative y if and only if total utility is greater in x
than in y. Although this principle is unconcerned with the distribution of any fixed
total utility, it is not unconcerned with income inequality or social provision for special
needs.
The utilitarian indifference to inequality of well-being has prompted the complaint
that "persons do not count as individuals in this any more than individual petrol
tanks do in the analysis of the national consumption of petroleum" [Sen and Williams
(1982, p. 4)]. This criticism does not apply to welfarist principles that are averse to
utility inequality, however: they rank more equal distributions of utility as better than
less equal ones.
A family of principles whose value functions have the same additively separable
mathematical structure as the utilitarian value function is the generalized-utilitarian
family of principles, which includes utilitarianism as a special case. Each of these
principles employs transformed utilities, and some exhibit aversion to utility inequality.
The generalized-utilitarian principles satisfy an important property: if a social change
affects the utilities of a particular group of individuals only, the ranking of such
changes is independent of the utility levels of others. This means that independent
subprinciples exist for subgroups (including generations) and are consistent with
the overall principle. Both the utilitarian and generalized-utilitarian families can be
extended to cover changes in population size and composition.
A social-evaluation functional assigns a ranking of all possible social alternatives
to every admissible profile of utility functions, where a profile contains one utility
function for each member of society. Arrow's (1951, 1963) seminal contribution to
social-choice theory employs a domain that uses individual preference information
only. His impossibility result can be avoided if the domain is changed to include
profiles of utility functions together with conditions that ensure that utilities are,
to some degree, interpersonally comparable 4 . There are, of course, other ways of
avoiding Arrow's impossibility result by removing or weakening one or more of
his axioms but, in our opinion, the most natural way to proceed is to allow for
interpersonal comparisons. Although the principles discussed in this chapter all
require a certain amount of interpersonal comparability, numerically meaningful utility

3 See also Mill (1861) and Sidgwick (1907) for other early formulations of utilitarianism.
4 See Sen (1970).
Ch. 11: Utilitarianismand the Theory of Justice 547

functions are not necessary for most of them. For example, utilitarianism requires
only cardinal measurability of individual utilities - with utility functions that are
unique up to increasing affine transformations - and interpersonal comparability of
utility gains and losses between pairs of alternatives (cardinal measurability and
unit comparability). In a two-person society, if one person gains more in moving
from y to x than the other loses, utilitarianism declares x to be socially better
than y.
In this chapter, we survey a set of results in social-choice theory that provides
an axiomatic basis for generalized utilitarianism and, in some cases, utilitarianism
itself. Although we are concerned, for the most part, with comparisons of alternatives
according to their goodness, we include a brief discussion of induced rankings.
Utilitarianism and generalized utilitarianism can be used to rank combinations of
institutions (including legal and educational ones), customs and moral rules, taking
account of the constraints of history and human nature. If each of these combinations
leads with certainty to a particular social alternative, they can be ranked with any
welfarist principle. If, however, consequences are uncertain, the problem is more
difficult. One way of doing it is to attach probabilities (possibly subjective) to a
set of 'states of nature' and use them to rank prospects: lists of corresponding
alternatives.
There are different ways to make use of the resulting rankings. One is to take an
uncomplicated maximizing approach and recommend to governments and individuals
alike that they choose the best feasible action, a position taken by act utilitarians such
as Sidgwick (1907). Other utilitarians realize that it may be impossible or unwise
for individuals to engage in complicated assessments of consequences and, instead,
follow general rules which are "constantly evolving, but on the whole stable, such that
their use in moral education, including self-education, and their consequent acceptance
by society at large, will lead to the nearest possible approximation to archangelic
thinking" [Hare (1982, p. 33)], a position sometimes called rule utilitarianism. It is
true, in addition, that the best actions may require an individual, state or generation
to make very great sacrifices. Because concern for others is bounded, it may be
important to choose rules that limit the sacrifices of agents or declare some actions
to be supererogatory: beyond the call of duty5 . Mill (1861), a utilitarian who took
such considerations seriously, was not a simple maximizer [see, for example, Brown
(1972)].
Section 2 introduces social-evaluation functionals. They make use of some or
all available utility and non-utility information to rank alternatives according to
their goodness. In addition, the section provides a formal account of information
environments and information-invariance conditions. Information environments are
described by partitions of profiles of utility functions into equivalence classes. They
determine which statements involving both intrapersonal and interpersonal utility

5 See also Blackorby, Bossert and Donaldson (2000) for a discussion of limited altruism.
548 C. Blackorbv et ai.

comparisons are meaningful. Information-invariance conditions specify the amount of

utility information that a social-evaluation functional may make use of 6 .
Section 3 turns to welfarism and it is shown that, given an unlimited domain
of utility profiles, welfarism is characterized by the axioms binary independence of
irrelevant alternatives and Pareto indifference. Section 4 contains a set of results that
characterize generalized utilitarianism. Section 5 focuses on utilitarianism itself and
presents characterization theorems for it. The axioms employed in Sections 4 and 5 are
of two types. Some are information assumptions and they require the social-evaluation
functional to make use of meaningful utility information only. Other axioms, such as
strong Pareto and anonymity, are ethical in nature. Anonymity, for example, captures
the idea of impartiality in social evaluation, an essential feature of many welfarist
principles.
Population issues are the concern of Section 6 which discusses extensions of
utilitarianism and generalized utilitarianism to environments in which alternatives may
differ with respect to population size and composition. The axioms used in that section
strengthen the case for the utilitarian and generalized-utilitarian principles and, in
addition, characterize families of principles that extend the fixed-population principles
in an ethically attractive way. These families are known as critical-level utilitarianism
and critical-level generalized utilitarianism. They require the specification of a fixed
'critical level' of lifetime utility above which additions to a population are, ceteris
paribus, valuable.
Section 7 makes use of subjective probabilities to rank prospects. We present a
multi-profile version of Harsanyi's (1955, 1977) social-aggregation theorem. Instead of
using lotteries as social alternatives, we employ prospects and assume that probabilities
are fixed and common to all individuals and the social evaluator. Both individual
ex-ante utilities and social preferences are assumed to satisfy the expected-utility
hypothesis [von Neumann and Morgenstern (1944, 1947)] and individual ex-ante
utilities are equal to the expected value of von Neumann-Morgenstern utilities [this
is called the Bernoulli hypothesis by Broome (1991a)]. Given that, we show that any
welfarist ex-ante social-evaluation functional satisfying anonymity and the weak Pareto
principle is utilitarian. Harsanyi (1953) presents another argument for utilitarianism
in his impartial-observer theorem. It is discussed in detail in Mongin (2001) and is
omitted in this survey.
Welfarist social evaluation is an attractive option but it is not the only one. Many
people who reject welfarism do not believe that welfare considerations are completely
irrelevant to social decision-making, however. Accordingly, the results of this chapter
should be of interest to most people who are concerned with social evaluation. Other
theories, such as Sen's (1985) treatment of functionings and capabilities, can be
modified to fit our framework. Functionings are the things that people can do, and the
idea can be used to provide an account of (some aspects of) well-being. Capabilities

6 For a more complete guide to these requirements, see d'Aspremont and Gevers (2002).
Ch. II: Utilitarianism and the Theory of Justice 549

are opportunities and include freedoms. If the two are aggregated into a single index of
'advantage' for each person, welfarist social-evaluation functionals can be used to rank
alternatives. On the other hand, it is possible to use welfarist principles to aggregate
functionings into an index of social functioning, which would be only one factor in
overall social evaluation .
A major challenge to welfarism has appeared in recent years. It replaces concern for
well-being with concern for opportunities for well-being on the grounds that individual
people are responsible for their choices (in certain circumstances they may be thought
to be responsible for their preferences as well) 8. In practice, welfarists often agree that
the provision of opportunities is socially warranted, but their concern is with actual
well-being. If autonomy is a significant aspect of well-being, people must be free to
make important choices for themselves, and this provides a constraint which restricts
the feasible set of social possibilities. By way of analogy, parents typically provide
opportunities to their children, but that does not mean that opportunities are what they
care about.

2. Social-evaluation functionals

Social-evaluation functionals use information about the members of a set of possible

social alternatives to rank them according to their social goodness. An alternative is
a complete history of the world from remote past to distant future. Let X be a set of
alternatives that contains at least three members (some slightly stronger requirements
on the minimal number of alternatives are employed in Sections 6 and 7). No other
restrictions are imposed on X: it may be finite, countably infinite, or uncountable.
The set of individuals in an n-person society is {1, ... , n}, where n E Z++ 9. Except
for our discussion in Section 6, we consider only comparisons of alternatives with the
same population. For any i {1, ... , n}, U: X - 1 is i's utility function 0 and
ui = U,(x) is the utility level of individual i C {1, ... , n} in alternative x E X. Utilities
are interpreted as indicators of lifetime well-being and measure how good a person's
life is from his or her own point of view. This does not mean that the utility function Ui
is a representation of person i's actual preferences. Preferences and utility functions
may be inconsistent because of individual non-rationalities, altruism or insufficient
information 11 .

7 Suzumura (1999) considers the value of social procedures in addition to the value of individual
well-being.
8 See, for example, Arneson (1989, 2000), Roemer (1996) and, for an unsympathetic critique, Anderson
(1999). See also Fleurbaey and Maniquet (2002) and Foster (2002).
9 Z_ is the set of positive integers and Z, is the set of nonnegative integers.
10 7Z, R,+and 7ZR+are the sets of all real numbers, nonnegative real numbers and positive real numbers
respectively. In addition, 1,, = (1 .. 1)E R".
1n See Broome (1991a) and Mongin and d'Aspremont (1998) for discussions of individual well-being
and its relationship to preferences, information and self-interest. Hammond (2001) offers a very different
account, interpreting "individual welfare as a purely ethical concept".
550 C. Blackorby et al.

A profile of utility functions is an n-tuple U = (U1, ... , U~) with one utility function
for each individual in society. The set of all possible profiles is denoted by U. For
U C U and x X, we write U(x) = (Ul(x),..., U(x)). This vector represents
the welfare information for alternative x given the profile U. In addition to welfare
information, non-welfare information may be available, and we assume that each x in
the set X contains a full description of all non-welfare aspects of alternatives that may
be considered relevant for social evaluation.
A social-evaluation functional is a mapping F: D --- , where 0 D C U and
O is the set of all orderings on X 12. D is the domain of admissible utility profiles
and it may consist of a single profile or many profiles. In the latter case, the social-
evaluation functional can cope with different profiles of utility functions, and inter-
profile consistency conditions such as binary independence of irrelevant alternatives
or various information-invariance conditions (see below) may be imposed on it. The
social-evaluation functional may make use of non-welfare information in addition to
welfare information contained in the profile U. For simplicity of notation, we write
Ru = F(U); Iu and Pu are the symmetric and asymmetric components of Ru. For
any x,y E X, xRuy means that x is socially at least as good as y, xluy means that x
and y are equally good, and xPuy means that x is socially better than y.
General social-evaluation functionals may make use of both welfare and non-
welfare information, but welfarist functionals ignore non-welfare information and
completely non-welfarist functionals ignore welfare information, making use of non-
welfare information only. In the latter case, a single ordering of the alternatives in X
is produced because only a single set of non-welfare information is available.
In a multi-profile environment, it is possible to restrict the welfare information
that the social-evaluation functional F may make use of. This is done by parti-
tioning D into subsets of informationally equivalent profiles called information sets.
Usable information in a profile in D is that which all informationally equivalent
profiles in the corresponding information set have in common. If utilities are
ordinally measurable and interpersonally noncomparable, for example, two profiles
U, V E D are informationally equivalent if and only if there exist increasing
functions 01, ... , 0q with i: -- 7 for all i C {1, ... , n} such that
(Vl(x), ... , V,(x)) = (l(U 1(x)), . ., 0,(U,2(x))) for all x C X. The utility comparison
Ul ( x) > UI(y) is meaningful in such an environment because it is true in all
informationally equivalent profiles or false in all of them. On the other hand, if ID= U,
the interpersonal comparison Ul(x) > U2(x) is not meaningful because it is true in
some informationally equivalent profiles and false in others.
Restrictions on social-evaluation functionals imposed by the available information
regarding the measurability and interpersonal comparability of individual utilities can
be summarized using information-invariance conditions. In order to represent the

12 An ordering is a reflexive, transitive and complete binary relation. Social-evaluation functionals are
also referred to as social-welfare functionals and were introduced by Sen (1970).
Ch. II: Utilitarianism and the Theory of Justice 551

informational environment, the set of admissible profiles is partitioned into information

sets, and an information-invariance condition requires F to be constant on each of
them. That is, if two profiles U and V are informationally equivalent, Ru and Rv are
identical.
A partition of D into information sets can be defined using an equivalence
relation - on D with an information set given by an equivalence class of 13. That is,
for U, V C D, U - V if and only if U and V are informationally equivalent. A social-
evaluation functional F satisfies information invariance with respect to the information
environment described by - if and only if it assigns the same social ordering to all
profiles in an equivalence class of -.

Information Invariance with Respect to~:

For all U, V D, if U V, then Ru =Rv.

The most commonly used approach to formalizing various types of informational

assumptions identifies the equivalence relation - by specifying a set of admissible
transformations of utility profiles that lead to informationally equivalent profiles 14
An invariance transformation is a vector 0 = (1, ... , O,) of functions i: 1Z - 7Z
for all i {1, ... , n} whose application to a profile U results in an informationally
equivalent profile. Let denote the set of invariance transformations used to generate
the equivalence relation -. That is, for all U, V cE D, U - V if and only if there exists
E 0 such that V = o U, where o denotes component-by-component function
composition 15
Various information assumptions that can be expressed in terms of admissible
transformations have been considered in contributions by Blackorby and Donaldson
(1982), Blackorby, Donaldson and Weymark (1984), d'Aspremont and Gevers (1977),
DeMeyer and Plott (1971), Dixit (1980), Gevers (1979), Roberts (1980b) and Sen
(1970, 1974, 1977a, 1986) among others. We restrict attention to the information
assumptions that are relevant for the purposes of this chapter, and refer the interested
reader to Bossert and Weymark (2001) or d'Aspremont and Gevers (2002) for more
detailed treatments. Each of these assumptions is defined by specifying the set of
invariance transformations which induces the equivalence relation - that partitions D7
into sets of informationally equivalent utility profiles. For each information assumption

13 An equivalence relation is a reflexive, transitive and symmetric binary relation.

14 This approach was developed in contributions such as d'Aspremont and Gevers (1977), Roberts
(1980a,b) and Sen (1974). See Basu (1983), Bossert (1991, 2000), Bossert and Stehling (1992, 1994),
Falmagne (1981), Fishburn, Marcus-Roberts and Roberts (1988), Fishburn and Roberts (1989) and
Krantz, Luce, Suppes and Tversky (1971) for discussions of information-invariance assumptions in
terms of meaningful statements and their relations to uniqueness properties of measurement scales.
15 We only consider sets of invariance transformations such that the resulting relation - is an
equivalence relation. See Bossert and Weymark (2001) for a discussion and for conditions guaranteeing
this.
552 C. Blackorby et al.

listed below, we implicitly assume that the domain D contains all profiles that are
informationally equivalent to U for each U D. This is true, in particular, for the
unrestricted domain D = / on which we focus for most of the chapter. Additional
information assumptions are introduced in Section 6 in a variable-population context.
If utilities are cardinally measurable, individual utility functions are unique up
to increasing affine transformations, thereby allowing for intrapersonal comparisons
of utility differences. If, in addition, some comparisons of utility are meaningful
interpersonally, these transformations must be restricted across individuals. An
example is cardinal unit comparability. In that information environment, admissible
transformations are increasing affine functions and, in addition, the scaling factor must
be the same for all individuals. This information assumption allows for interpersonal
comparisons of utility differences, but utility levels cannot be compared interpersonally
because the intercepts of the affine transformations may differ across individuals.

Cardinal Unit Comparability (CUC):

0E if and only if there exist al, ... ,an e k and b E R,+_ such that (r) =
ai + b for all t E R and for all i E 1,... n}.

An information environment that provides more information than CUC is one in

which the unit in which utilities are measured is numerically significant. In this case, we
say that utilities are translation-scale measurable. Utility differences are interpersonally
comparable and, in addition, their numerical values are meaningful. Because the
functions (0, ... , 0, may be different for each person, utility levels are, again, not
interpersonally comparable.

Translation-Scale Measurability (TSM):

p E r if and only if there exist al, ... , a, C1 such that ¢i(T) = ai + T for all
TEC and for all iE 1,..., n}.

If utilities are cardinally measurable and fully interpersonally comparable, both

utility levels and differences can be compared interpersonally. In this case, utility
functions are unique up to increasing affine transformations which are identical across
individuals.

CardinalFull Comparability (CFC):

0 E ( if and only if there exist a E R and b E ?++ such that pi(r) = a bT
for all r E R and for all i {1, ... , n}.

CFC defines a finer partition of the set of admissible utility profiles than CUC and,
therefore, places weaker invariance requirements on the social-evaluation functional.
Ch. I. Utilitarianism and the Theory of Justice 553

If all the information in a profile is meaningful, we say that utilities are numerically
measurable and fully interpersonally comparable. In this case, each information set
consists of a singleton.

Numerical Full Comparability (NFC):

0 e i if and only if i(T)= r for all 'r e R and for all i {1,..., n}.

In general, increases in available information reduce the restrictions on F implied by

the information-invariance condition. For example, information invariance with respect
to TSM is a weaker restriction than information invariance with respect to CUC, and
invariance with respect to NFC provides no restriction at all.

3. Welfarism

The orderings on X generated by welfarist social-evaluation functionals compare any

two alternatives x,y E X solely on the basis of the individual utilities experienced in x
and iny. All non-welfare information is ignored when establishing the social ranking.
Welfarism is a consequence of three axioms, the first of which is an unrestricted-
domain assumption. This axiom requires the social-evaluation functional F to be
defined on the set of all possible utility profiles.

UnrestrictedDomain (UD): = U.

The next axiom is an independence condition which links the orderings associated
with different profiles. It requires the social ranking of any two alternatives to be
independent of the utility levels associated with other alternatives.

Binary Independence of Irrelevant Alternatives (BI):

For all x,y e X, for all U, V E D, if U(x) = V(x) and U(y) = V(y), then
xRuy if and only if xRvy.

The above independence axiom for social-evaluation functionals is weaker than the
corresponding independence axiom for social-welfare functions [see Arrow (1951,
1963) and Sen (1970)]. Arrow's independence axiom requires the social ordering of a
pair of alternatives to depend only on the individual rankings of the two alternatives.
BI is equivalent to Arrow's binary-independence axiom if the social-evaluation
functional satisfies information invariance with respect to ordinally measurable,
interpersonally noncomparable utilities. As formulated above, binary independence
is compatible with any assumption concerning the measurability and interpersonal
comparability of individual utilities.
The final axiom used to generate welfarism is Pareto indifference. If all individuals
554 C. Blackorby et al.

are equally well off in two alternatives, it requires the social-evaluation functional to
rank them as equally good.

Pareto Indifference (PI):

For all x,y C X, for all U C D, if U(x) = U(y), then xluy.

Pareto indifference is an attractive axiom if utility functions measure everything that

is of value to individuals. For that reason, we endorse comprehensive accounts of
lifetime utility such as the ones provided by Griffin (1986) and Sumner (1996)16
Griffin includes enjoyment, pleasure and the absence of pain, good health, autonomy,
liberty, understanding, accomplishment and good human relationships as aspects of
well-being. He argues, in addition, that there is a moral dimension to well-being.
Sumner also discusses the role that individual attitudes can play.
Pareto indifference is a consequence of the fundamental welfarist view that, if one
alternative is ranked as better than another, it must be better for at least one individual
[see Goodin (1991)]. We call this axiom minimal individual goodness, and it is defined
formally as follows:

Minimal Individual Goodness (MIG):

For all x,y e X, for all U e D, if xPuy, then there exists k C 1, ... , n}
such that Uk(x) > Uk(y).

Minimal individual goodness is equivalent to the conjunction of Pareto indifference

and the following Pareto-weak-preference axiom:

Pareto Weak Preference (PWP):

For all x,y E X, for all U C , if Ui(x) > Ui(y) for all i E 1, ... , n} with at
least one strict inequality, then xRuy.

We obtain
Theorem 1. A social-evaluationfunctional F satisfies MIG if and only if F satisfies
PI and PWP
Proof: Suppose F satisfies MIG. We first prove by contradiction that PI is satisfied.
Suppose not. Then there exist x,y X and U D such that U(x) = U(y) and
not xluy. Because Ru is complete, we must have either xPuy or yPux. In either case,
we obtain a contradiction to MIG.
Now suppose F violates PWP. Then there exist x,y X and U D such that
Ui(x) > Ui(y) for all i E 1, ... , n} with at least one strict inequality and not xRuy.
By the completeness of Ru, we must have yPux, again contradicting MIG.

16 See also Broome (1991a) and Mongin and d'Aspremont (1998) for accounts based on self-interested
preferences under conditions of full information.
Ch. 11: Utilitarianism and the Theoty of Justice 555

Finally, suppose F satisfies PI and PWP but violates MIG. Then there exist x,y e X
and U E D such that xPuy and U(y) > Ui(x) for all i {1, ... , n}. If
Ui(y) = U(x) for all i {1, ... , n}, we obtain a contradiction to PI, and if there
exists k E {1, ... , n} such that Uk(y) > Uk(x), we obtain a contradiction to PWP. C
In the presence of unrestricted domain, BI and PI together imply that non-welfare
information about the alternatives must be ignored by the social-evaluation functional.
If, in one profile, utility numbers for a pair of alternatives are equal to the utility
numbers for another pair in a possibly different profile, the rankings of the two pairs
must be the same. This property is called strong neutrality.

Strong Neutrality (SN):

For all x,y,z, w EX, for all U, V C D, if U(x) = V(z) and U(y) = V(w),
then xRuy if and only if zRvw.

We obtain the following theorem [see, for example, Blau (1976), d'Aspremont and
Gevers (1977), Guha (1972) and Sen (1977a), for this and related results]:
Theorem 2. Suppose that a social-evaluationfunctional F satisfies UD. F satisfies
B and PI if and only if F satisfies SN.
Proof: First, suppose that F satisfies UD and SN. That BI is satisfied follows
immediately by setting x = z and y = w in the definition of SN. Setting U = V
and y = z = w, SN implies that xRuy if and only if yRuy when U(x) = U(y). Because
Ru is reflexive, this implies xIuy, which demonstrates that PI is satisfied.
Now suppose that F satisfies UD, BI and PI. Suppose that U(x) = V(z) = u and
U(y) = V(w) = v. By UD, there exists an alternative x E X and profiles U, U, U · D
such that U(x) = U(x) = u and U(y) = v, U(z) = U(x) = u and U(w) = v, and
U(x) = u and U(y) = U(w) = v. By BI, xRuy if and only if xRoy. By PI and the
transitivity of Ru, it follows that xRoy if and only if SRoy. A similar argument implies
that .RUy if and only ifS Roy if and only if SRow. Applying the same argument once
again, we have RfR&w if and only if iRw if and only if zR&w. By BI, zRow if and only
if zRvw. Therefore, xRuy if and only if zRvw which proves that F satisfies SN. D1

Given unrestricted domain, SN is equivalent to the existence of an ordering R on R"

which can be used to rank the alternatives in X for any utility profile U. The social
betterness (strict preference) relation and the equal-goodness (indifference) relation
corresponding to R are denoted by P and I, respectively. We refer to R as a social-
evaluation ordering. 17 Combined with Theorem 2, this observation yields the following
welfarism theorem [see d'Aspremont and Gevers (1977) and Hammond (1979)] 8:

'7 Gevers (1979) calls R a social-welfare ordering.

18 Bordes, Hammond and Le Breton (1997) and Weymark (1998) prove variants of this theorem with
specific domain restrictions, that is, with weaker domain assumptions than UD.
556 C. Blackorby et al.

Theorem 3. Suppose that a social-evaluationfunctional F satisfies UD. F satisfies

BI and PI if and only if there exists a social-evaluation ordering R on R" such that,
for all x,y C X andfor all U E D,

xRuy 4~ U(x)RU(y). (3.1)

Proof: If there exists a social-evaluation ordering R such that, for all x,y i X and all
U E D, Equation (3.1) is satisfied, BI and PI are satisfied.
Now suppose that F satisfies UD, BI and PI. By Theorem 2, F satisfies SN. Define
the relation R as follows. For all u, v E R", uRv if and only if there exist a profile
U C D and two alternatives x,y e X such that U(x) = u, U(y) = v, and xRuy. By SN,
the relative ranking of any two utility vectors u and v does not depend on the profile U
or on the alternatives x and y used to generate u and v and, therefore, R is well-
defined. That R is reflexive and complete follows immediately because Ru is reflexive
and complete for all U · D. It remains to be shown that R is transitive. Suppose
u, v, w E 1Zn are such that uRv and vRw. By UD and the maintained assumption that
X contains at least three elements, there exists a profile U D and three alternatives
x,y,z E X such that U(x) = u, U(y) = v, and U(z) = w. Because U(x)RU(y) and
U(y) hU(z), it follows that xRuy and yRuz by definition of R. Transitivity of Ru then
implies that xRuz. Hence, U(x)RU(z) or, equivalently, uw which shows that R is
transitive. O
Note that the social-evaluation ordering R in the statement of Theorem 3 is profile-
independent. Pairs of alternatives whose utility vectors are the same are ranked in the
same way, regardless of the utility profile. If D consists of a single profile, the result
is true but BI is not needed [Blackorby, Donaldson and Weymark (1990)].
For notational convenience, we concentrate on the social-evaluation ordering in
most of the remainder of the chapter. All axioms and results regarding this ordering
can be reformulated in terms of the social-evaluation functional F by defining the
properties analogous to those defined for R and adding the welfarism axioms.
Given a set of invariance transformations qP (and, hence, an equivalence relation -)
and the welfarism axioms UD, BI and PI, the information-invariance axiom for the
social-evaluation functional F is equivalent to an analogous condition formulated in
terms of the corresponding social-evaluation ordering R:

Information Invariance with Respect to P):

For all u, v, u', v' · In, if there exists 0 e P such that u' = (u) and v' = (u),
then uRv if and only if u'Rv'.
Ch. 11: Utilitarianismand the Theory of Justice 557

Next, we introduce some axioms that are commonly required in welfarist social
evaluation. Continuity is a regularity condition. It ensures that 'small' changes in
individual utilities do not lead to 'large' changes in social rankings.
Continuity (C):
For all u E R"', the sets {v E R'n uRu} and {u E Rn I uRv} are closed in n".

Anonymity ensures that the ordering R treats individuals impartially, paying no

attention to their identities. That is, any permutation of a given utility vector must
be as good as the utility vector itself. Note that this is a strengthening of Arrow's
(1951, 1963) condition that prevents the existence of a dictator.
Anonymity (A):
For all u e Rn, for all bijective mappings sr: {1, ... , n} - {1, ... , n},
uI (u~,(1), ·. Uff(,)).
, (3.2)
The weak Pareto principle requires an increase in everyone's utility to be regarded
as a social improvement.
Weak Pareto (WP):
For all u,v E Rn, if ui > vi for all i E {1, ... , n}, then uPv.
A strengthening of both weak Pareto and Pareto indifference is the strong Pareto
principle. In addition to Pareto indifference, it requires that if no one's utility has
decreased and at least one person's utility has increased, the change is a social
improvement.
Strong Pareto (SP):
For all u, v E Rn,
(i) if ui = vi for all i E {1, ... , n}, then ulv; and
(ii) if ui > vi for all i E 1, ... , n} with at least one strict inequality, then uP.
Note that (i) in the definition of SP is redundant in a welfarist framework - this
restriction is implied by the reflexivity of R. We have chosen to include it in the
definition of strong Pareto in order to follow the conventional terminology.
Finally, we introduce an axiom which prevents the social ordering from exhibiting
a strong version of inequality preference.
Minimal Equity (ME):
There exist i, j E {1, ... , n} and u, v e R'"such that uk = k for all
k {1, ... , n} \{i,j},vj > uj > ui > vi, and uRv.
See d'Aspremont (1985), d'Aspremont and Gevers (1977) and Deschamps and Gevers
(1978) for this axiom and Hammond (1976) for a related condition.
558 C. Blackorbv et al.

Continuity and the weak Pareto principle ensure the existence of a continuous
representation of the social-evaluation ordering R. We obtain

Theorem 4. If a social-evaluation ordering R satisfies C and WPI then, .for each

u E R ' , there exists a unique ~ = E(u) [min{ul, ... u},max{ul, ... , ,,}] such
that ui*1,,.

Proof: If n = 1, WP and reflexivity imply that, for all u, v E 7, uRv if and only if
u > v. Consequently, the result follows from letting ~ = -(u) = u (which is equal to
both maximum and minimum utility) for all u c R.
Suppose n > 2, let u E RI be arbitrary and suppose, by way of contradiction,
that min{ul, ... , u}lnPu. By C, there exists a neighborhood of u such that
min{ul, ... , u}ln, is preferred to all points in that neighborhood according to R.
Because this neighborhood contains points that strictly dominate u and, thus,
min{ul, ... , u,}l,, this contradicts WP. Therefore, uRmin{ul, ... , u,}l,,. Analo-
gously, it follows that max{ul,..., u,,}lRu. By C, it follows that there exists
E [min{ul,..., u,},max{ul,..., u,}] such that ull1. WP implies that 2 must
be unique for each u and thus can be written as a function : ' - 7Z. C]
The representative utility 5 is analogous to the equally-distributed-equivalent income
used in ethical approaches to income-inequality measurement 19. The function . is
the representative-utility function corresponding to R, and it is easy to see that it is a
representation of R - that is, for all u, v C RZ',
uRv X E(u) > E(v). (3.3)
Furthermore, is continuous because R is, and WP implies that is weakly
increasing.
We conclude this section with some examples of welfarist social-evaluation
orderings. The utilitarian social-evaluation ordering uses the sum of the individual
utilities to make social comparisons. According to utilitarianism, for all u, v E Rn,
n al

uRv~
XE i >LV
E i (3.4)
i=l i-

The class of social-evaluation orderings that respect all strict rankings of utility
vectors according to utilitarianism is the class of weakly utilitarian orderings [see
Deschamps and Gevers (1978)]. R is weakly utilitarian if and only if

E
i 1
ui > E v
i-
" uv (3.5)
for all u, u E R ' .

'1 See, for example, Atkinson (1970), Dalton (1920), Kolm (1969) and Sen (1973).
Ch. 11: Utilitarianism and the Theory of Justice 559

The members of the class of generalized-utilitarian orderings perform social

comparisons by adding the transformed utilities of the members of society. If the
transformation applied to individual utilities is (strictly) concave, the resulting ordering
represents (strict) aversion to utility inequality. Formally, a social-evaluation ordering R
is generalized utilitarian if and only if there exists a continuous and increasing function
g: -- such that, for all u, v R'n,
n n

uRv X Cg(ui) > Eg(vi) (3.6)

i=l i=l

All generalized-utilitarian social-evaluation orderings satisfy C, A, and SP (and thus

WP). If g is concave, ME is satisfied as well.
Utilitarianism is a special case of generalized utilitarianism in which the transfor-
mation g is affine. An example of a class of generalized-utilitarian orderings is the
class of symmetric global means of order r. R is a symmetric global mean of order r
if and only if there exist /3,r R,++ such that, for all u, v cE ',

uRv >7 - >E |uJ > E7 7-- /3 7 ivjr.

iE{1.n}: iE{l .. n}: iE{1 ... n}: iE{1 . n}:
i >0 ui <0i
o >o0 i <0
(3.7)
In this case, g(T) = Tr for all r > 0 and g(r) = -/31
r for all r < 0. The special cases of
Equation (3.7) with r = 1 and /3 > 1 are of particular interest. They represent the only
cases that exhibit (weak) inequality aversion in all of "' [Blackorby and Donaldson
(1982)]. The resulting ordering is a modification of utilitarianism such that negative
utilities get an equal or higher weight than positive utilities. This principle requires
that the zero level of utility has some meaning. In the variable-population principles
of Section 6, it is used to represent the value of a 'neutral' life. Above neutrality, an
individual life, taken as a whole, is worth living from the viewpoint of the individual;
below neutrality, it is not (see Section 6).
The utilitarian social-evaluation ordering pays no attention to utility inequality
and, in contrast, the maximin ordering exhibits complete inequality aversion. It pays
attention to the utility of the worst-off individual only. For u E 7' " , let (u(l), ... , u(,))
be a permutation of u such that u(i) > U(i+l) for all i {1, ... , n - 1}. For all u, v E Rn,
the maximin ordering requires

uRv #>u¢i) > V(n). (3.8)

Leximin is a variant of maximin in which the utility vector u is socially preferred

to the utility vector v if the worst-off individual in u is better off than the worst-off
individual in v. In case of a tie, however, the two are not necessarily equally good.
Instead, the utilities of the next-to-worst-off individuals are used to determine the social
preference, and the procedure continues until either there is a strict preference or the
560 C. Blackorby et a/.

two utility vectors are permutations of each other, in which case they are declared
equally good. Formally, the leximin ordering is given by

uRv =:u is a permutation of v

or there exists aj E {1, ... , n} such that u(i) = V(i) for all i >j and u(j) > u(i)
(3.9)
for all u, v E R '". Maximin is continuous and violates the strong Pareto principle
(but satisfies weak Pareto) and leximin satisfies strong Pareto but not continuity. Both
orderings satisfy A and ME.
The extremely equality-averse counterpart of leximin is the leximax ordering, which
is defined by

uRv X= u is a permutation of v
or there exists aj E {1, ... , n} such that U(i) = (i) for all i <j and u(j) > (j)
(3.10)
for all u, v E 1Zn. Leximax satisfies A and SP (and therefore WP) but violates
C and ME.
The class of single-parameter Gini social-evaluation orderings provides another
possibility for a generalization of utilitarianism - their level sets are linear in rank-
ordered subspaces of T'. A social-evaluation ordering R is a single-parameter Gini
ordering if and only if there exists a real number 6 > 1 such that, for all u, v E n,

- -
uRv zi- E -(i 1)]U(i) > E i-(i 1)](i). (3.11)
i=1 i=1

The single-parameter Ginis are special cases of the generalized Ginis introduced by
Weymark (1981) and discussed in Bossert (1990a) and Donaldson and Weymark
(1980) in the context of ethical inequality measurement. For 6 = 1, we obtain
utilitarianism, and as 6 approaches infinity, maximin is obtained in the limit. The
case 6 = 2 yields the social-evaluation ordering corresponding to the Gini index of
inequality [Blackorby and Donaldson (1978)]. All single-parameter Ginis satisfy C, A,
SP and ME.

4. Generalized utilitarianism

A distinguishing feature of generalized-utilitarian social-evaluation orderings is that

they possess an additively separable structure. This separability property is closely
related to several plausible independence conditions which deal with the influence of
the well-being of unconcerned individuals on the social ordering, and those conditions
can, together with some of our earlier axioms, be used to provide characterizations of
generalized utilitarianism.
Ch. II: Utilitarianism and the Theory of Justice 561

Suppose that a social change affects only the utilities of the members of a population
subgroup. Independence of the utilities of unconcerned individuals requires the social
assessment of the change to be independent of the utility levels of people outside the
subgroup.

Independence of the Utilities of Unconcerned Individuals (IUUI):

For all M C {1, ... , n}, for all u, v, u', v' E R n, if ui = vi and u = uv for all
i E M and ui = uj and u = j for allj {1, ... , n} \M, then
uRv X u RV'. (4.1)

In this definition, the individuals in M are the unconcerned individuals - they are
equally well off in u and v and in u' and v'. IUUI requires the ranking of u and v to
depend on the utilities of the concerned individuals - those not in M - only. In terms
of a real-valued representation, this axiom is referred to as complete strict separability
in Blackorby, Primont and Russell (1978). The corresponding separability axiom
for social-evaluation functionals can be found in d'Aspremont and Gevers (1977)
where it is called separability with respect to unconcerned individuals. d'Aspremont
and Gevers' separability axiom is called elimination of (the influence of) indifferent
individuals in Maskin (1978) and Roberts (1980b).
In the case of two individuals, this axiom is implied by strong Pareto. Therefore, its
use is typically restricted to societies with at least three individuals. In that case, we
obtain the following characterization of generalized utilitarianism.
Theorem 5. Suppose that n > 3. A social-evaluation ordering R satisfies C, A, SP
and IUUI if and only if R is a generalized-utilitariansocial-evaluationordering.
Proof: Applying Debreu's (1959, pp. 56-59) representation theorem, continuity
implies that there exists a continuous functionf: RIn - such that, for all u, v E RI",

uRvo : f(u) >f(v). (4.2)

By SP,f is increasing in all arguments, and A implies thatf is symmetric.

IUUI requires that { 1, ... , n} \ M is separable from its complement M for any
choice of M C {1, ... , n}. Gorman's (1968) theorem on overlapping separable sets
of variables [see also Acz61 (1966, p. 312), and Blackorby, Primont and Russell
(1978, p. 127)] implies thatf is additively separable. Therefore, there exist continuous
and increasing functions H: 7 -1 7. and g: 71 -- 7 for all i E {1, ... , n} such that

f(u)= H( g(ui)) (4.3)

562 C. Blackorbv et al.

for all u E R n. Becausef is symmetric, each gi can be chosen to be independent of i,

and we define g = gi for all i { 1, ... , n}. Therefore, because f is a representation
of R,

uRv H g(ui)) > H( g(vi)) X g(ui) > g(vi) (4.4)

i=l i=l i=l i-t

for all u, v G R".

See also Debreu (1960) and Fleming (1952) for variants of this theorem. Due to the
presence of A, IUUI could be weakened by suitably restricting the possible sets of
unconcerned individuals.
An alternative to independence of the utilities of unconcerned individuals is
the population substitution principle. It considers a sequence of social-evaluation
orderings {R } c z++, one for each population size in Z++. Given, for each n 2Z+,
anonymity and the axioms guaranteeing the existence of the representative-utility
functions {},, c z+ (see Theorem 4), the population substitution principle requires
that replacing the utilities of a subgroup of the population with the representative utility
of that subgroup is a matter of indifference [see Blackorby and Donaldson (1984)].

Population Substitution Principle (PSP):

For all n > 3, for all u E1n, for all M C 1, ... , n},

u7 n (EM
1 ((i)ic i)lM,(Uj)jE{ . \sM). (4.5)

As for IUUI, PSP implies that for all n > 3, the representation ?n of R" must
be additively separable and, therefore, it provides an alternative way of characterizing
generalized utilitarianism in the presence of continuity, anonymity and strong Pareto.
The proof of this result is analogous to that of Theorem 5 and is thus omitted.

Theorem 6. A sequence of social-evaluation orderings {R },, z+ satisfies C, A, SP

and PSP if and only if there exists a continuous and increasingfunction g: R -- R
such that, for all n Z_+, Rn is a generalized-utilitariansocial-evaluation ordering
with

uR% > g(ui) > g(vi) (4.6)

i=l i l

for all u, u C Wn.

See Blackorby and Donaldson (1984) for a related result in a variable-population
framework.
Ch. 11: Utilitarianism and the Theory of Justice 563

Generalized utilitarianism can also be characterized in an intertemporal framework.

In that case, a very weak and natural separability axiom can be employed. This
condition, the variable-population version of which was introduced in Blackorby,
Bossert and Donaldson (1995), requires social evaluations to be independent of the
utilities of individuals whose lives are over in both of any two alternatives and who
had the same birth dates, lengths of life, and lifetime utilities in both.
Consider a model where each alternative x X contains (among other features that
may be considered relevant for social evaluation) information about individual birth
dates and lengths of life. We assume that no one can live longer than L E Z++ periods
(L may be arbitrarily large). For i e {1, ... , n} and x C X, let si = S(x) E Z+ be the
period before individual i is born in alternative x, and let i = Li(x) {1, ... , L} be
i's lifetime in x (in periods). Thus, individual i is alive in periods si + 1 to si + li, and
ui = U/(x) is i's lifetime utility in alternative x. Let s = (si, Sr), I = (II, ... , l,)
s..,
and, as before, u = (ul, ... , un).
In order to extend our model to this intertemporal framework, instead of a social-
evaluation ordering R, we employ an ordering R on A = Z+ x {1, ... , L} n x R" and
the objects to be ranked are vectors (s, I, u) of the birth dates, lifetimes, and lifetime
utilities of everyone in society. It is straightforward to reformulate intertemporal
versions of the axioms continuity, anonymity and strong Pareto in this framework.
We use the following definitions, each of which is stated for an arbitrary n E Z++.

Intertemporal Continuity (IC):

For all (s, 1,u) E A, the sets {v R"· (s,, v)R(s, 1, u)} and
{v e TRn I (s,l,u)R(s,l,v)} are closed in 1Z".
IntertemporalAnonymity (IA):
For all (s,l,u) C , for all bijective mappings : {1, ... , n} -- 1... ni,

(s, 1, u)I ((Sz(l) .. . s (,)) ( ... l())

) ,,, ..
() , (,))). (4.7)

Intertemporal Strong Pareto (ISP):

For all (s, , u), (r, k, v) C 4,
(i) if ui = vi for all i C 1, . n}, then (s, l, u)I (r, k, v); and
(ii) if ui > vi for all i {1, ... , n} with at least one strict inequality, then
o
(s, 1,u) P(r, k, v).

At any time, all of the possible alternatives have a common past: the history that
has actually obtained. This means that we can think of history as having a branching
structure. Decisions taken at a particular time have the effect of selecting the branch
along which events will unfold. At any time, some individuals' lives may have ended
in all possible alternatives or their lives may be over in some and not in others.
564 C. Blackorbv et al.

The axiom independence of the utilities of the dead requires that, in any period
t Z++, the relative ranking of any two alternatives is independent of the
utilities of those individuals whose lives are over in t and who had the same
birth dates, lifetimes, and lifetime utilities in both alternatives. To define this axiom
formally, we need more notation. Let, for all (s, , u) E A and all t E Z,
Dt(s,l,u)= {i {1, ... , n} I si+li < t} and Bt(s,l,u)={i E {1,..., n} | si+l < t}.
The individuals in D,(s, 1,u) are those individuals whose lives are over before period t,
and Bt(s, 1,u) contains the individuals who are born before t. We can now define our
intertemporal independence condition.

Independence of the Utilities of the Dead (IUD):

For all (s, 1,u), (r, k, v), (s', 1', u'), (r',k', v') A, for all t e Z++, if
Bt(s, 1, u) = Bt(s', 1', u') = Bt(r, k, v) = Bt(r', k', v')
= D(s, 1,u) = Dt(s', 1', u') = Dt(r, k, v) = Dt(r', k', v') = Mt, (4.8)
(si, li, ui) = (ri, ki, vi) and (s, I', u) = (r',k, vl) for all i E Mt, and
(sj, 1/,u) = (s, lj',uh) and (rj,kj, v) = (rjlkj,vj) for all C {l, ..., n} Mt, then
(s, , u) R (r, k, v) 0 (s', l', u')R (r',k, v'). (4.9)

IUD is a very weak separability condition because it applies to individuals whose lives
are over only, and not to all unconcerned individuals. That is, if all generations overlap,
it does not impose any restrictions. However, when combined with the intertemporal
version of the strong Pareto principle, this axiom has important consequences. In
particular, independence of the utilities of the dead and intertemporal strong Pareto
together imply an intertemporal version of independence of the utilities of unconcerned
individuals, which is defined as follows.

Intertemporal Independence of the Utilities of Unconcerned Individuals (IIUUI)

For all M C {1, .. ., n}, for all (s, , u), (r, k, v), (s', l', u'), (r',k', v') C A, if
ui = vi and u = v; for all i · M and uj = Uj and vj = vj for all
j { 1,... , n} \ M, then
(s, l, u)R (r, k, v) (s', l', u')R (r', k', '). (4.10)

We obtain
Theorem 7. Suppose that n > 3 and an intertemporal social-evaluation ordering R
satisfies ISP R satisfies IUD if and only if R satisfies IIUUI.
Proof: Clearly, IIUUI implies IUD. Now suppose R satisfies ISP and IUD. Let
M C {1, ... , n}, and suppose (s, 1, u), (r, k, v), (s', 1', u'), (r', k', v') E A are such
that u = vi and u = vj for all i M and ui = uj and v = for all
j {1, ... , n}\ M. Let s' = r" = O and I = k' = 1 for all i M, and
Ch. 11: Utilitarianism and the Theory of Justice 565

s? = rJ = 1 and l = kj' = 1 for allj {1, ... , n} \M. By ISP, (s", l", u)I(s, I, u),
(s", 1", u') I (s', 1', u'), (r", k", v) I (r, k, v), and (r", k", v') I (r',k', v'). Therefore,

(s, , u) R (r,k, v) ~ (s", 1", u) R? (r", k", v) (4.11)

and

(s', F, u')R (r',k, v') A> (s , l", u)R (r",k, v'). (4.12)

Furthermore, by definition,

B2 (s", 1", u) = B2 (s", I", u/) = B2 (r",k", v) = B2 (r", k", v')

= D2 (s", 1", u) = D2 (s", 1", u I) = D2 (r", k", v) = D2 (r", k", v') = M2 = M,
(4.13)
and IUD implies

(s",, U) (r", k t , v)
u, X (s", I , u') (r", k", v ). (4.14)

Together with Equations (4.11) and (4.12), this implies that IIUUI is satisfied. C
The conclusion of Theorem 7 remains true if ISP is weakened by requiring part (i),
an intertemporal version of Pareto indifference, only; note that part (ii) of ISP is not
used in the proof.
As an immediate consequence of Theorem 7, a result analogous to Theorem 5 can be
obtained in this intertemporal setting. Thus, the characterization result for generalized-
utilitarian principles is remarkably robust. R is an intertemporal generalized-utilitarian
social-evaluation ordering if and only if there exists a continuous and increasing
function g: - R such that, for all (s, 1,u), (r, k, v) E A,
n n

(s , u)R(r, k, v) e> g(ui) > g(vi). (4.15)

i-I i=l

The following theorem is an immediate consequence of Theorem 7 and a change

in notation that allows us to adapt Theorem 5 to the intertemporal model; the proof is
therefore omitted.
Theorem 8. Suppose that n > 3. An intertemporal social-evaluation ordering R
satisfies IC, I1, ISP and IUD if and only ifR is an intertemporalgeneralized-utilitarian
social-evaluation ordering.
In addition to providing further support for generalized utilitarianism, the results of
the intertemporal model discussed above illustrate an alternative way of obtaining
fully welfarist social-evaluation functionals. Instead of imposing binary independence
of irrelevant alternatives in an atemporal model, a limited version of welfarism that
566 C. Blackorby et al.

includes, in addition to lifetime utilities, birth dates and lengths of life as the available
data, can be used to obtain welfarism by means of the strong Pareto principle alone.
Weakenings of the strong Pareto principle that allow for birth dates or lifetimes to
matter in intertemporal social evaluation are discussed in Blackorby, Bossert and
Donaldson (1997a, 1999b).

5. Utilitarianism

The ethical appeal of generalized utilitarianism rests, in part, on its separability

properties. Utilitarianism is but one possibility within that class of social-evaluation
orderings, and it is appropriate to ask whether it should have special status. The
arguments for utilitarianism that we present in this section are based, for the most
part, on information-invariance properties.
It is easy to verify that all of the information-invariance assumptions introduced
formally in Section 2 are compatible with utilitarianism. In an informational envi-
ronment that allows for cardinal unit comparability at least, the utilitarian social-
evaluation ordering can be employed. This is not the case for generalized utilitarianism:
many generalized-utilitarian orderings do not satisfy information invariance with
respect to cardinally measurable and fully comparable utilities (CFC) or with
respect to translation-scale measurable (TSM) utilities 2 0. The inequality aversion
that generalized utilitarianism permits has, therefore, an informational cost. The
application of generalized utilitarianism is restricted to informational environments
that allow (at least) for the comparability properties described by the set of admissible
transformations in the following theorem.
Theorem 9. Suppose that n > 2 and I contains n-tuples of continuous and increasing
functions only. Generalized utilitarianismwith a continuous and increasingfunction g
satisfies information invariance with respect to (P if and only iffor each 0 CE , there
exist al, ... , a EcR and b E R,+such that

0i() = g-t (a, + bg(r)) (5.1)

for all rT R andfor all i {1, ... , n}.

Proof: That generalized utilitarianism satisfies information invariance with respect
to if Equation (5.1) is satisfied can be verified by substitution.
Now suppose generalized utilitarianism, generated by a function g, satisfies
information invariance with respect to 0, and contains continuous and increasing

20 In addition, some generalized-utilitarian principles fail to satisfy other information-invariance

conditions such as ratio-scale full comparability or translation-scale full comparability which are not
discussed in this chapter. See, for example, Blackorby and Donaldson (1982).
Ch. 11: Utilitarianism and the Theory of Justice 567

transformations only. Information invariance requires that an admissible transformation

¢ = (l, ... , tn) E (P must satisfy the condition
n n n n

g(_ii)) > geg(t0i)) X Cg(ui) > g(ui) (5.2)

i=l i=l i=l i=l

for all u, v E TRn. This is equivalent to the functional equation

g (((ui) = H( g(ui)) (5.3)

i=l i=l

for all u E RZ', where H is increasing. Letting zi = g(ui) and Gi = g o i o g I for all
i E {1, ... , n}, Equation (5.3) can be rewritten as

Z Gi(zi)= H( Zi), (5.4)

i=l i-i

a Pexider equation, and it follows that Gi(r) = ai + bT with b E R,++ and ai R

for all i E {1, ... , n} 2 1 . Substituting back, we obtain 0i(T) = g-' (ai + bg(T)) for all
i {1, ... , n}.
Condition (5.1) says, in effect, that the information environment must support
cardinal unit comparability of transformed utilities (g(ul), ... , g(u,)). It is difficult
to justify such an information environment unless the function g is affine, in which
case we are back to utilitarianism. Therefore, the informational difficulties involved
in applying generalized-utilitarian principles other than utilitarianism suggest that the
utilitarian social-evaluation functional has an important advantage over its competitors
within that class.
The first characterization of utilitarianism we present is new. It does not require
an information-invariance assumption but, as demonstrated below, it can be used to
provide an alternative proof of a characterization result that does. We employ an axiom
that we call incremental equity. In its definition, 1 is the vector x GC7" with xj = 1
and xi = 0 for all i {1, ... , n}\ {j}.
Incremental Equity (IE):
For all u E 1Zn , for all 6 E R, for allj,k {1, ... , n},
(u + 61) ( + n6). (5.5)
IE requires a kind of impartiality with respect to utility increases or decreases. If a
single individual's utility level changes by the amount , IE requires the change to be

21 See AczMl (1966, Chapter 3) for a detailed discussion of Pexider equations and their solutions.
Because the functions qi and g (and thus the inverse of g) are continuous, the domains of H and the Gi
are nondegenerate intervals, which ensures that the requisite functional-equations results apply. Pexider
equations are also discussed in Eichhorn (1978).
568 C. Blackorby et al.

ranked as equally good, no matter who receives the increment [see Hare (1982, p. 26)].
Incremental equity and weak Pareto together characterize utilitarianism.

Theorem 10. A social-evaluation ordering R satisfies WP and IE if and only if is

the utilitariansocial-evaluation ordering.
Proof: That the utilitarian social-evaluation ordering satisfies WP and IE is easily
checked.
If n = 1, WP alone implies the result. Now let n > 2. Applying IE to (u - c1ll),
Equation (5.5) implies that

ul(u- 61i + 61). (5.6)

For any u czR", Equation (5.6) implies

Ui - u -E .Un, + U l- ui

in ii

Using WP, this implies

uRv: X ui > Lvi (5.8)

i=l i=

for all u, v E 7Z. Z

Equation (5.6) in the proof shows that IE requires social indifference about transfers
of utility from one individual to another. Consequently, all distributions of the same
total must be regarded as equally good and utilitarianism results.
There is an interesting link between Theorem 10 and a well-known characterization
of utilitarianism by means of A, WP and information invariance with respect to TSM.
The following theorem illustrates this relationship.

Theorem 11. Ia social-evaluation orderingR satisfies and information invariance

with respect to TSM, then R satisfies IE.
Ch. 11: Utilitarianism and the Theory of Justice 569

Proof: Suppose R satisfies A and information invariance with respect to TSM. By

information invariance with respect to TSM, we have

(U+ 61s R) (u +
+ ) (5 9)

for all u E 1?", for all 6 c 7Z, for all j, k c {1, ... , n}. By anonymity, 61& 61I and,
thus, Equation (5.9) implies

(u+6h)7 (uI+ ),+ (5.10)

which establishes that IE is satisfied. D

As can be verified easily, it is also the case that IE implies A, and WP and E
together imply information invariance with respect to TSM.
By combining Theorems 10 and 11, we obtain an alternative proof of the
above-mentioned characterization result, stated in the following theorem. It is a
strengthening of a result due to d'Aspremont and Gevers (1977) who use the stronger
axiom information invariance with respect to cardinal unit comparability instead
of information invariance with respect to translation-scale measurability [see also
Blackwell and Girshick (1954), Milnor (1954) and Roberts (1980b)].

Theorem 12. A social-evaluation ordering R satisfies A, WP and information

invariance with respect to TSM if and only if R is the utilitarian social-evaluation
ordering.
Proof: That the utilitarian social-evaluation ordering satisfies the required axioms is
easily verified. Conversely, suppose R satisfies A, WP and information invariance with
respect to TSM. By Theorem 11, R satisfies WP and IE. By Theorem 10, X must be
utilitarian. D[
An alternative characterization of utilitarianism can be obtained for the case n > 3
by employing information invariance with respect to CFC and the separability axiom
IUUI together with C, A and SP. This theorem is due to Maskin (1978) - see also
Deschamps and Gevers (1978).

Theorem 13. Suppose that n > 3. A social-evaluationordering R satisfies C, A, SP

IUUI and information invariance with respect to CFC if and only if R is the utilitarian
social-evaluation ordering.
Proof: That utilitarianism satisfies the required axioms is easy to verify. Now suppose a
social-evaluation ordering R satisfies C, A, SP, IUUI and information invariance with
respect to CFC. By Theorem 5, h is generalized utilitarian with a continuous and
increasing function g. Because any increasing affine transformation of g leads to the
same ordering of utility vectors, we can without loss of generality assume that g(O) = 0
570 C. Blackorby et al.

and g(l) = 1. It remains to be shown that, given this normalization, g must be the
identity mapping.
Information invariance with respect to CFC requires

g(a +bu) > g(a+ i) * g(u) > ,g(ui) (5.11)

=1 i-i i- i=1

for all u, v E R", a G R, b C R++. This is equivalent to the functional equation

i=l
g(a + bui) =Ha,b ( i 1
g(ui)) (5.12)

for all u Rn, a 1, b R++, where Ha,b is increasing. Letting zi = g(ui) and
Ga,b(Zi) = g (a + bg-'(zi)), Equation (5.12) can be rewritten as

Gab(Zi) = Hl,b ( E Zi) (5.13)

i=1 i=1

Our continuity and monotonicity assumptions ensure that all solutions to this Pexider
equation are such that Gab(Zi) = A(a, b) + B(a, b)zi. Substituting back, we obtain the
equation
g(a + br) = A(a, b) +B(a, b) g(T) (5.14)
for all a, r E and all b R++, where we use instead of ui for simplicity.
Setting T = 0 and using the normalization g(O) = 0, we obtain A(a, b) = g(a), and
choosing = 1 in Equation (5.14) yields, together with the normalization g(l) = 1,
B(a, b) = g(a + b) - g(a). Therefore, Equation (5.14) is equivalent to
g(a + b ) = g(T) [g(a +b) - g(a)] +g(a) (5.15)
for all a,r C R and all b R+. Setting a =0, we obtain
g(bT) = g(b)g(T) (5.16)
for all T C R and all b E Z++. Analogously, choosing b = 1 in Equation (5.15) yields
g(a + ) = g(r)[g(a + 1) - g(a)] +g(a) (5.17)
for all a, T E R. This is a special case of Equation 3.1.3(3) in Aczel (1966, p. 150)22
and, together with the increasingness of g, it follows that either there exists a c E T,+
such that
c r
e - 1
g(T)= ec
ec - l (5.18)
for all T CER, or g(T) = for all T E 7. Because Equation (5.18) is incompatible
with Equation (5.16), this completes the proof. [

22 To see this, setf(x) = k(x) = g(r) and h(y) = g(a + 1) -g(a) in Equation 3.1.3(3) of Aczel (1966).
Ch. 11: Utilitarianism and the Theory of Justice 571

Continuity plays a crucial role in Theorem 13. Deschamps and Gevers (1978)
examine the consequences of dropping C from the list of axioms in the above
theorem. Among other results, they show that if a social-evaluation ordering R satisfies
A, SP, IUUI and information invariance with respect to CFC, then R must be weakly
utilitarian, leximin or leximax. It is remarkable that these axioms narrow down the class
of possible social-evaluation orderings to that extent. When minimal equity is added,
only weakly utilitarian principles and leximin survive because leximax obviously
violates ME. Therefore, we obtain the following theorem, which is due to Deschamps
and Gevers (1978). Because the proof is very lengthy and involved, we state the
theorem without proving it and refer interested readers to the appendix of their paper.

Theorem 14. Suppose that n > 3. If a social-evaluation ordering A satisfies A, SP

ME, IUUI and information invariance with respect to CFC, then R is the leximin
social-evaluation orderingor a weakly utilitariansocial-evaluation ordering.
It should be noted that the above theorem is not a characterization result because
its statement is an implication rather than an equivalence. The reason is that not all
weakly utilitarian orderings satisfy all the required axioms.

6. Variable-population extensions

Utilitarian and generalized-utilitarian social-evaluation orderings may be extended

to a variable-population framework in different ways. As an example, average and
classical utilitarianism, which use the average and total utility of those alive to rank
alternatives, coincide on fixed-population rankings but may order alternatives with
different population sizes differently. As a full description of the corresponding state
of affairs, an alternative contains, in particular, information regarding the number and
the identities of those who are alive in the state.
X is the set of possible alternatives. For each x X, let N(x) = N denote the
set of individuals alive in x, where N C Z__ is finite and nonempty 23 . Furthermore,
let Z++ be the set of potential people and define X, = {x E X I i C N(x)} to be
the set of all alternatives in which individual i E Z++ is alive. Individual i's utility
function is Ui: Xi - R and a profile of utility functions is U = (U)i Ez. We follow
the standard convention in population ethics and normalize lifetime utilities so that a
lifetime-utility level of zero represents neutrality. A life, taken as a whole, is worth
living for an individual if and only if lifetime utility is above neutrality. Consequently,
a fully informed self-interested and rational person whose lifetime-utility level is below

23 If the empty set were included as a possible population, all results in this section would still be valid.
See, for example, Blackorby, Bossert and Donaldson (1995) for details.
572 C. Blackorby et al.

neutrality would prefer not to have any of his or her experiences 24 . We assume that,
for each non-empty and finite iN C Z++, the set {x X I N(x) = N} contains
at least three elements. This assumption, which is analogous to the fixed-population
assumption that X contains at least three elements, ensures that a variable-population
version of the welfarism theorem is valid. The vector of lifetime utilities of those
alive in alternative x X is (Ui(x))i N() = (Ui)iEN E is the set of all possible
utility profiles (U)i Ez++, which extends the domain U employed in earlier sections to
a variable-population framework.
A variable-population social-evaluation functional is a mapping FE: DE 0,
where DE C IUE is the set of admissible profiles. For all U E DE, the social no-worse-
than relation is RE = FE(U) and I and PEf denote its symmetric and asymmetric
components. As in the fixed-population case, variable-population welfarism is the
consequence of three axioms.
Population UnrestrictedDomain (PUD): D E = E.

PopulationBinary Independence of Irrelevant Alternatives (PBI):

For all x,y c X, for all U, V i DE , if Ui(x) = Vi(x) for all i C N(x) and
U/(y) = V(y) for all i e N(y), then xREy if and only if xREy.

PopulationPareto Indifference (PPI):.

For all x,y C X such that N(x) = N(y), for all U C DE, if Ui(x) = Ui(y)
for all i E N(x), then xIUy.
Note that population Pareto indifference is a fixed-population axiom (it applies
to comparisons of alternatives with the same people alive in each only), whereas
population binary independence of irrelevant alternatives imposes restrictions on the
comparison of alternatives that may involve different populations and population sizes.
PBI requires the social ranking of any pair of alternatives to be the same if two profiles
coincide on the pair.
Results analogous to Theorems 2 and 3 are valid in this variable-population model -
see Blackorby, Bossert and Donaldson (1999a) for details. Because we restrict attention
to anonymous variable-population social-evaluation functionals in this section, we do
not provide formal statements of the corresponding generalizations and, instead, state
a related result that incorporates a variable-population anonymity condition.
Population Anonymity (PA):
For all U, V E DE, for all bijective mappings 3T: Z++, - Z,++such that
Ui = V,() for all i E Z++,
Ru = R. (6.1)

24 It is also true, of course, that such a person would want any change that increases his or her lifetime
utility. See Broome (1993, 1999) for discussions of neutrality and its normalization to zero.
Ch. II: Utilitarianism and the Theory of Justice 573

T
Let 2 = U, E z++ Rn. An ordering RE on Q2is anonymous if and only if the restriction
of hE to 7Z" satisfies A for all n C Z++.
We now obtain the following anonymous variable-population version of the
welfarism theorem. Since the proof of this theorem is analogous to its fixed-population
version, it is omitted. See Blackorby, Bossert and Donaldson (1999a) and Blackorby
and Donaldson (1984) for details.
Theorem 15. Suppose that a variable-population social-evaluationfunctional FE
satisfies PUD. FE satisfies PBI, PPIand PA if and only if there exists an anonymous
ordering *E on Q2 such that, for all x,y e X andfor all U · DE,

E <:R
xRjy ~ (Ui(x))
.E k (Ui(y)) . (6.2)
(6.2)

We call E a variable-population social-evaluation ordering, and we use IE and ~E to

denote its symmetric and asymmetric components.
We say that the ordering RE satisfies the fixed-population axioms continuity, weak
Pareto, strong Pareto and independence of the utilities of unconcerned individuals if
and only if for all n C Z++, the restriction of KE to Rn satisfies the appropriate fixed-
population axiom.
If C and WP are satisfied, population size and representative utility are all the
information that is needed to rank alternatives. No variable-population axioms are
required for this result, which is due to Blackorby and Donaldson (1984). The existence
of a value function representing this ranking is not guaranteed by the fixed-population
axioms alone. For that, continuity must be strengthened to extended continuity.

Extended Continuity (EC):

I
For all n, m Z++and for all u E 7Zn, the sets {v CE7 REu} and
{V C 7Zm I uhEv} are closed in 7Rm.

The following representation theorem is due to Blackorby, Bossert and Donaldson

(2001). In the theorem statement, _n is the representative-utility function for the
restriction of }E to Rn.
Theorem 16. If an anonymous variable-populationsocial-evaluation ordering RE
satisfies EC and WP,then there exists a valuefunction W: Z++ x 7 - , continuous
and increasing in its second argument, such that, for all n, m G Z++, for all u E Rn,
andfor all v e RTm ,

uFv <, W - (U)> (m, (u)). (6.3)

574 C Blackorby et al.

See Blackorby, Bossert and Donaldson (2001) for a proof of Theorem 1625. In
the theorems that follow, extended continuity is not assumed and the existence of
representations is derived from other axioms.
A natural generalization of IUUI requires its separability property to hold for all
comparisons, not only those that involve utility vectors of the same dimension. See
Blackorby, Bossert and Donaldson (1998) for a discussion.

Extended Independence of the Utilities of Unconcerned Individuals (EIUUI):

For all u,v,w Q2, uREv if and only if (u,w)RE (v,w).

The next axiom establishes a link between different population sizes. It requires,
for each possible alternative, the existence of another alternative with an additional
individual alive that is as good as the original alternative, where the individuals alive
in both alternatives are unaffected by the population augmentation. This assumption
rules out social orderings that always declare an alternative with a larger population
better (worse) than an alternative with a smaller population, thereby ensuring nontrivial
trade-offs between population size and well-being.

Expansion Equivalence (EE):

For all u E 2, there exists c e 7t such that (u, c)I E u.

The number c in the definition of EE is a critical level of lifetime utility for u E Q2. If
RE satisfies SP and EE, the critical level for any u cE is unique and can be written
as c = C(u), where C: Q2 - R is a critical-level function.
For most of the results in this section, it is not necessary to impose EE - a weaker
version which requires the existence of a critical level only for at least one u E is
sufficient. Therefore, we define

Weak Expansion Equivalence (WEE):

There exist i E 2 and c E R such that (i, c)IE U.

If a variable-population social-evaluation ordering RE satisfies EIUUI and WEE, then

RE satisfies EE. Moreover, if c is a critical level for , then c is a critical level for all
u CE . See Blackorby, Bossert and Donaldson (1995) for an analogous result in an
intertemporal model.

Theorem 17. If an anonymous variable-population social-evaluation ordering RE

satisfies EIUUI and WEE, then RE satisfies EE and there exists a 7? such that a is
a critical level for all u 6 Q2.

25 Blackorby and Donaldson (1984) falsely claim that the existence of a representation is guaranteed
by C and WP (without EC). See Blackorby, Bossert and Donaldson (2001) for a discussion.
Ch. I]: Utilitarianism and the Theory of Justice 575

Proof: By WEE, there exist i E 2 and c E R such that (u,c)IE i. Let u 2

be arbitrary. By EIUUI, (u, i, c) IE (u, ii), and applying EIUUI again, it follows that
(u, c)I*E u. Letting a = c, the theorem is established.
If SP is added to EIUUI and WEE, Theorem 17 implies that the critical-level
function C is constant 2 6.
There are several possibilities for extending utilitarianism to a variable-population
framework. For example, average utilitarianism is defined as follows. For all n, m
Z++, for all u E ?Rn, for all v · R m ,

In 1 7I
uREV = - E i - Vi. (6.4)
i-1 i=

Average utilitarianism satisfies EE with average utility as the critical level for any
utility vector u E Q2. In addition, average utilitarianism satisfies IUUI but not
EIUUI 27 .
Critical-level utilitarianism [Blackorby, Bossert and Donaldson (1995), Blackorby
and Donaldson (1984)] uses the sum of the differences between individual utility
levels and a fixed critical level of utility as its value function. If a person is added
to a population that is unaffected in terms of utilities, the critical level is that level
of lifetime utility that makes the two alternatives equally good. RE is a critical-level
utilitarian social-evaluation ordering if and only if there exists a E R7 such that, for
all n, m e Z++, for all u e 7Zn, for all v E Rn',

uREV [ui - a] > [vi- a]. (6.5)

i=1 i=1

Critical-level utilitarianism satisfies EE with a constant critical level a and, in addition,

satisfies EIUUI. Classical utilitarianism is a special case of critical-level utilitarianism
with a equal to zero, the level of lifetime utility representing neutrality.
We believe that a positive value (that is, a value above neutrality) should be chosen
for the critical level a. If a is at or below neutrality, the 'repugnant conclusion' results
[Parfit (1976, 1982, 1984)]. A variable-population social-evaluation ordering leads to
the repugnant conclusion if, for any level of utility experienced by each individual in
a society of a given size (no matter how far above neutrality this utility level is) and
for any utility level a above neutrality but arbitrarily close to it, there exists a larger

26 Hammond (1988, 2001) and Dasgupta (1993) argue for constant critical levels and normalize them
to zero. Both set the critical level above neutrality, which implies that the utility level that represents
neutrality is negative.
27 See Blackorby and Donaldson (1984, 1991), Bossert (1990b,c) and Hurka (1982) for discussions of
average utilitarianism.
576 C. Blackorby et al.

population size m so that an alternative in which each of m individuals has a utility

of p is considered better than the former alternative. Therefore, a situation with mass
poverty is superior to an alternative in which every person leads a good life. Because
we consider the repugnant conclusion ethically unacceptable, we recommend a level
of a above neutrality.
Positive critical levels are incompatible with the Pareto plus principle [Sikora
(1978)], which requires the ceteris paribus addition of an individual above neutrality
to a given population to be desirable. Classical utilitarianism satisfies the Pareto plus
principle and leads to the repugnant conclusion. On the other hand, average
utilitarianism does not lead to the repugnant conclusion and it does not satisfy the
Pareto plus principle. Although it performs well in this regard, it possesses the ethically
unattractive property of declaring the ceteris paribus addition of a person below
neutrality desirable if the average utility of the existing population is even lower.
The incompatibility between avoiding the repugnant conclusion and the Pareto plus
principle is not restricted to utilitarian principles, however. More general impossibility
results are reported in Arrhenius (2000), Blackorby, Bossert, Donaldson and Fleurbaey
(1998), Blackorby and Donaldson (1991), Carlson (1998) and Ng (1989).
The variable-population versions of generalized utilitarianism use the same transfor-
mation of utilities g for all population sizes. Moreover, because this function is unique
up to increasing affine transformations, we can, without loss of generality, assume that
g(0) = 0; this ensures that the utility level that represents neutrality is preserved when
applying the transformation.
RE is the average generalized-utilitarian social-evaluation ordering if and only if
there exists a continuous and increasing function g: R - 7 with g(0) = 0 such that,
for all n, m E Z++, for all u E RI', for all v E Rm,

uREv - g(ui) > - Eg(i). (6.6)

i=1 i=1

jE is a critical-level generalized-utilitarian social-evaluation ordering if and only if

there exist a R and a continuous and increasing function g: D-, R satisfying
g(0) = 0 such that, for all n, m Z,+, for all u RZ', for all v R"m,

uREv > > [g(ui)-g(a)] > [g(vi)-g(a)]. (6.7)

i=I i=I

Setting a = 0 yields classical generalized utilitarianism.

As is the case for average utilitarianism, average generalized utilitarianism satisfies
EE with average utility as the critical level and satisfies IUUI but violates EIUUI.
Critical-level generalized utilitarianism satisfies EE with the constant critical level a
and EIUUI. Classical generalized utilitarianism satisfies the Pareto plus principle
and leads to the repugnant conclusion but average generalized utilitarianism and
Ch. 11: Utilitarianism and the Theory of Justice 577

critical-level generalized utilitarianism with a positive critical level avoid the repugnant
conclusion and violate Pareto plus.
The representative-utility function for generalized utilitarianism is given by

n(u) = g-t (- g(ui) (6.8)

for all n E Z++, u e RZ. The value function for average generalized utilitarianism can
be written as

WAGu(n, = g(~), (6.9)

and the value function for critical-level generalized utilitarianism can be written as

WcLGU(n, ) = n [g(~) - g(a)]. (6.10)

All members of the two families satisfy extended continuity and the value functions
are continuous and increasing in their second arguments.
Analogously to Theorem 5, EIUUI can be used to characterize critical-level
generalized utilitarianism in the variable-population case. The following theorem is
due to Blackorby, Bossert and Donaldson (1998).
Theorem 18. An anonymous variable-population social-evaluation ordering RE
satisfies C, S EIUUI and WEE if and only if RE is a critical-level generalized-
utilitariansocial-evaluationordering.
Proof: By Theorem 5, fixed-population comparisons for population sizes n > 3
must be made according to fixed-population generalized utilitarianism with continuous
and increasing functions g. Because each gn is unique up to increasing affine
transformations only, we can without loss of generality assume g"(0) = 0 for all n > 3.
EIUUI requires
n n n n

g "+(ui) i=
Egn+17(vi).¢,
1 ii
)g(u)
> Eg(vu)
=1
(6.11)

for all n,m > 3, for all u,v E Rn, which implies that the g" can be chosen
independently of n, and we define g = gn for all n > 3. By Theorem 17 and SP,
there exists a unique constant critical level a E R. Consider u E RI" and u e RI"' with
n > 3 and, without loss of generality, n > m. Because a is a critical level for all utility
vectors, it follows that

uR
Kv uR (v, aln m)j (6.12)
578 C. Blackorby et al.

Because u and (v, al,, -_,) are of the same dimension n > 3, it follows that

uREv X uR (, al_,)
n In

] g(ui) > A g(v) + (n - m) g(a)

i=l i=l (6.13)
a m1

X A [g(u1 ) - g(a) > [g(vi) - g(a)]

i=l i-1

If n < 3, the definition of a critical level can be used again to conclude uIE (u, a1 3 ,),
and the above argument can be repeated with u replaced by (u, al3-,). D
If the requirement that the repugnant conclusion be avoided is added to the axioms
of Theorem 18, it follows immediately that the critical level must be positive.
As in the fixed-population case, critical-level generalized utilitarianism can be
characterized in an intertemporal model with a variable-population version of
independence of the utilities of the dead. This extended version of IUD is obtained
from IUD in the same way EIUUI is obtained from IUUI; see Blackorby, Bossert and
Donaldson (1995) for details.
Alternative intertemporal consistency conditions are explored in Blackorby, Bossert
and Donaldson (1996). In the intertemporal setting, individuals are assumed to
experience utilities in each period which aggregate into lifetime utilities. Forward-
looking consistency requires that, in any period, future utilities are separable from past
utilities. In Blackorby, Bossert and Donaldson (1996), it is shown that consistency
between forward-looking social evaluations and intertemporal social evaluations
implies, together with some other axioms, classical generalized utilitarianism and
thus the repugnant conclusion. The same results are obtained for a full intertemporal
consistency requirement which is stronger than forward-looking consistency by itself
but is equivalent to it in the presence of other axioms. The consequences of weakening
the intertemporal strong Pareto principle are examined in Blackorby, Bossert and
Donaldson (1997a,b), where versions of critical-level generalized utilitarianism and
classical generalized utilitarianism that allow for discounting are characterized.
We conclude this section with a discussion of information-invariance assumptions in
the variable-population framework. Let E denote an equivalence relation defined on
the domain DE of a variable-population social-evaluation functional FE. Information
invariance with respect to ~E is defined as follows:

Information Invariance with Respect to -:

For all U, V DE , if U E V, then RE = RE.

As in the fixed-population case, one possible way to define an information

assumption is to specify a set of admissible vectors of utility transformations. In the
variable-population case, the elements of such a set P can be written as 0 = (i)i E Z_ ,
Ch. 11. Utilitarianism and the Theory of Justice 579

where each i is a function Ai: -i R that transforms individual i's utility

ui into i(ui).
The following information assumptions are used in this section.

OrdinalFull Comparability (OFC):

0 E ( if and only if there exists an increasing function 0o: 7R -- 7Z such that
gi = o0 for all i E Z++.

CardinalMeasurability (CM):
0 E if and only if there exist ai E R and bi C 7Z++ for each i E Z++ such
that gi(T) = ai + bir for all T C R7 and all i E Z++.

CardinalFull Comparability (CFC):

0 E if and only if there exist a E7Z and b E R,++ such that i(r) = a + br
for all E 7Z and all i E Z++.

Numerical Full Comparability (NFC):

0 e if and only if Oi(r) = r for all Tr R and all i E Z++.

In the presence of welfarism, information invariance can alternatively be defined in

terms of RE. For information-invariance assumptions defined with sets of admissible
transformations, we obtain the following definition:

Information Invariance with Respect to 0P:

For all 0 E P, for all n, m e Z++, for all u, u' R", for all v, v' E Zm,
if u' = gi(ui) for all i {1, ... , n} and v = i(v) for all i c {1, .. ., m}, then
uREv > u'uREv' (6.14)

In the variable-population framework, information assumptions are considerably

more restrictive than in the fixed-population case. In the presence of anonymity and the
weak Pareto principle, for example, the possibility of level comparisons is necessary for
the existence of a variable-population social-evaluation functional. This observation,
which is stated in the following theorem, implies that stronger information-invariance
requirements than information invariance with respect to OFC cannot be satisfied. We
use POFC to denote the set of admissible vectors of utility transformations according
to ordinal full comparability.

Theorem 19. If an anonymous variable-population social-evaluation ordering RE

satisfies WP and information invariance with respect to c4, then 4P C PoFC.

Proof: Suppose RE is anonymous and satisfies WP and information invariance with

respect to P. By way of contradiction, suppose that P \ POFC 0. Then there
580 C Blackorby et at.

exist e , y R, and i,j e Z+, such that ¢j(y) • Oj(y). Consider the one-
dimensional utility vectors (us), (uj) E Q such that ui = uj y=. Because .RE is reflexive,
E
we must have (ui) (uj). By information invariance with respect to 4P, we obtain
(i(ui))IE ((uj)). Let vi = j(u,). This implies (¢i(i))7E(vi). Because Oi(ui) vi,
this contradicts weak Pareto. D
Some information-invariance assumptions impose significant restrictions on ethical
parameters such as critical levels, which clearly is undesirable. It can be shown
that information invariance with respect to CFC leads to average utilitarianism (for
comparisons involving at least three individuals) in the presence of some other axioms
including IUUI, and if IUUI is strengthened to EIUUI, an impossibility result is
obtained. These results are proved and discussed in Blackorby, Bossert and Donaldson
(1999a).
Because of these negative observations, we suggest an alternative way of formulating
information invariance in a variable-population framework. The fundamental difficulty
appears to be that the standard welfarist framework with an unrestricted domain is
inadequate to define norms, such as the utility level associated with a neutral life,
which permit interpersonal comparisons of utility at a single utility level. Therefore, we
suggest the use of a systematic procedure for incorporating such norms [see Blackorby,
Bossert and Donaldson (1999a)]. In particular, we propose the use of norms to restrict
the domain of admissible utility profiles2 8. In addition to avoiding the difficulties
associated with extending the traditional taxonomy of comparability and measurability
assumptions to a variable-population framework, we think that this approach using
norms is more intuitive.
For U C UE and i E Z++, let ]i(U,) denote the level of utility individual i assigns to
a neutral life, given the utility function Ui. Suppose, in addition, that a second norm
denotes a life above neutrality at some satisfactory or 'excellent' level (not necessarily
a critical level). It is possible, given these norms, to represent the value of a neutral
life with a utility level of zero and the value of an excellent life with a utility level of
one. Letting ei(Ui) denote the utility level representing an excellent life according to
i's utility function Ui, the restricted domain that respects both normalizations is given
by

DE = Uld= U E UE j i(Ui) = 0 and r(Ui) = 1 Vi E Z++}. (6.15)

These normalizations allow us to start with very demanding information-invariance

assumptions on the unrestricted domain UE and yet have remarkable flexibility
in designing social-choice rules if we restrict attention to the profiles respecting
our normalizations. We obtain the following theorem [see Blackorby, Bossert and
Donaldson (1999a)]:

28 See Tungodden (1999) for a discussion of a single norm in combination with ordinally measurable
utilities. For a different approach using normalized utilities, see Dhillon (1998).
Ch. 11: Utilitarianismand the Theory of Justice 581

Theorem 20. If a variable-population social-evaluation functional FE satisfies

information invariance with respect to CM and the utility levels representinga neutral
life and an excellent life are normalized to zero and one respectively, then the
restriction of FE to DE = E satisfies information invariance with respect to NFC.
Proof: Suppose FE satisfies information invariance with respect to CM. Let U C ZF.
By definition, ?ri(Ui) = 0 and Ei(Ui) = 1, and it follows that i(0) = 0 and ¢i(l) = 1 for
all i C Z+. Consequently, ai = 0 and bi = 1 for all i E Z++. 1
Note that only cardinal measurability is required in Theorem 20; full interpersonal
comparability is provided by the two norms. Thus, the theorem shows that, if utilities
are cardinally measurable and two norms are employed, utilities on the resulting
restricted domain are numerically measurable and fully interpersonally comparable.
Therefore, the norms generate sufficient additional information to apply any social-
evaluation functional. Similar results involving a single norm can be found in
Blackorby, Bossert and Donaldson (1999a).

7. Uncertainty

Suppose that a government or an individual must take an action, which might be a

simple one or a more complex one that leads, for example, to the establishment of an
institution, custom or moral rule, from a set of feasible actions. If the agent knows with
certainty the alternative that results from each action, a social-evaluation functional can
be used to rank them. In that case, a functionf maps actions into alternatives and, for
any two actions a and b, action a is at least as good as action b if and only iff (a) is
socially no worse thanf(b)2 9 .
In most cases, the consequences of actions are not known with certainty at the
time a choice of action has to be made. It may be possible, however, to attach
probabilities to the outcomes that may materialize and, in that case, actions can
be ranked by ranking prospects. Prospects can be identified with vectors of social
alternatives if probabilities are fixed. For such an approach to make normative sense,
probabilities may be subjective but must be based on the best information available at
the time decisions are taken. If probabilities represent uninformed individual beliefs,
the normative force of this approach is weakened substantially.
In any normative investigation, rationality plays an important role and this
suggests that both social and individual preferences should satisfy the expected-utility
hypothesis [von Neumann and Morgenstern (1944, 1947)]. Given that, two different
versions of welfarism are possible. Ex-ante welfarism bases social evaluations of
prospects on individual valuations while ex-post welfarism orders alternatives after the

29 See Broome (1991b) for a discussion.

582 C. Blackorby et al.

uncertainty has been resolved and aggregates these judgements into a social ordering
of prospects.
Harsanyi (1955, 1977) investigates ex-ante welfarism and shows that it has
surprising consequences for social evaluation. In his formulation, individuals have
ex-ante utility functions that satisfy the Bernoulli hypothesis [Broome (1991a)], a
condition that is stronger than the expected-utility hypothesis (see the discussion
below) 30 . The Bernoulli hypothesis requires that individual ex-ante utilities are equal
to the expected value of von Neumann-Morgenstern (vNM) utilities. There are in > 2
'states of nature' with probabilities p = (pa..., p,) E R", Ijpj = 1, and they
are agreed upon by individuals and by the social evaluator. Individual ex-ante utilities
are given by

U; = puI = Ep Ui(x) (7.1)

j=- j 1

for all i {1, ... , n}, where u is individual i's utility in state j and xi is the social
alternative that occurs in state j. There is a single profile of utility functions and
x j is fixed for allj E 1, ... , m}. Consequently, the utility level u is fixed for all
i C , .. ., n} and all j E 1, ... , m}. Each probability vector p is called a lottery
and all lotteries p C R4+ with En" I pj = 1 are permitted. Social ex-ante preferences
are represented by

uo = piUo(xJ) = (7.2)
j=1 j=1

where u' is social utility in statej and UO: X - R is a social utility function. Harsanyi
requires social preferences over lotteries to satisfy ex-ante Pareto indifference, which
requires society to rank any two lotteries as equally good whenever they are equally
valuable for each individual. This axiom alone has the consequence that there exist
y E 1Z and 6 C R such that social utilities are weighted sums of individual utilities
with u = = I yiu{ + 6 for all E {1, ..., m} and u = 74= yiu/' + . If p and q
are any two lotteries, then

p q y pju >E y qj,, (7.3)

i-= j= i-= j=1
where p >- q means that p is socially at least as good as q. This result is
called Harsanyi's (1955) social-aggregation theorem 3t . The weights (yl, ... , y,) in

30 Arrow (1964) provides an account of the expected-utility hypothesis that is consistent with our
approach.
31 Border (1981) presents an elegant proof which is reproduced in expanded form in Weymark (1994).
See also Blackorby, Bossert and Donaldson (1998), Blackorby, Donaldson and Weymark (1999, 2001),
Broome (1990, 1991a), Coulhon and Mongin (1989), Domotor (1979), Fishburn (1984), Hammond
(1981, 1983), Mongin (1994, 1995, 1998) and Mongin and d'Aspremont (1998).
Ch. 11: Utilitarianism and the Theory of Justice 583

Equation (7.3) are, in general, not unique and need not be positive. The imposition
of stronger Pareto conditions implies some restrictions on their signs, however. If
strong Pareto is satisfied, Equation (7.3) can be satisfied with positive weights 3 2 and
if weak Pareto is satisfied, Equation (7.3) can be satisfied with non-negative weights,
at least one of which is positive 33.
Because the above-described model employs a single profile of utility functions and,
thus, differs from the approach of this survey, we do not include a proof. Instead, we
present a variant of Harsanyi's theorem that uses the basic model employed in the rest
of this chapter. It is a multi-profile model which allows for interpersonal comparisons
of utilities and permits the application of the anonymity axiom 34 . X is a set of
alternatives with at least four elements. A prospect is a vector x = (x', ... , xm ) C X "
with m > 2, and the prospect xc = (x, ... , x) E Xm is one in which x C X occurs for
certain. The vector of positive probabilities is fixed at p = (pl, Pm) E Zm with
~mY Ipj = 135
As in Harsanyi (1955, 1977), we assume that individual utilities satisfy the Bernoulli
hypothesis with the ex-ante utility function UiA: X" - R1 given by
m
UA(x) = EU(x) = pjiUi(xi) (7.4)
j=1

for all x E X". Ui(x) is the value of the prospect x to person i, U: X -* R is

individual i's vNM utility function and EUi(x) is i's expected utility for prospect x.
Equation (7.4) implies that, for all x C X, UiA(x) = EUi(xc) = U(x). A profile of
ex-ante utility functions is UA = (U, . . ., U,,), and a profile of vNM utility functions
is U = (Ul, ... , U,). Writing UA(x) = (UI(x), ... , U,A(x)) = EU(x) = (EUi(x),....
EU,(x)) for all x E X' and U(x) = (Ul(x), ... , U,(x)) for all x e X,

UA(x) = EU(x) = Zpj U(xj) (7.5)

j=-

for all x C X m . The functions U 1, ... , U, do double duty in this formulation: they are
the individuals' vNM utility functions and they measure individual well-being.
An ex-ante social-evaluation functional F': DA 4 OA is a function which maps
.

each profile of ex-ante utility functions into an ordering on xm. We say that the
domain DA is the Bernoulli domain DA if and only if it consists of all profiles of

32 See Domotor (1979), De Meyer and Mongin (1995), Weymark (1993) and Zhou (1997).
33 See Weymark (1993, 1994, 1995) and Zhou (1997).
34 Multi-profile models are presented in Blackorby, Donaldson and Weymark (2001), Hammond (1981,
1983), Mongin (1994) and Mongin and d'Aspremont (1998).
35 Because probabilities are assumed to be fixed, any state of nature with a probability of zero may be
dropped.
584 C Blackorby et al.

ex-ante utility functions UA = (U, ... , U,) such that Equation (7.5) is satisfied for
some U=(U1,..., U,) E .
FA is (ex-ante) welfarist if and only if there exists an ordering RA on R" such that
xRy X (UA(x), . UA(x)) RA (Us).U(y UA )) (7.6)

for all x,y E Xm, where RA = F(UA) is the social ordering of prospects. On
the Bernoulli domain, ex-ante welfarism is a consequence of the assumptions binary
independence of irrelevant alternatives and Pareto indifference (applied to ex-ante
utilities) 36
Social preferences satisfy the expected-utility hypothesis if and only if there exists
a function UO: X x DA -- R such that, for all x, y C X,
m
xRUy E pjUo(Xj, UA) > piUo(yj, UA) (7.7)
j=l j=l
Note that this is somewhat weaker than Equation (7.4) because there is no need to
measure a social ex-ante utility level in this case. The social vNM function is allowed
to be profile-dependent. In our multi-profile setting, if the social vNM utility function
were written without UA, an imposed social ranking would result. In Harsanyi's lottery
problem, there is only a single profile of vNM utility functions and the second argument
of U is not needed. If the value of the social-evaluation functional satisfies the
expected-utility hypothesis for every UA C DA we say that the range of the functional
is OAu
)A
Now suppose that the domain of the welfarist social-evaluation functional is DB, so
that individual utilities satisfy the Bernoulli hypothesis, and its range is OCA, so that
social preferences satisfy the expected-utility hypothesis. Thus, we consider a social-
evaluation functional FBEU: DA OEAU. Then it must be true that, for all x,y e X"',
m m
xRUy X ZpjUo(xi, UA) > pjUo(yj, UA)
j=1 j=

X (EUI(x),.. EU,(x)) A (EU y), . EU,(y)) (7.8)

X EU(x)KA EU(y),
where RA = FBAEu(UA). Setting x = xc and y =y, in Equation (7.8) results in
xR yc Uo(x, UA) > Uo(y, UA)

(U(x), ... , U,(x)) RA (Ui(y), U,(y)) (7.9)

A
U(x)* U(y).

36 See Blackorby, Donaldson and Weymark (2001), Mongin (1994) and Mongin and d'Aspremont
(1998).
Ch. 11: Utilitarianismand the Theory of Justice 585

This implies that there is a single social-evaluation ordering which is the same for
all states, and it is the ex-ante social-evaluation ordering RA. The ordering RA orders
prospects and it also orders alternatives once the uncertainty has been resolved. Such
a social-evaluation functional is both ex-ante and ex-post welfarist.
Next, we prove a theorem that shows that any welfarist ex-ante social-evaluation
functional on the Bernoulli domain whose social preferences satisfy the expected-
utility hypothesis must possess a property that is equivalent to the requirement
that RA satisfy information invariance with respect to translation-scale measurabil-
ity (TSM)37. In order to find the largest class of functions satisfying our axioms, no
information-invariance restriction is placed on the social-evaluation functional.
Theorem 21. Suppose that XI > 4. If an ex-ante social-evaluation functional
FBAEU: DB OEu is welfarist, then, for all u, v, a C R',

uR v (u + a) RA (v + a). (7.10)

Proof: For any u,v,a E R7, choose a profile UA e DA such that there exist
x,y,z,w E X with U(x) = u/pl, U(y) = v/pl, U(z) = 1O and U(w) = a/Zm 2pj
where U is the vNM profile corresponding to UA. Consider x,y E Xm with x' = x,
yI = y and x = y = z for all j E 2, ... , m}. Because EU(x) = u and EU(y) = v,
Equation (7.8) implies
m

uAv p Uo(x, UA ) + pj Uo(Z, UA) >plUo(y, UA) + pUo( z, UA)

j=2 j=2

<pl Uo(x, UA)>pi Uo(Y, UA).

(7.11)
Now consider w, z E X m with wl = x, zl = y and w = z = w for allj E {2, ... , m}.
Because EU(w) = (u + a) and EU(z) = (v + a), Equation (7.8) implies
m m

(u+a)RA(v+a) p Uo(x,UA) + pjUo(w,UA) >pUo(yUA) +pjU(w,UA)

j=2 j=2
plIUo(x,UA) > Uo (y, UA).
(7.12)
Because the second lines of Equations (7.11) and (7.12) are identical, Equation (7.10)
is immediate. l
The property described by Equation (7.10) is the same as information invariance
with respect to translation-scale measurability, and we use the result of Theorem 12
to show that, given anonymity and weak Pareto, RA must be the utilitarian ordering.

37 Mongin (1994) and Mongin and d'Aspremont (1998) prove a similar theorem for lotteries.
586 C. Blackorby et al.

Theorem 22. Suppose that XI > 4. An ex-ante social-evaluation functional

FJE: DA - OEU is welfarist and satisfies A and WP if and only iffor all u, v C R'",
n I

uRn v X E U: > E vi (7.13)

and, for all x,y E X m and all profiles UA e D,

1 n

xRy EEU,(x) > EEU(y)

i-1 i-=
n M n (7.14)

i=1 jI= i=i =

Proof: Necessity follows from Theorems 12 and 21. Sufficiency is immediate.

Note that continuity of RA is not needed in Theorem 22. A variant can be proved
by adding continuity and dropping anonymity. In that case, Equation (7.13) becomes
n n
uRnv y/iui > YiVi, (7.15)
i=l i-

where y E R {01,}. The social-evaluation ordering RA is weighted utilitarian and

the weights are non-negative with at least one that is positive. Because anonymity is
such an important axiom in welfarist social ethics, however, we have presented the
theorem that uses it.
An objection that is sometimes made to Harsanyi's theorem is that vNM utility
functions are not unique (increasing affine transformations represent the same
preferences) and, thus, equal weights on utilities are meaningless. Our framework does
not suffer from this difficulty. Because of our assumptions regarding the measurability
and comparability of individual utilities, a particular vNM function is selected for
each person. The result of the theorem implies that the information structure (for both
vNM utility functions and the ex-ante utility functions (UI, ... , U,A)) must support
cardinal unit comparability.
An interesting question of interpretation is whether Theorem 22 provides a
convincing argument for utilitarianism. If the answer is 'yes', it should be noted that the
theorem requires the Bernoulli hypothesis to be satisfied, a stronger requirement than
the expected-utility hypothesis. Thus, the utility functions (U 1, ... , U,) must represent
people's good, free of the irrationalities of compulsive gambling, for example. It is
not usual to present arguments in favour of the Bernoulli hypothesis, over and above
the requirements of the expected-utility hypothesis, but it has been done by Broome
(1991a).
Ch. II: Utilitarianism and the Theory of Justice 587

The theorem requires individual and social probabilities to coincide. If, however,
individual probabilities are subjective and can differ across individuals, impossibility
theorems emerge 38. The same-probability requirement is a demanding one but it might
be justified by regarding probabilities as 'best-information' probabilities.
Suppose that, instead of the Bernoulli hypothesis, individual ex-ante utilities satisfy
the expected-utility hypothesis. In that case, writing UTNM as individual i's vNM utility
function,
m m
UiA(x) > UiA(y) pjUi M (xi) > pjUNM (yi) (7.16)
j=l j=1

for all x,y Xm 1 and for all i

{1, ... , n}. It follows that there exist increasing
functions h1, ... , h, with hi: R 1- R for all i e {1, ... , n} such that, for all x E X m
and for all i c {1, ... , n},

Ui(x) = hi (EpU' (x)). (7.17)

Ifx = xc in Equation (7.17), UIA(x<) = hi(UNM(x)) and this utility level can be regarded
as person i's actual utility level when alternative x is realized. Writing Ui as i's actual
utility function, Ui(x) hi(UVM(x)) for all x E X, and Equation (7.17) becomes

UiA(x) = hi( pJUNM (xi)) = hi (ph (Ui(xi))). (7.18)

j=1 j=1

The domain of an ex-ante social-evaluation functional is the expected-utility

domain DEA if and only if for each i e {1, ... , n}, Equation (7.18) is satisfied for
some vNM utility function UNM and some increasing function hi. On such a domain,
no welfarist ex-ante social-evaluation functional exists.
Theorem 23. Suppose that IXI > 4. There exists no ex-ante social-evaluation
functional FEAU: DEU - 0 EA that is welfarist and satisfies A and WP
Proof: First consider the subdomain DB of DEU and assume that FEAU is welfarist and
satisfies A and WP Theorem 22 implies that, for all u, v C Rn

n n

uRAv X ui > vi. (7.19)

i-I i=I

38 See Hammond (1981, 1983) and Mongin (1995).

588 C. Blackorby et al.

Now consider the subdomain DA of DU in which, for all i {1, ... ,n},

UiA(x) = h ( pjU (xi)) (7.20)

j=l

where h: 1Z - IZ is increasing but not affine. Define the ordering RA on R" by

URh Uv (h(ul), ... , h(u,,)) R (h(v),..., h(v,)) (7.21)

for all u,vu R". Writing EU'vM(x) = I'

jpZU,'VM(xj) for all x E X' 1 and for all
i {1, ... , n}, we know that

UA(X A UA(y)

(h(EU (x)),
... , h(EUvM(x))) A. (h(EUiM(y)), h(EU<M'(y)))

(EUVM(x)..., EUfM(x)) R (EUi~M(y) ., EU, (y)).

(7.22)
Because h is the same for each individual, .Rh is anonymous and Theorem 22 implies
that, for all u, v E R n,

* n
uRh Ui > Vi. (7.23)
i=l i=l

Because h is increasing, Equation (7.21) can be rewritten as

uRAv (h l(ul), ... ,

A (h l(vL), ... , h-l(v,)),
h-'(u,))Rh (7.24)

and Equations (7.19), (7.23) and (7.24) together imply

uAv ZUi
' > vi , h(ui) > I h (vi) (7.25)
i=l i=l i-l i=l

for all u, v E Rn. Equation (7.25) can be satisfied if and only if h-1 is affine. This
requires h to be affine, and a contradiction results. []
The proof of Theorem 23 shows that, when the transforms hi, ... , h, in Equa-
tion (7.18) are identical, prospects must be ranked by using the sums of expected
utilities. This means, however, that the social-evaluation ordering .RA must depend
on h, and welfarism, which requires a single social-evaluation ordering, is contradicted.
A variant of Theorem 23 can be proved by dropping anonymity and adding
Ch. II: Utilitarianismand the Theory of Justice 589

continuity 3 9. A way out of this impossibility result can be provided by restricting the
domain of the social-evaluation functional to a single utility profile or information
set. If the expected-utility hypothesis is satisfied, Blackorby, Donaldson and Weymark
(2001) prove that, on a single information set, A must be generalized utilitarian if
the Bernoulli hypothesis is not satisfied and utilitarian if it is satisfied.
One possible escape from Harsanyi's theorem is to embrace ex-ante welfarism
without requiring social preferences to satisfy the expected-utility hypothesis, a move
that has been suggested by Diamond (1967) and Sen (1976, 1977b, 1986). They argue
that social preferences that satisfy the expected-utility hypothesis cannot take account
of the fairness of procedures by which outcomes are generated [see also Weymark
(1991)]. An ex-ante social-evaluation functional of the type they advocate is given
by

xRy : - A (UA(x)) > A (UA(y)) (7.26)

for all x,y C X m and all profiles UA C DA. In the equation, 7A: Rn - R is a strictly
concave and, therefore, inequality-averse, ex-ante representative-utility function. If the
social-evaluation functional satisfies anonymity, =A must be symmetric. See Weymark
(1991) for a discussion.
An important question to consider, however, is whether it is appropriate to require
ex-ante welfarism. This form of welfarism is not applied to actual well-being, and that
suggests that ex-post welfarism may be more appropriate and ethically more basic. It is
true of course that, given the Bernoulli hypothesis, ex-ante welfarism implies ex-post,
but the converse is not true. A second way out of the result of Theorem 22, therefore, is
provided by requiring ex-post welfarism only. Suppose, for example, that P: RZn - R
is an ex-post representative-utility function, which expresses a social attitude toward
utility inequality. Then ex-post welfarism is satisfied by a principle given by

xRpuy EpjUP ( P(U(x))) > pjU(OP

( (U(yj))) (7.27)
j=1 j=1

for all x,y c Xm and all profiles U e D. In Equation (7.27), UP is a social vNM utility
function which expresses a social attitude toward representative-utility uncertainty.
Even if society is neutral toward such uncertainty (Us' is affine), such a principle
is not consistent, in general, with ex-ante Pareto indifference if individual ex-ante
utilities satisfy the expected-utility hypothesis. This means that x may be regarded as
better than y even though the same standard of rationality that is used socially ranks
prospect y as better for each person. With such a principle, therefore, social rationality
trumps individual rationality.

39 See also Blackorby, Donaldson and Weymark (1999, 2001), Roemer (1996), Sen (1976) and Weymark
(1991).
590 C. Blackorby et Ul.

As an example of this last claim, let there be two individuals, two states with
equal probabilities, let U be the identity map, and let 'EP(ul,U2 ) = 4u() + iU(2). In
prospect x, each individual's utility level is 20 in both states, so P(U(xL)) = -P(U(x2))
= 20 and expected social value is 20. In prospect y, utilities are (ul, u2 ) = (40, 4) in
state 1 and (4,40) in state 2. Consequently, 'P(U(yl))=E'(U(y2 )) = 13, expected
social value is 13, and society ranks x as better than y. Each individual's expected
utility is 20 in x and 22 in y, however, and each is better off, ex-ante, in y.

8. Conclusion

The idea that a just society is a good society can be an attractive one if the good
receives an adequate account. Welfarist social-evaluation functionals are capable of
performing well as long as the notion of well-being that they employ captures
everything of value to individual people. Given that, principles for social evaluation
that are non-welfarist run the risk of recommending some social changes from which
no one benefits. This is the lesson of Theorems 1 and 3.
If a welfarist principle is to be used to rank alternatives that are complete histories
of the world, a single social-evaluation ordering is sufficient to do the job for every
profile of utility functions. Although such orderings can be used to rank changes
which affect a population subgroup, such as the citizens of a single country or the
people in a particular generation, the induced ordering over their utilities is not, in
general, independent of the utilities of others. Independence is guaranteed by the axiom
independence of the utilities of unconcerned individuals and, in conjunction with
continuity, anonymity and strong Pareto, that axiom leads to generalized utilitarianism.
In a dynamic framework, the same result is the consequence of independence of the
utilities of the dead together with intertemporal versions of continuity, anonymity and
strong Pareto.
Generalized-utilitarian social-evaluation functionals are ethically attractive, but some
of them may require utility information that is difficult to acquire. In parsimonious
information environments, utilitarianism itself may prove to be more attractive than
the other members of that family of orderings. The only generalized-utilitarian social-
evaluation ordering that satisfies information invariance with respect to cardinal
full comparability is utilitarianism. And if individual utilities are translation-scale
measurable, anonymity and weak Pareto alone imply that the social-welfare ordering
must be utilitarian. Information restrictions are not the only axioms that generate
utilitarianism, however. Incremental equity is an axiom that requires a kind of
impartiality with respect to utility increases or decreases. If one person's utility
increases or decreases, the axiom requires the change to be equally good no matter
who the affected person is. This axiom, together with weak Pareto, characterizes
utilitarianism.
The utilitarian and generalized-utilitarian social-evaluation functionals can be
extended to environments in which population size and composition may vary across
Ch. 11: Utilitarianismand the Theory of Justice 591

alternatives. Two properties can be considered particularly desirable in this framework:

extended independence of the utilities of unconcerned individuals and avoidance of
the repugnant conclusion. Given continuity, anonymity and strong Pareto, extended
independence of the utilities of unconcerned individuals implies that the social-
evaluation ordering must be critical-level generalized utilitarian. The critical level is a
parameter that represents the smallest utility level above which additions to a utility-
unaffected population have value. The repugnant conclusion is avoided if and only if
the critical level is above neutrality.
If utilities are cardinally measurable, interpersonal comparisons at any two norms
are sufficient to produce numerical full comparability. We might choose norms, for
example, at neutrality and at a utility level that represents a satisfactory or excellent
life. Utility numbers such as zero and one may be chosen for these, and NFC results. It
follows that cardinal measurability and two norms are sufficient to employ any welfarist
social-evaluation functional.
Social-evaluation functionals can be extended to rank prospects as long as
probabilities can be attached to the various states of nature. If individual ex-ante
utilities satisfy the Bernoulli hypothesis, social preferences satisfy the expected-utility
hypothesis and all subjective probabilities coincide, the only ex-ante social-evaluation
functional that satisfies anonymity and (ex-ante) weak Pareto is the utilitarian one. If,
however, the domain is expanded to include all individual ex-ante utility functions that
satisfy the expected-utility hypothesis, or if subjective probabilities can be different
for different people, an impossibility results. If the von Neumann-Morgenstern utility
functions, in addition to representing people's good, express an attitude toward
uncertainty that is rational and has some normative standing, it can be argued that
the Bernoulli hypothesis is a reasonable assumption. In that case, Harsanyi's social-
aggregation theorem provides support for utilitarianism.
Together, these results make a strong case for utilitarian and generalized-utilitarian
social evaluation. When these social-evaluation functionals are coupled with an
adequate account of lifetime well-being, the resulting principles are ethically attractive
and perform well in environments in which other principles perform poorly. Social-
contract theories, for example, are not able to give an adequate account of justice
between generations when the existence of people in one generation is contingent on
decisions made by another. On the other hand, the critical-level generalized-utilitarian
principles can cope with fully dynamic environments in which history has a branching
structure and the identities of those alive, their numbers, quality of life and length of
life can vary across alternatives.

References

Acz6l, J. (1966), Lectures on Functional Equations and Their Applications (Academic Press, New York).
Anderson, E. (1999), "What is the point of equality?", Ethics 109:287-337.
Arneson, R. (1989), "Equality of opportunity for welfare", Philosophical Studies 56:77-93.
592 C. Blackorby et al.

Arneson, R. (2000), "Luck egalitarianism and prioritarianism", Ethics 110:339-349.

Arrhenius, G. (2000), "An impossibility theorem for welfarist axiologies", Economics and Philosophy
16:247-266.
Arrow, K.J. (1951), Social Choice and Individual Values, 1st Edition (Wiley, New York).
Arrow, K.J. (1963), Social Choice and Individual Values, 2nd Edition (Wiley, New York).
Arrow, K.J. (1964), "The role of securities in the optimal allocation of risk-bearing", Review of Economic
Studies 31:91-96.
Atkinson, A.B. (1970), "On the measurement of inequality", Journal of Economic Theory 2:244-263.
Basu, K. (1983), "Cardinal utility, utilitarianism, and a class of invariance axioms in welfare analysis",
Journal of Mathematical Economics 12:193-206.
Bentham, J. (1789), An Introduction to the Principles of Morals and Legislation (Payne, London).
Reprinted in 1973 in: The Utilitarians (Anchor, Garden City) pp. 5-398.
Blackorby, C., and D. Donaldson (1978), "Measures of relative equality and their meaning in terms of
social welfare", Journal of Economic Theory 18:59-80.
Blackorby, C., and D. Donaldson (1982), "Ratio-scale and translation-scale full interpersonal
comparability without domain restrictions: admissible social evaluation functions", International
Economic Review 23:249-268.
Blackorby, C., and D. Donaldson (1984), "Social criteria for evaluating population change", Journal of
Public Economics 25:13-33.
Blackorby, C., and D. Donaldson (1991), "Normative population theory: a comment", Social Choice and
Welfare 8:261-267.
Blackorby, C., and D. Donaldson (1992), "Pigs and guinea pigs: a note on the ethics of animal
exploitation", Economic Journal 102:1345-1369.
Blackorby, C., D. Primont and R. Russell (1978), Duality, Separability, and Functional Structure: Theory
and Economic Applications (North-Holland, Amsterdam).
Blackorby, C., D. Donaldson and J.A. Weymark (1984), "Social choice with interpersonal utility
comparisons: a diagrammatic introduction", International Economic Review 25:327-356.
Blackorby, C., D. Donaldson and J.A. Weymark (1990), "A welfarist proof of Arrow's theorem",
Recherches Economique de Louvain 56:259-286.
Blackorby, C., W Bossert and D. Donaldson (1995), "Intertemporal population ethics: critical-level
utilitarian principles", Econometrica 63:1303--1320.
Blackorby, C., W. Bossert and D. Donaldson (1996), "Intertemporally consistent population ethics:
classical utilitarian principles", in: K. Arrow, A. Sen and K. Suzumura, eds., Social Choice Re-
Examined, Vol. 2 (Macmillan, London) pp. 137-162.
Blackorby, C., W Bossert and D. Donaldson (1997a), "Birth-date dependent population ethics: critical-
level principles", Journal of Economic Theory 77:260-284.
Blackorby, C., W Bossert and D. Donaldson (1997b), "Intertemporally consistent population ethics:
birth-date dependent classical principles", Japanese Economic Review 48:267-292.
Blackorby, C., W Bossert and D. Donaldson (1998), "Uncertainty and critical-level population principles",
Journal of Population Economics 11:1-20.
Blackorby, C., W. Bossert, D. Donaldson and M. Fleurbaey (1998), "Critical levels and the (reverse)
repugnant conclusion", Journal of Economics 67:1-15.
Blackorby, C., D. Donaldson and J.A. Weymark (1999), "Harsanyi's social aggregation theorem for
state-contingent alternatives", Journal of Mathematical Economics 32:365-387.
Blackorby, C., W. Bossert and D. Donaldson (1999a), "Information invariance in variable-population
social-choice problems", International Economic Review 40:403-422.
Blackorby, C., W. Bossert and D. Donaldson (1999b), "Functional equations and population ethics",
Aequationes Mathematicae 58:272-284.
Blackorby, C., W. Bossert and D. Donaldson (2000), "The value of limited altruism", Journal of Economic
Theory 95:37-70.
Ch. 11: Utilitarianism and the Theory of Justice 593

Blackorby, C., W Bossert and D. Donaldson (2001), "Population ethics and the existence of value
functions", Journal of Public Economics 81:301-308.
Blackorby, C., D. Donaldson and J.A. Weymark (2001), "Social aggregation and the expected utility
hypothesis", in: M. Salles and J. Weymark, eds., Justice, Political Liberalism, and Utilitarianism:
Themes from Harsanyi and Rawls (Cambridge University Press, Cambridge) forthcoming.
Blackwell, D., and M. Girshick (1954), Theory of Games and Statistical Decisions (Wiley, New York).
Blau, J.H. (1976), "Neutrality, monotonicity, and the right of veto: a comment", Econometrica 44:603.
Border, K. (1981), "Notes on von Neumann-Morgenstern social welfare functions", Unpublished
manuscript (California Institute of Technology, Department of Economics).
Bordes, G.A., P. Hammond and M. Le Breton (1997), "Social welfare functionals on restricted
domains and in economic environments", Unpublished manuscript (Stanford University, Department
of Economics).
Bossert, W. (1990a), "An axiomatization of the single-series Ginis", Journal of Economic Theory
50:82-92.
Bossert, W (1990b), "Maximin welfare orderings with variable population size", Social Choice and
Welfare 7:39-45.
Bossert, W (1990c), "Social evaluation with variable population size: an alternative concept",
Mathematical Social Sciences 19:143-158.
Bossert, W (1991), "On intra- and interpersonal utility comparisons", Social Choice and Welfare
8:207-219.
Bossert, W. (2000), "Welfarism and information invariance", Social Choice and Welfare 17:321-336.
Bossert, W, and F Stehling (1992), "A remark on admissible transformations for interpersonally
comparable utilities", International Economic Review 33:739-744.
Bossert, W., and E Stehling (1994), "On the uniqueness of cardinally interpreted utility functions", in:
W. Eichhorn, ed., Models and Measurement of Welfare and Inequality (Springer, Berlin) pp. 537-551.
Bossert, W, and J.A. Weymark (2001), "Utility in social choice", in: S. BarberA, P. Hammond and
C. Seidl, eds., Handbook of Utility Theory, Vol. 2, Extensions (Kluwer, Dordrecht) forthcoming.
Broome, J. (1990), "Bolker-Jeffrey expected utility theory and axiomatic utilitarianism", Review of
Economic Studies 57:477-502.
Broome, J. (1991a), Weighing Goods (Basil Blackwell, Oxford).
Broome, J. (1991b), "The structure of good: decision theory and ethics", in: M. Bacharach and S. Hurley,
eds., Foundations of Decision Theory (Basil Blackwell, Oxford) pp. 123-146.
Broome, J. (1993), "Goodness is reducible to betterness: the evil of death is the value of life", in:
P. Koslowski, ed., The Good and the Economical: Ethical Choices in Economics and Management
(Springer, Berlin) pp. 69-83.
Broome, J. (1999), Weighing Lives (Oxford University, UK) unpublished.
Brown, D.J. (1972), "Mill on liberty and morality", Philosophical Review 81:133-158.
Carlson, E. (1998), "Mere addition and the two trilemmas of population ethics", Economics and
Philosophy 14:283-306.
Coulhon, T, and P. Mongin (1989), "Social choice theory in the case of von Neumann-Morgenstern
utilities", Social Choice and Welfare 6:175-187.
Dalton, H. (1920), "The measurement of the inequality of incomes", Economic Journal 30:348-361.
Dasgupta, P (1993), An Inquiry into Well-Being and Destitution (Clarendon, Oxford).
d'Aspremont, C. (1985), "Axioms for social welfare orderings", in: L. Hurwicz, D. Schmeidler and
H. Sonnenschein, eds., Social Goals and Social Organizations: Essays in Memory of Elisha Pazner
(Cambridge University Press, Cambridge) pp. 19-76.
d'Aspremont, C., and L. Gevers (1977), "Equity and the informational basis of collective choice",
Review of Economic Studies 44:199-209.
d'Aspremont, C., and L. Gevers (2002), "Interpersonal comparability of welfare and social choice
theory", in: K.J. Arrow, A.K. Sen and K. Suzumura, eds., Handbook of Social Choice and Welfare
(Elsevier, Amsterdam) Chapter 10, this volume.
594 C. Blackorby et al.

De Meyer, B., and P. Mongin (1995), "A note on affine aggregation", Economics Letters 47:177-183.
Debreu, G. (1959), Theory of Value: An Axiomatic Analysis of Economic Equilibrium (Wiley, New
York).
Debreu, G. (1960), "Topological methods in cardinal utility theory", in: K.J. Arrow, S. Karlin and
P. Suppes, eds., Mathematical Methods in the Social Sciences, 1959; Proceedings (Stanford University
Press, Stanford) pp. 16-26.
DeMeyer, E, and C.R. Plott (1971), "A welfare function using 'relative intensity' of preference", Quarterly
Journal of Economics 85:179-186.
Deschamps, R., and L. Gevers (1978), "Leximin and utilitarian rules: a joint characterization", Journal
of Economic Theory 17:143-163.
Dhillon, A. (1998), "Extended Pareto rules and relative utilitarianism", Social Choice and Welfare
15:521-542.
Diamond, P. (1967), "Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility:
comment", Journal of Political Economy 75:765-766.
Dixit, A.K. (1980), "Interpersonal comparisons and social welfare functions", Unpublished manuscript
(University of Warwick, Department of Economics).
Domotor, Z. (1979), "Ordered sum and tensor product of linear utility structures", Theory and Decision
11:375-399.
Donaldson, D., and J.A. Weymark (1980), "A single-parameter generalization of the Gini indices of
inequality", Journal of Economic Theory 22:67-86.
Eichhorn, W. (1978), Functional Equations in Economics (Addison-Wesley, Reading).
Falmagne, J.-C. (1981), "On a recurrent misuse of a classical functional equation result", Journal of
Mathematical Psychology 23:190-193.
Fishburn, P.C. (1984), "On Harsanyi's utilitarian cardinal welfare theorem", Theory and Decision 17:
21-28.
Fishburn, P.C., and E Roberts (1989), "Uniqueness in finite measurement", in: F. Roberts, ed., Applications
of Combinatorics and Graph Theory to the Biological and Social Sciences (Springer, New York)
pp. 103-137.
Fishburn, P.C., H. Marcus-Roberts and E Roberts (1988), "Unique finite difference measurement", SIAM
Journal of Discrete Mathematics 1:334-354.
Fleming, M. (1952), "A cardinal concept of welfare", Quarterly Journal of Economics 66:366-384.
Fleurbaey, M., and E Maniquet (2002), "Compensation and responsibility", in: K.J. Arrow, A.K. Sen
and K. Suzumura, eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam)
forthcoming.
Foster, J.E. (2002), "Freedom, opportunity and well-being", in: K.J. Arrow, A.K. Sen and K. Suzumura,
eds., Handbook of Social Choice and Welfare, Vol. 2 (Elsevier, Amsterdam) forthcoming.
Gevers, L. (1979), "On interpersonal comparability and social welfare orderings", Econometrica 47:
75-89.
Goodin, R. (1991), "Utility and the good", in: P. Singer, ed., A Companion to Ethics (Blackwell, Oxford)
pp. 241-248.
Gorman, W.M. (1968), "The structure of utility functions", Review of Economic Studies 32:369-390.
Griffin, J. (1986), Well-Being: Its Meaning, Measurement, and Moral Importance (Clarendon, Oxford).
Guha, A.S. (1972), "Neutrality, monotonicity, and the right of veto", Econometrica 40:821-826.
Hammond, PJ. (1976), "Equity, Arrow's conditions, and Rawls' difference principle", Econometrica
44:793-804.
Hammond, P.J. (1979), "Equity in two person situations: some consequences", Econometrica
47:1127-1135.
Hammond, P.J. (1981), "Ex-ante and ex-post welfare optimality under uncertainty", Economica 48:
235-250.
Hammond, P.J. (1983), "Ex-post optimality as a dynamically consistent objective for collective choice
Ch. 11: Utilitarianism and the Theory of Justice 595

under uncertainty", in: P. Pattanaik and M. Salles, eds., Social Choice and Welfare (North-Holland,
Amsterdam) pp. 175-205.
Hammond, P.J. (1988), "Consequentialist demographic norms and parenting rights", Social Choice and
Welfare 5:127-146.
Hammond, PJ. (2001), "Interpersonally comparable utility", in: S. BarberA, P Hammond and C. Seidl,
eds., Handbook of Utility Theory, Vol. 2, Extensions (Kluwer, Dordrecht) forthcoming.
Hare, R. (1982), "Ethical theory and utilitarianism", in: A.K. Sen and B. Williams, eds., Utilitarianism
and Beyond (Cambridge University Press, Cambridge) pp. 23-38.
Harsanyi, J.C. (1953), "Cardinal utility in welfare economics and in the theory of risk-taking", Journal
of Political Economy 61:434 435.
Harsanyi, J.C. (1955), "Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility",
Journal of Political Economy 63:309-321.
Harsanyi, J.C. (1977), Rational Behavior and Bargaining Equilibrium in Games and Social Situations
(Cambridge University Press, Cambridge).
Hurka, T. (1982), "Average utilitarianism", Analysis 42:65-69.
Kolm, S.-Ch. (1969), "The optimal production of social justice", in: J. Margolis and S. Guitton, eds.,
Public Economics (Macmillan, London) pp. 145-200.
Krantz, D., R.D. Luce, P Suppes and A. Tversky (1971), Foundations of Measurement, Vol. I, Additive
and Polynomial Representations (Academic Press, New York).
Maskin, E. (1978), "A theorem on utilitarianism", Review of Economic Studies 45:93-96.
Mill, J.S. (1861), Utilitarianism (Collins, London). Reprinted in 1969 in: Collected Works of John Stuart
Mill, Vol. X (University of Toronto Press, Toronto) pp. 203-259.
Milnor, J. (1954), "Games against nature", in: R. Thrall, C. Coombs and R. Davis, eds., Decision
Processes (Wiley, New York) pp. 49-59.
Mongin, P. (1994), "Harsanyi's aggregation theorem: multi-profile version and unsettled questions",
Social Choice and Welfare 11:331-354.
Mongin, P (1995), "Consistent Bayesian aggregation", Journal of Economic Theory 66:313-351.
Mongin, P. (1998), "The paradox of the Bayesian experts and state-dependent utility theory", Journal of
Mathematical Economics 29:331-361.
Mongin, P. (2001), "The impartial observer theorem and social ethics", Economics and Philosophy
17:147-180.
Mongin, P., and C. d'Aspremont (1998), "Utility theory and ethics", in: S. Barbera, P Hammond and
C. Seidl, eds., Handbook of Utility Theory, Vol. 1, Principles (Kluwer, Dordrecht) pp. 371-481.
Murphy, L. (1998), "Institutions and the demands ofjustice", Philosophy and Public Affairs 27:251-291.
Ng, Y.-K. (1989), "What should we do about future generations? Impossibility of Parfit's Theory X",
Economics and Philosophy 5:235-253.
Parfit, D. (1976), "On doing the best for our children", in: M. Bayles, ed., Ethics and Population
(Schenkman, Cambridge) pp. 100-102.
Parfit, D. (1982), "Future generations, further problems", Philosophy and Public Affairs 11:113-172.
Parfit, D. (1984), Reasons and Persons (Oxford University Press, Oxford).
Rawls, J (1971), A Theory of Justice (Harvard University Press, Cambridge).
Roberts, K.W.S. (1980a), "Possibility theorems with interpersonally comparable welfare levels", Review
of Economic Studies 47:409-420.
Roberts, K.WS. (1980b), "Interpersonal comparability and social choice theory", Review of Economic
Studies 47:421-439.
Roemer, J.E. (1996), Theories of Distributive Justice (Harvard University Press, Cambridge).
Sen, A.K. (1970), Collective Choice and Social Welfare (Holden-Day, San Francisco).
Sen, A.K. (1973), On Economic Inequality (Oxford University Press, Oxford).
Sen, A.K. (1974), "Informational bases of alternative welfare approaches: aggregation and income
distribution", Journal of Public Economics 3:387-403.
Sen, A.K. (1976), "Welfare inequalities and Rawlsian axiomatics", Theory and Decision 7:243-262.
596 C Blackorby et al.

Sen, A.K. (1977a), "On weights and measures: informational constraints in social welfare analysis",
Econometrica 45:1539-1572.
Sen, A.K. (1977b), "Non-linear social welfare functions: a reply to Professor Harsanyi", in: R. Butts and
J. Hintikka, eds., Foundational Problems in the Special Sciences (Reidel, Dordrecht) pp. 297-302.
Sen, A.K. (1979), "Personal utilities and public judgements: or what's wrong with welfare economics?",
Economic Journal 89:537-558.
Sen, A.K. (1985), Commodities and Capabilities (Elsevier, Amsterdam).
Sen, A.K. (1986), "Social choice theory", in: K. Arrow and M. Intriligator, eds., Handbook of
Mathematical Economics, Vol. III (North-Holland, Amsterdam) pp. 1073-1181.
Sen, A.K., and B. Williams (1982), "Introduction", in: A.K. Sen and B. Williams, eds., Utilitarianism
and Beyond (Cambridge University Press, Cambridge) pp. 1-21.
Sidgwick, H. (1966), The Methods of Ethics (Dover, New York). Originally published in 1907.
Sikora, R. (1978), "Is it wrong to prevent the existence of future generations?", in: R. Sikora and
B. Barry, eds., Obligations to Future Generations (Temple, Philadelphia) pp. 112-166.
Sumner, L. (1996), Welfare, Happiness, and Ethics (Clarendon, Oxford).
Suzumura, K. (1999), "Consequences, opportunities, and procedures", Social Choice and Welfare 16:
17-40.
Tungodden, B. (1999), "Social choices with independent norm levels", Unpublished manuscript
(Norwegian School of Economics and Business Administration, Norway).
von Neumann, J., and O. Morgenstern (1944), Theory of Games and Economic Behavior (Princeton
University Press, Princeton).
von Neumann, J., and 0. Morgenstern (1947), Theory of Games and Economic Behavior, 2nd Edition
(Princeton University Press, Princeton).
Weymark, J.A. (1981), "Generalized Gini inequality indices", Mathematical Social Sciences 1:409-430.
Weymark, J.A. (1991), "A reconsideration of the Harsanyi-Sen debate on utilitarianism", in: J. Elster and
J. Roemer, eds., Interpersonal Comparisons of Well-Being (Cambridge University Press) pp. 255-320.
Weymark, J.A. (1993), "Harsanyi's social aggregation theorem and the weak Pareto principle", Social
Choice and Welfare 10:209-222.
Weymark, J.A. (1994), "Harsanyi's social aggregation theorem with alternative Pareto principles", in:
W. Eichhorn, ed., Models and Measurement of Welfare and Inequality (Springer, Berlin) pp. 869-887.
Weymark, J.A. (1995), "Further remarks on Harsanyi's social aggregation theorem and the weak Pareto
principle", Social Choice and Welfare 12:87-92.
Weymark, J.A. (1998), "Welfarism on economic domains", Mathematical Social Sciences 36:251-268.
Zhou, L. (1997), "Harsanyi's utilitarianism theorems: general societies", Journal of Economic Theory
72:198-207.
Chapter 12

INEQUALITY, POVERTY AND WELFARE*

BHASKAR DUTTA
University of Warwick, Coventry, England, UK

Contents
Abstract 598
Keywords 598
1. Introduction 599
2. Preliminaries 600
3. Measurement of inequality 601
3.1. The Atkinson Kolm-Sen approach 601
3.2. Dominance and inequality 607
3.3. Characterization theorems 612
3.4. Measuring mobility 616
4. Measurement of poverty 619
5. Concluding remarks 627
References 628

* I am most grateful to S.R. Chakravarty, James Foster, Amartya Sen and John Weymark for very
helpful comments and suggestions.

Handbook of Social Choice and Welfare, Volume 1, Edited by K.J Arrow, A.K. Sen and K. Suzumura
© 2002 Elsevier Science B. V All rights reserved
598 B. Dulta

Abstract

This chapter is concerned with issues arising from the construction of ethical measures
of inequality and poverty. The recent literature on measurement of inequality and
poverty emphasizes the close connection between social welfare functions and ethical
indices of inequality and poverty. This chapter surveys the main issues in this literature.
In particular, we discuss how indices of inequality can be constructed from social
welfare functions, and vice versa. Other issues include the equivalence theorems which
provide the analytical foundations of the approach which declares one distribution to
be more equal than another only when all "sensible" measures agree on the ranking.
The chapter also discusses the measurement of mobility. Finally, the chapter describes
some of the parallel issues which arise in the measurement of poverty.

Keywords

inequality, social welfare functions, Lorenz curve, decomposition, poverty

JEL classification: D6
Ch. 12: Inequality, Poverty and Welfare 599

1. Introduction

Most people subscribe to the view that other things remaining the same, a more equal
distribution of incomes is preferable to a less equal distribution. Similarly, almost
everyone will agree that the reduction of poverty is an important goal of public
policy. Unfortunately, the congruence of views about reducing inequality and poverty
exists along with a wide disagreement about how best to measure these complex
phenomena. Like the proverbial elephant, concepts such as equality and poverty are
easy to recognise but much harder to describe.
Issues such as the promotion of greater equality lie squarely in the domain of
welfare economics. But, as Sen (1973) points out, traditional welfare economics offers
very little help in so far as distributional issues are concerned. Welfare rankings of
alternative income distributions must inevitably involve a comparison of the gains and
losses of utility of different income groups. Unfortunately, Paretian welfare economics
eschews interpersonal comparison of utilities, and is thus unable to pass judgements
on inequality.
Perhaps it is this failure of traditional welfare economics which explains the
dominant tendency in empirical work to define inequality as some (statistical) measure
of dispersion of the frequency distribution of incomes. Although Dalton (1920) had
earlier pointed out the close connection between social welfare and income distribution,
Atkinson (1970) and Kolm (1969) have been instrumental in initiating the modern
approach to the measurement of inequality 1. This approach explicitly endowed indices
of inequality with a normative interpretation by establishing a relationship between
these indices and social-evaluation orderings defined on income distributions. Each
index derived through this approach is based on a specific set of distributional value
judgements. Thus, a comparison of different indices of inequality can be made by
comparing the corresponding sets of value judgements. This essentially involves an
axiomatic approach in which inequality measures could be characterized by different
sets of axioms.
This chapter provides a brief introduction to the literature on ethical measurement of
inequality and poverty. In order to keep the chapter to a manageable length, I have had
to restrict attention to only some of the most important themes in the recent literature.
This has resulted in the omission of several significant issues. The reader is referred
to several recent books which provide much deeper and extensive treatments of this
important topic 2.

1 Sen (1976b) played an analogous role in stimulating research on the measurement of poverty.
2 See, for instance, Cowell (1995), Chakravarty (1990), Lambert (1989), Kakwani (1980a) and Silber
(1999).
600 3 Dutta

2. Preliminaries

Indices of inequality and poverty are defined on sets of income distributions 3. Different
domains may be used for different indices. We will use D" to stand for any of three
possible domains corresponding to a population of size n: 91n, 91, 9++. In order to
allow comparisons to be made when population sizes vary, let D = U,EN cND, where
N is the set of natural numbers.
An index of inequality is a function I: D - 9. Each inequality index I is associated
with a sequence In: n2"- 91}, eN, one for each population size n. Each I" is a
continuous and strictly S-convex function 4. We will sometimes find it convenient to
refer to indices of equality. These are real-valued, continuous and strictly S-concave
functions.
An index ofpoverty is a function P: - 91. A typical index of poverty P is also
associated with a class P": 2" - 91}, G v, where each P" is an S-convex function.
The main focus of this chapter is on ethical indices of inequality and poverty. These
ethical indices are closely related to society's value judgements summarized in a social
welfare function 5 . We will assume that the social welfare function W: 2) -- 9 is
associated with a class { W": 2)" --*n , oN, where each W" is continuous, S-concave
and increasing along rays, that is, W"(Aly) > Wn(y) whenever A. > 1 and y e 9+.
It will often be convenient to drop the superscript n and simply write W, I or P
instead of W", I" or P".
These properties of a social welfare function together guarantee the existence and
uniqueness of the equally-distributed equivalent income (EDEI) for each income
distribution y E 2)n. Given any y E D", the EDEI is that income which if received
by each person, results in a distribution that is socially indifferent to y. Formally, it is
given by

W"(il) = W"(y), (2.1)

where 1, = (1,..., 1) E 9++. Equation (2.1) can be used to write the EDEI as an
explicit function of the income distribution y: ~ = ?"(y). Notice that continuity of W
implies that is also continuous. It is also clear that and W are monotonic
transformations of one another. That is,

Vy, E '", W"(y) > Wn"() X: 7"(y) > ?n(y). (2.2)

3 In referring to income distributions, we are not distinguishing between real and nominal incomes,
although this distinction should clearly be made in intertemporal or interspatial comparisons.
4 A function f:D" - 91 is S-convex if and only if f(Qy) < fv() for all y D and all nxn
bistochastic matrices Q. Strict S-convexity requires strict inequality whenever Qy is not a permutation
of y. Correspondingly, a function f: D" - 9 is S-concave if and only iff(Qy) >_f(y) for all y E D"
and all nxn bistochastic matrices.
5 We will sometimes use the term social-evaluation function instead of social welfare function.
Ch. 12: Inequality, Poverty and Welfare 601

In the subsequent sections, we will also use orderings over D instead of functions
on D when we want to emphasize the ordinal characteristics or properties of these
functions. For instance, I, p and t w will correspond to the inequality, poverty
and social welfare orderings over D.

3. Measurement of inequality

3.1. The Atkinson-Kolm-Sen approach

Since our concern with inequality springs mainly from the feeling that a reduction
in inequality results in an increase in social welfare provided mean income remains
unchanged, it makes sense to explicitly construct inequality measures from social
welfare functions. This section describes the close connection between social welfare
functions and ethical indices of inequality 6 . In particular, we discuss how indices
of inequality can be constructed from social welfare functions, and vice versa. The
move from welfare functions to inequality can be used to construct new measures of
inequality, while the move in the reverse direction helps in uncovering the implicit
value judgements that are embedded in specific measures of inequality.
This is the methodology associated with Atkinson (1970), Kolm (1969) and Sen
(1973) (AKS). Given any distribution y D, an AKS index of inequality is the
percentage of total income that can be discarded without affecting social welfare if
income is distributed equally. Hence, the AKS index is given by

Vn E N, Vy E Dn2, /AKS(Y) = 1- -,(y) (3.1)

where tn(y) is the mean income corresponding to the distribution y. Note that
since W" (and hence ?E) is S-concave, w-n(y) < /t(y). Hence, 0 < I'KS(Y) < 1.
Moreover, every AKS index of inequality is normatively significant. That is, for
every pair of income distributions y and y' such that unf(y) = u(y'),

IAKS(Y) < IAKS(Y') iff W (y) > W (y'). (3.2)

In other words, an inequality index is normatively significant if whenever two

distributions have the same mean income, greater inequality is associated with lower
social welfare.

6 This has been labelled the Atkinsonian perspective by Foster and Sen (1997). See Blackorby, Bossert
and Donaldson (1999) for an elegant survey of the normative approach to measurement of inequality. Sen
(1978) points out some conceptual difficulties associated with this approach. See also Foster (1994).
602 B. Dutta

Atkinson (1970) specified the following single-parameter 7 class of functions:

6 l, 0,
W (y) (3.3)
E = 0.

This specification of the welfare function yields the following class of inequality
indices:

I- , e6<1, e 0,
iYc
i~y) (3.4)
1 - 6())
e = O.

These AKS indices are relative indices because they have the property that equal
proportionate changes in individual incomes do not change the level of inequality.
While this method of constructing inequality indices is perfectly general and can be
applied to any specification of the welfare function, it will not always yield relative
indices.
It is also possible to construct the social welfare functions which are implied by
various inequality measures commonly used in the literature on income distribution.
This enables us to analyze explicitly the value judgements which are embodied in these
measures. Consider, for instance, the well-known Gini index of relative inequality. This
is given by
n n
2 (3.5)
'(Y) = 2n ,(y) Yi - jl

From Sen (1973), this can be written as

I(y)= 1 nW 1(2i- l)i (3.6)

where 9 is a permutation of y such that Pi > i +1 for all i = 1,..., n I.

7 The parameter c is the degree of inequality aversion. The lower the value of e, the greater is the
aversion to inequality.
Ch. 12: Inequality, Poverty and Welfare 603

So, 7a(y) and WG(y) are given by 8

~(y) = n2 (2i l)i (3.7)

il

WG(y) = 'P(3;(y)), (3.8)

where is some increasing monotonic transformation.
This specification shows that the welfare weights associated with individual incomes
are increasing functions of the rank orders of the individuals. Thus, the preference
for greater equality is built into the social welfare function - the income of a poorer
individual has a larger weight.
Kolm (1969) also suggested an alternative to the AKS family of indices. In this
alternative formulation, an inequality index measures the absolute amount of income
per person that can be saved without affecting welfare if the remaining income is
distributed equally. The counterpart to Equation (3.1) now becomes
Vn E N, Vy C D", I(y) = ut"(y)- "(y). (3.9)
The Kolm class of inequality indices is also normatively significant since the EDEI
must be a decreasing function of the level of inequality.
Kolm (1976a,b) investigated the question of scale independence in some detail.
While members of the Atkinson family as well as the Gini coefficient are independent
of mean incomes, this is not the only possible form of scale independence. An absolute
index is one that is invariant to equal absolute changes in individual incomes. Bossert
and Pfingsten (1990) and Pfingsten (1986) also develop intermediateinequality indices
which are compromises between relative and absolute indices.
A family of welfare functions which gives rise to absolute inequality indices is the
Kolm-Pollak family 9:

Vn c N, Vy E , W(y) = -n[
In exp(-6yi) for 6 > 0. (3.10)

The corresponding inequality indices are given by

Vn E N, Vy e Dn, In(y) = ,u(y)+ Iln Zexp(-6yi)j for 6 > 0. (3.11)

Here, 6 corresponds to the degree of inequality aversion with higher values of 6

signifying a greater degree of inequality aversion. As 6 approaches o, the Kolm-
Pollak index approaches the absolute maximin index:
Vn N, y C Dn, In(y) = t'(y)-min{yl,... y,. (3.12)
Since the inequality indices are derived from social welfare functions, each
particular type of scale independence imposes a restriction on the functional form of

8 See Sen (1973) and Sheshinski (1972).

9 See Pollak (1971).
604 B. Dutta

the social welfare function. Inspection of Equation (3.1) suggests that the AKS family
can be relative indices if, and only if, the social welfare function is homothetic lo
This follows because homotheticity of the social welfare function implies that the
corresponding EDEI function - is linearly homogeneous. Since u" is also linearly
homogeneous, this ensures that the AKS family is homogeneous of degree zero in
individual incomes.
The property of translatability1 of social welfare functions ensures that the
corresponding Kolm indices of inequality belong to the family of absolute inequality
measures. If the social welfare function W ' is translatable, then the EDEI function is
unit translatable. From Equation (3.9), it follows that the corresponding Kolm index
is an absolute index.
An important class of social welfare functions, which have been labelled dis-
tributionally homothetic, 12 are those which are both homothetic and translatable.
Distributionally homothetic social welfare functions can be used to derive both relative
and absolute inequality indices. Of course, whether any specific measure derived from
a distributionally homothetic function is a relative or absolute measure depends upon
whether Equations (3.1) or (3.9) is employed 3.
An example of a distributionally homethetic social welfare function is the Gini
welfare function described in Equation (3.8)14. Thus, using Equation (3.9), we get
the absolute 15 Gini index as
n

IAG(Y) = (Y) - (Y) = U(y) - (2i-I)i. (3.13)

i=l1

It would be unfortunate if the set of relative inequality measures had to be restricted

to the narrow class of indices which can be derived from homothetic welfare functions.
Fortunately, Blackorby and Donaldson (1978) present an alternative procedure which
yields relative inequality measures for non-homothetic social welfare functions 16
Their procedure involves the construction of distance functions which are (linear)

'0 A function is homothetic if it is a monotonic transformation of a linear homogeneous function.

1 A functionf: " -D 91 is unit-translatable ifVy D' and C 9,f(y +cl,) =f(y) + . A functions
is translatable if it is a monotone increasing transformation of a unit-translatable function.
12 See Blackorby and Donaldson (1978, 1980a).
13 Eichhorn (1988) and Eichhorn and Gehrig (1982) show that an index cannot simultaneously be a
relative and an absolute index.
14 The social welfare function which yields the coefficient of variation as the (relative) inequality
measure is also distributionally homothetic.
15 Yitzhaki (1979) provides a novel interpretation of the absolute Gini index by showing that it represents
a measure of deprivation for the society as a whole. This is generalized by Chakravarty and Mukherjee
(1999) who show that many of the existing inequality indices can be interpreted as measures of
deprivation. On the general concept of deprivation, see Runciman (1966).
16 In an analogous exercise Blackorby and Donaldson (1980a) derive absolute inequality indices from
welfare functions which are not translatable.
Ch. 12: Inequality, Poverty and Welfare 605

homogeneous and implicit representations of social welfare functions. Let Wn }. e N

be any sequence of welfare functions. Choose a reference level of welfare w. Define
the distance functions {Fn: R x Dn - },,
n N by 17

Vy E D), Fn(w,y) = max{/A E 9l I Wn"( ) > w}. (3.14)

Take any arbitrary income distribution y. Then since W'(y/F"(w,y)) = w, the

social welfare function is implicitly defined by F'(w,y) = 1. Clearly, F n is linearly
homogeneous in y, and decreasing in w. A measure of relative inequality is obtained
as the percentage of total income which can be discarded if the remainder is distributed
equally so as to keep social welfare at the reference level w. That is,

Fn(w,y)
Vn c N, Vy E Dn, IBD(Y) = 1- F(w, n(y)ln) (3.15)

It is obvious that linear homogeneity of F" ensures that IBD is a relative inequality
index.
Of course, no assumption has been made about homotheticity of the underlying
welfare function W " . However, the derivation of relative inequality measures from
any arbitrary welfare function has been obtained at some cost. Blackorby and
Donaldson point out that the inequality measure is no longer normatively significant.
So, a redistribution of income can change inequality and welfare in the same
direction. Obviously, this cannot happen with the AKS indices which are always
normatively significant. Note that while the Blackorby-Donaldson indices are always
relative indices, they will also be normatively significant if the welfare function is
homothetic 18
Blackorby and Donaldson (1978) also discuss how their procedure can be "inverted"
in order to recover the social welfare function which is implied by a given inequality
index. But, they also point out a problem with their own procedure. Starting from an
ordinal social welfare function, they derive a cardinally significant inequality index -
"it makes sense to speak of reducing inequality by 17%" 19. However, inequality is
typically considered to be an ordinal concept - in common parlance, one usually
speaks of distribution x being more or less unequal than y. Unfortunately, two ordinally
equivalent inequality indices - for instance, G and G2 where G is the Gini coefficient -
imply social welfare functions which are not ordinal transformations of one another.
This is problematic because it suggests that two ordinally equivalent inequality indices
are associated with different sets of value judgements.

17 We are glossing over some technical details since the maximisation in Equation (3.14) may not be

well-defined on the boundary of D'".

18 If Wn is homothetic, then F(w,y) is multiplicatively separable in w and y. Inspection of
Equation (3.10) implies that the inequality index must be normatively significant in this case.
19 See Blackorby and Donaldson (1978).
606 B. Dutta

Ebert (1987) proposes a procedure which avoids this difficulty. Ebert recognises
that social welfare depends both on the size of the "cake" as well as its distribution.
In other words, there is no one-to-one correspondence between social welfare and
inequality since the level of social welfare is also an increasing function of mean
income. Ebert explicitly introduces an ordering t, defined on R++ x Yf, where Yil is
the set of indifference classes of an inequality ordering >-. Then, consistency of TVJ
with t7 requires that the ranking of t' and t' coincides over all income distributions
y,yJ c D" if y and y have the same mean income. Ebert then derives a sequence of
social welfare orderings ( ) N for each pair (, ) N. These social welfare
orderings are defined by
Vn N, Vy,J fE 2, wy X0 , (),)
(y),Yl), (3.16)
where Yl,Yi denote the indifference class of -7nto which y and y belong. Ebert also
shows that the procedure can be reversed. That is, one can start with a sequence of
welfare orderings and derive a pair of orderings >-n and t'b which are consistent with
one another2 0 .
In the preceding discussion, we have implicitly assumed a homogeneous population.
When the population is heterogeneous, then individuals of different types may have
different needs, and the social welfare function will in general no longer treat
individuals alike2 1 . In particular, a perfectly equal division of the cake may not be
the distribution maximising social welfare. This implies that the concept of the EDEI
needs to be modified in order to derive AKS inequality indices when the population
is heterogeneous. Weymark (1999) presents an elegant discussion of how to make
the necessary adjustments. Weymark starts from the observation that under standard
regularity assumptions on the social welfare function, n.E(x) is the minimum amount
of aggregate income required to obtain an income distribution which is socially
indifferent to the given distribution x. Now, let A = ( 1,...., A,), and define the function
A:913 - 9iby

A(x) = min Ai subject to W'() = W'(x). (3.17)

The heterogeneous counterpart to Equation (3.1) is obtained by replacing the EDEI

by A(x) and the mean income by aggregate income. That is,

IH1(x) = 1 A(x ) . (3.18)

Notice that A(x) is the minimum amount of aggregate income required to generate
the same social welfare as the actual distribution of income x. Of course, if

20 See Blackorby and Donaldson (1984), Dutta and Esteban (1992) and Sen (1976a) on related issues.
21 So, the social welfare function may no longer satisfy symmetry, a condition which we define in the
next section.
Ch. 12: Inequality, Poverty and Welfare 607

the population is homogeneous, then A(x) equals n(y). Clearly, just like the
AKS indices for homogeneous populations, IH(x) measures the percentage of aggregate
income that can be discarded without affecting social welfare.
Throughout the rest of this chapter, we revert to the assumption that the population is
homogeneous. This qualification is important since some of the axioms which will be
imposed on inequality (or welfare functions) are compelling only when the population
is homogeneous.

3.2. Dominance and inequality

The discussion in the preceding section suggests that a wide variety of inequality
indices can be constructed, each index being "justified" or rationalized by the welfare
function from which it is derived. This rather large menu of available choices actually
constitutes an embarrassment of riches. It is obvious that two inequality indices can
produce contradictory rankings of different income distributions. This failure to pass
unambiguous judgements about the level of inequality implies a severe constraint on
how widely the theory can be used in making clear policy recommendations.
To a large extent, this ambiguity is due to the attempt to measure inequality too
precisely, because this precision has resulted in inequality rankings which are complete
orderings. The sets of value judgements embodied in different welfare functions
(and hence in the inequality indices implied by them) can often be in conflict with
each other. In such cases, the inequality measures implied by them will produce
contradictory rankings. An alternative approach is to base inequality judgements on a
notion of dominance - that is, to declare one distribution to be more unequal than
another only when all "sensible" inequality indices agree on the ranking of these
distributions. Of course, a price has to be paid for this conservative approach - there
will be cases in which it will not be possible to compare income distributions in terms
of their extent of inequality since the dominance ranking will typically be incomplete.
Nevertheless, a useful research agenda has been the attempt to narrow down the range
of cases in which inequality judgements are incomplete.
The first step in this direction is to specify a set of value judgements which all
sensible measures of inequality should satisfy. One such basic principle is the principle
of transfers due to Dalton ( 1920). This says that an appropriate transfer of income from
a richer person to a poorer person must reduce the extent of inequality.
Let ei E 9+ denote the n-tuple (0,... 0, 1,0,... .,0), with 1 occurring in the ith
position.
Definition. Given x,y D,x is obtained from y by a progressive transfer if
(i) x - y = 6(ei - ej) for > 0, and (ii) yj > xi > i 22

22 Fields and Fei (1978) use a more restrictive concept of progressive transfer in which the magnitude
of the transfer (6) is small enough to ensure that individuals' rank orders are preserved.
608 B. Dutta

Definition. A sequence of inequality indices {I": D" - 9i} satisfies the Principle of
Transfers (PT) if for all n E N, for all x,y e D,',In(x) > I'(y) whenever x is obtained
from y by means of a progressive transfer2 3.
A principle which is very similar to the familiar condition of anonymity in social
choice theory is that the inequality index should not discriminate between individuals.
Definition. A sequence of inequality indices In: Dn R} satisfies Symmetry (S) if
1n(x) = I(7rx) for all x E D n and all permutation matrices Jr.
The classic papers of Atkinson (1970) and Kolm (1969) promoted the use of
dominance rankings by establishing the close connection between the hitherto
statistical concept of Lorenz curves, inequality indices of the Atkinson family and
the Principle of Progressive Transfers. This result has subsequently been extended by
Dasgupta, Sen and Starrett (1973), Shorrocks (1983), Foster (1985), Shorrocks and
Foster (1987) and Davies and Hoy (1995). An illuminating survey of this literature is
contained in Moyes (1999).
Given any x E D, let i denote the permutation of x so that incomes are arranged
in increasing order. That is, 1l < 2 ... < x,. Let Xk = =1 X denote the partial sum
of the k lowest incomes in x. The Lorenz curve represents the percentage of income
accruing to the l00p%poorest individuals in x for all p e (0,1). Hence, it is obtained
by first plotting the (n - 1) points (, x) for k = 1,2,... ,n - 1, and then joining
adjacent points starting with (0, 0) and ending with (1, 1). The Lorenz curve ordinate
corresponding to the lowest 100p% income is denoted L(x,p).
Definition. For all x,y E D, x Lorenz-dominates y, denoted xLy, if L(x,p) > L(y,p)
for all p E (0, 1) with strict inequality for some p2 4 .
It is clear that if x and y have the same population size and the same means, then
xLy iffXk > Yk, for all k and 25
Hardy, Littlewood and Polya established the analytical connections that underlie the
linkage between Lorenz domination and progressive transfers.
Lemma 3.1. Suppose x,y G D" with pl(x) = [t"(y). Then, xLy iff i is obtainedfrom )
by a finite sequence of progressive transfers.
Atkinson exploited the formal similarity between the ranking of income distributions
and the ranking of probability distributions in terms of expected utility 26. In particular,

23 The variance of logarithms is an index of inequality which violates the Principle of Transfers. On
this, see Creedy (1977) and Foster and Ok (1999).
24 See Gastwirth (1971) for a general representation of the Lorenz curve. See also Kakwani and Podder
(1973).
25 In the terminology of Hardy, Littlewood and Polya (1934) y is said to "majorize" x if Xk > Y for
all k, with strict inequality for some k. On majorization, see Marshall and Olkin (1979).
26 It is interesting that the book by Hardy, Littlewood and Polya does not even figure in Atkinson's list
of references.
Ch. 12: Inequality, Poverty and Welfare 609

he used results of Rothschild and Stiglitz (1970) to show that Lorenz dominance
was equivalent to second-order stochastic dominance 2 7 . We first define notions of
stochastic dominance before describing Atkinson's classic result.
Let q(x;z) = I{i I xi < z}J. That is, q(x;z) is the number of individual incomes in x
which do not exceed z. Also, define ¢r(x; z) L(z - i)
Let SSD denote second-order stochastic dominance and TSD denote third-order
stochastic dominance.
Definition. Let n E N, and x,y E D" with an (x) = Vt"(y). Then,
(i) x SSD y iff ¢2 (y; z) > 02(x; z) Vz E 9+ with strict inequality for some z E 91+.
(ii) x TSD y iff 03 (y; Z) > ¢3 (x; ) Vz E 9+ with strict inequality for some z E 9+.
The preceding definition sets the stage for the basic set of equivalences in the theory
relating to the measurement of inequality.
Theorem 3.1. For any n C N, let x,y E Dn with pu"(x) = Un'(y). Then, the following
statements are equivalent2 8 :
(i) x can be obtainedfrom by a finite sequence of progressive transfers.
(ii) xLy.
(iii) x SSD y.
(iv) W'(x) > Wn(y) for all strictly S-concave functions W' .
(v) I"(x) < In(y) for all I" satisfying PT and symmetry.
Theorem 3.1 provides an elegant justification for the use of Lorenz domination in
the measurement of inequality. Suppose the distribution x Lorenz-dominates y. Then,
since can be obtained from by a finite sequence of progressive transfers, all
inequality indices satisfying the basic principles of PT and symmetry must agree that
x is less unequal than y. It must also be the case that social welfare is higher in x than
in y according to all strictly S-concave social welfare functions. So, Lorenz domination
provides a "safe" criterion to conclude that one distribution is more unequal than
another. Conversely, if neither distribution Lorenz-dominates the other, then one can
always find a pair of inequality indices satisfying PD and S which disagree on the
ranking of the two distributions.
Theorem 3.1 is somewhat limited in scope because it provides a basis for ranking
two income distributions only when they have the same mean income and involve
the same number of people. This is problematic because interspatial or intertemporal
comparisons will typically involve different population sizes as well as different mean
incomes. Fortunately, the relevance or applicability of Theorem 3.1 can be extended
to account for these differences.

27 See Bawa (1976) or Hadar and Russell (1969) on stochastic dominance. See also Newbery (1970)
and Rothschild and Stiglitz (1973).
28 In Atkinson's (1970) formulation, the social welfare function was taken to be the sum of strictly
concave individual utility functions. Dasgupta, Sen and Starrett (1973) replaced this with strict
S-concavity. See Foster (I985) for an interesting account of the history of this result.
610 B. Dutta

Let x C D". Call y an m-replication of x if for some integer m, y = (y(), .. ,y(",))

where each y(i) = x. The following property, suggested by Dalton (1920), facilitates
comparisons of inequality across different population sizes:

Definition. An inequality index I: D i 9 satisfies the Population Principle (PP) if

I(y) = I(x) whenever y is a replication of x.
A related axiom was defined for social welfare functions by Dasgupta, Sen and Starrett
(1973).

Definition. W satisfies the Symmetry Axiom for Population (SAP) if mW(x) = W(y)
whenever y is an m-replication of x.
Thus, if W satisfies SAP, then per capita welfare remains unchanged under replica-
tions.
The following theorem, due to Dasgupta, Sen and Starrett (1973), shows that Lorenz
domination with the same mean implies a higher per capita welfare level even when
population sizes differ.

Theorem 3.2. Let x E DI,y Dm and ['t(x) = ui"'(y). If xLy, then w(x) > w"(l) if
the class W = { W }, E N. satisfies strict S-concavity and SAP
A theorem in a similar vein was proved by Foster (1985). Define an inequality index
to be Lorenz consistent if for all distributions x,y E D, xLy X I(x) > I(y) and
L(x,p) = L(y,p) V P E (0, 1) = I(x) = (y).

Theorem 3.3. An inequality measure I: D -- 9l is Lorenz consistent iff I is a relative

inequality measure satisfying PP PT and S.
Shorrocks (1983) discusses how to extend Theorem 3.1 to cover the case of variable
mean incomes. Shorrocks defines a Generalized Lorenz curve, which is essentially
the Lorenz curve scaled up by the mean income. Thus, the Generalized Lorenz curve
is obtained by plotting the points (,Xk) for k = 1,... , n - 1, and joining adjacent
points. Let GL(x; p) represent the Generalized Lorenz curve ordinate corresponding
to the lowest lOOp% income for all p E [0, 1]. Clearly, one can now define GL,
the binary relation representing Generalized Lorenz domination, in a straightforward
manner. Shorrocks shows that xGLy is equivalent to W(x) > W(y) for all increasing,
strictly S-concave functions.
Despite these extensions covering variable population sizes and variable mean in-
comes the scope of Theorem 3.1 is extremely limited. This becomes clear when one
considers the very stringent condition which has to be satisfied in order to ensure
that Lorenz curves of two ordered distributions do not intersect - one distribution
has to be obtained from the other by a sequence of progressive transfers. Another
way of viewing this stringency is to consider the large class of inequality indices
satisfying the principle of progressive transfers. Since Lorenz domination is equivalent
to the unanimous judgement of all inequality indices in this very large class, it is not
Cl. 12: Inequality, Poverty and Welfare 611

surprising that this unanimity will prevail over a very small set of pairs of income
distributions.
Shorrocks and Foster (1987) provide a very interesting extension of the basic Lorenz
dominance criterion. They consider "composite" transfers which consist of a transfer
of a "small" amount from individual j to individual i, combined with a simultaneous
transfer of the same amount from k to e. The amount has to be small so as to ensure
that the rank order of individual incomes is preserved after the transfer. Suppose now
that one is a regressive transfer while the other is a progressive transfer. Shorrocks and
Foster argue that if the progressive transfer occurs at the lower end of the income scale,
then a "sensible" measure of inequality should record a decrease in equality because
it should be more sensitive to transfers at the lower end of the income scale.
Definition. Let x,y E 2)". Then, x is obtained from y by a favourable composite
transfer (FACT) if (i) x - y = 6(ei - ej) + 6(e - ek), (ii) 2(x) = o2 (y), and
(iii) yi <Yj <Yk < ye and xi < xj <Xk < Xe.
Definition. An inequality index I: D - 9 is transfer sensitive iffl(x) < I(y) whenever
x is obtained from y by a FACT.
Shorrocks and Foster prove the following.
Theorem 3.4. For x,y CED", the following statements are equivalent:
(i) i can be obtainedfrom y by a finite sequence of rank-preservingprogressive
transfers and/or FACT;
(ii) x TSD y;
(iii) I(x) < I(y) for all transfer-sensitive inequality measures I.
It is worth pointing out that although transfer sensitivity seems a natural axiom,
it is not satisfied by the Gini index. To see why the Gini index violates transfer
sensitivity, notice that for income profiles with a given rank order of incomes, the
Gini index is linear in individual incomes. Hence, Theorem 3.4 implies that even if
x TSD y, the Gini index need not record a lower inequality for x. However, Chakravarty
(1988) constructs a family of Extended Gini indices satisfying transfer sensitivity. Let
¢: [0, 1] -- 91+ be some continuous, strictly increasing, strictly convex function with
O(0) = 0. Then, a (relative) 2 9 Extended Gini index is given by

IEG(X) = 2 [k (l ) (3.19)

Chakravarty points out that the Gini index does not satisfy transfer sensitivity because
it is linear in the difference (k -_ ). The insistence on strict convexity of 0 ensures
that the class of Extended Gini indices satisfies transfer sensitivity, and is consistent
with third-order stochastic dominance.

29 See Chakravarty (1988) for the absolute version.

612 B. Dutta

Atkinson and Bourguignon (1982) and Kolm (1977) have also explored the relation
between stochastic dominance, social welfare and inequality for distributions of more
than one variable. Multidimensional inequality and welfare comparisons are interesting
in a variety of contexts. For instance, one could be interested in comparisons of
inequality between vectors of income distributions, where elements in each vector
represent the distribution of income at a point in time. Alternatively, one may want to
assess the extent of inequality taking into account not just differences in incomes but
also other attributes such as educational attainment, access to health-care facilities,
and so on. It is often assumed that individual utilities (and hence social welfare)
are additive in "present" and "future" incomes. Kolm shows that in such cases,
there is a relatively straightforward extension of results for the one-dimensional case.
Unambiguous conclusions about inequality or social welfare can be drawn only if each
marginal distribution of one vector stochastically dominates the corresponding element
of the other vector. However, Atkinson and Bourguignon (1982) tackle the harder case
where social welfare is not separable in the different components. In this case, the
conditions for dominance must involve the joint distributions of the variables 30.

3.3. Characterizationtheorems

Since different inequality measures will rarely produce the same ranking of income
distributions, it is important to specify criteria or principles which enable us to evaluate
inequality measures. This provides a rationale for the axiomatic characterization of
inequality indices. An index can now be evaluated in terms of the properties or axioms
satisfied by it. In the preceding sections, some general properties such as the Principle
of Transfers, Symmetry, or the Population Principle have been discussed. However,
as Foster (1985) shows (see Theorem 3.3), these properties still let in a large class
of inequality indices. The purpose of this section is to discuss the role of additional
properties which help in narrowing down the class of permissible indices.
The notion of decomposability has proved to be extremely powerful in weeding
out measures of inequality. Loosely speaking, an inequality measure is decomposable
if it is possible to express overall inequality in a population as a function of
the inequalities within major subgroups. This is a very useful property if one is
interested in a disaggregated analysis of the causes of inequality. In particular,
inequality measures which are decomposable by population subgroups allow the overall
inequality to be resolved into contributions due to (a) inequality within groups formed
according to characteristics such as sex, race, religion, etc.; and (b) inequality between
groups arising due to variations in mean incomes across groups. Another type of
decomposition disaggregates overall inequality by inequality from different sources
of earnings or factor components.

30 See also Foster, Majumdar and Mitra (1990), Nlaasoumi (1986, 1989), Tsui (1995) and Weymark
(1999).
Ch. 12: Inequality, Poverty and Welfare 613

The definitive work on decomposition by population subgroups is that of Shorrocks

(1980, 1984) 3 1. Shorrocks (1980) focuses on additively decomposable inequality
measures. These are measures which can be written in the form

G
In(x) = E wg(p,n)l g(x g ) +I"(1l,
In) * t (3.20)

Here the population is partitioned into G disjoint subgroups. Subgroup g has

ng individuals, and has the income distribution x g with mean income pug. Also,
P= ( ...G, G) and n = (nl,... ,nG).
Notice that the weight attached to subgroup g, Wg, is allowed to vary with the
vectors u and n. The second term in Equation (3.20) represents the between-group
inequality. Since this is supposed to represent the inequality across groups due to
different levels of mean incomes, the income distribution in each subgroup g is
assumed to be perfectly equal 32 .
Shorrocks' principal result was to show that amongst the class of continuous Lorenz-
consistent inequality measures normalized so that inequality is zero at perfect equality,
the only measures which are additively decomposable belong to the single-parameter
Generalized Entropy family. This family is represented by

n c(c-1) [((x) -1 ,
IC'(x) = log( )) c = 0, (3.21)

(X)\ i/ · =
·
x log c=1.

When c = 2, 1' corresponds to the square of the coefficient of variation. I' is the Theil
index of inequality, while I' has also been suggested as a measure of inequality by
Theil (1967) 33.
Thus, additive decomposability implies a severe restriction on the form of the
inequality measure. Can we get away from this constricting straightjacket by imposing
a weaker form of decomposability? This is the question posed by Shorrocks (1984).

31 See also Bourguignon (1979), Cowell (1980, 1988), Cowell and Kuga (1981b), Foster and Shneyerov
(2000), Maasoumi (1986), Pyatt (1976) and Russell (1985).
32 Blackorby, Donaldson and Auersperg (1981) alter the decomposition formula by using EDEI of
subgroup distributions rather than the subgroup means in the between-group term.
33 For alternative axiomatizations of the Theil index, see Cowell and Kuga (1981a) and Foster (1983).
614 B. utta

Instead of Equation (3.20), Shorrocks (1984) imposed a weaker restriction. Call

an inequality measure decomposable or aggregative iff there exists an aggregator
function A such that

l"(x,y) = A(I"'(x), p(x), I'" (y), y(y)) (3.22)

where nx, n, are the dimensions of the vectors x and y, and n = n, + n,. Equation (3.22)
states that if the distributions x and y are combined into a single population, then
aggregate inequality depends only on the size, mean income and inequality of the
component distributions.
Unfortunately, Shorrocks shows that replacement of Equation (3.20) by (3.22)
does not really expand the menu of available choices. The only additions to the
Generalized Entropy family are indices which are monotonic transformation of the
members of this family! This means that the Atkinson family of indices are also
"weakly" decomposable. But if inequality is viewed as an ordinal concept, then even
the weak form of decomposability rules out all indices which are not members of the
Generalized Entropy family!
There have been attempts to examine the extent to which the Gini index can be
decomposed. As Foster and Sen (1997) remark, the Gini index can be given "an
additive but somewhat artificial form of decomposition". The "artificially" arises
because we are left with a residual term R after accounting for within-group and
between-group inequality. The residual term arises because the rank-order weighting
of individual incomes depends on incomes in the entire population. Clearly, the
pattern of weights changes if subgroup income distributions overlap. This suggests
that an exact decomposition of the Gini index should be possible when the population
is partitioned into groups whose distributions are non-overlapping. Indeed, Ebert
(1988) showed that the Generalized Entropy family and the Gini index are the
only indices amongst the class of continuous Lorenz-consistent indices satisfying a
differentiability 34 condition which can be additively decomposed for non-overlapping
partitions. Other papers which have addressed the issue of decomposability of the Gini
index include Bhattacharya and Mahalanobis (1967), Chakravarty (1990), Lambert and
Aronson (1993), Pyatt (1976), Love and Wolfson (1976) and Silber (1989).
There is also a sizeable literature [Fei et al. (1978), Fields (1980), Pyatt et al.
(1980), Rao (1969), Shorrocks (1982a)] proposing methods for decomposing overall
inequality into inequality arising from different factor components. While many of
these studies advocate the use of essentially ad hoc decomposition rules, Shorrocks
(1982a) provides an axiomatic basis for potential decomposition methods. However,
the focus of this analysis is to isolate a unique decomposition rule for any inequality
measure. While this has the advantage that the relative contribution of any component
to overall inequality is independent of the choice of inequality measure, this analysis
does not help in narrowing down the class of permissible inequality indices.

34 This condition requires the existence of first- and second-order partial derivatives.
Ch. 12.: Inequality, Poverty and Welfare 615

Although there is no natural decomposition of the Gini index, it has remained

a popular choice in many applications. This is not surprising because it has many
advantages. In particular, the Gini index can accommodate nonpositive incomes, which
is a common feature of income-distribution data. This gives it a decided advantage over
the Atkinson and Generalized Entropy family since these are only defined for positive
incomes. Not surprisingly, there have also been attempts to provide a characterization
of the Gini index 35.
Building on the work of Weymark (1981), Donaldson and Weymark (1980) and
Bossert (1990) have characterized the class of single-series Gini social-evaluation
functions. They note that the EDEI function implied by the Gini index, given in
Equation (3.7), can also be written as3 6

VneN, Vx Dn , .E1 (x) - 1 (2i- )i (3.23)

i l(2i - 1)
Of course, the weight of (2i - 1) is essentially arbitrary. This suggests the class of
Generalized Gini social-evaluation functions, which correspond to the class of EDEI
functions of the form

VxN, ,
Vn = ) l(a7).i
i=
Vln c N Vdx C Dn, _g -( = 1 (). (3.24)
Ki = (ai)
S-concavity of the EDEI function implies that the weights a' satisfy the restriction
a, < a, '. The Generalized Gini functions are both linear homogeneous and
unit translatable. In other words, the corresponding social-evaluation functions are
distributionally homothetic, and can be used to yield both absolute and relative
inequality indices. Donaldson and Weymark (1980) define the class of single-series
Gini indices where the weights are independent of n. The class of inequality indices
derived from the single-series social-evaluation functions do not necessarily satisfy
Dalton's Principle of Population. However, Donaldson and Weymark show that the
subclass of single-parameterGini functions yield inequality indices which satisfy the
Principle of Population. The single-parameter family (also called S-Ginis by Donaldson
and Weymark) is obtained by the following specification of weights:
-
Vi, ai = i (i - 1) where 6 > 1. (3.25)

Notice that the Gini social-evaluation function is obtained when 6 = 2.

Bossert (1990) uses an interesting aggregation property to obtain a characterization
of this class of social-evaluation functions. The aggregation property requires that the
income distribution of the richest members of the society can be evaluated without

35 On the Gini index, see also Mehran (1976), Kakwani (1980a), Pyatt (1976) and Yitzhaki (1983).
36 This follows because the sum of the first n odd integers is n2.
616 3. Ourta

reference to the incomes of the remaining individuals. This property is defined formally
below.
Definition. The sequence of social-evaluation functions W"'},, satisfies the High
Income Aggregation Property (HIGAP) iff the corresponding sequence {('",,\
satisfies the following:

Vn E N, Vx E D( +l ), -'T'`1)(x) = O("(l.... ,. r),, +I) (3.26)

for some function 0: R9+x i+ - 9,.

Bossert (1990) proves the following
Theorem 3.5. A sequence of social-evaluationfunctions { W" },, v satisfies HIGAP
S-concavity, distributionalhomotheticity and the Principleof Populationiff { W" },, ,v
is a sequence of S-Ginis.

3.4. Measuring mobility

In the preceding sections, we have discussed various issues concerned with the
evaluation of inequality from a static perspective. However, relative incomes, absolute
income differences as well as individuals' ranks in the income distribution change over
time, and social welfare clearly depends on the dynamics of income distribution. The
literature on the measurement of income mobility is essentially concerned with changes
in inequality of incomes over time. In this section, we review some of the main issues
concerned with measuring income mobility. The reader is referred to Fields and Ok
(1999) for a comprehensive and insightful survey of the literature.
From a purely descriptive perspective, the existence of income mobility between
two points of time creates a distance between the corresponding income distributions.
Although there has been some work on the measurement of distance between
income distributions 37, caution has to be exercised before the "distance" between
income distributions can be used as a measure of the extent of income mobility.
These measures are essentially designed to measure the difference in social welfare
between two distributions, and hence incorporate anonymity. Consider, for instance,
E(y)l, which is the measure of distance between distributions x and y
-(x)-

suggested by Chakravarty and Dutta (1987). Suppose x = (1,0) and y = (0,1).

Then, the "distance" between x and y will be recorded as 0 by the Chakravarty
Dutta measure, and indeed by most measures designed to measure the distance between
income distributions from a (static) welfare perspective. On the other hand, there is an
obvious sense in which the distributions x and y exhibit a great deal of mobility - the

37 See Dagum (1980), Shorrocks (1982b), Ebert (1984) and Chakravarty and Dutta (1987) for the
construction of distance functions between income distributions.
Ch. 12: Inequality, Pouerty and Welfare 617

person who is poor in distribution x occupies the position of the rich in distribution y
and vice versa.
The problem arises because of the implicit use of anonymity. While this makes sense
from a static welfare perspective, any measure of income mobility must recognize that
changes in individual rank orders are an intrinsic feature of income mobility. One
way in which changes in individual rank orders can be taken into account is to define
the distance between two distributions to depend upon changes in individual incomes.
This is the approach adopted by Fields and Ok (1996) and Mitra and Ok (1998), who
specify sets of axioms in order to single out indices of mobility. For instance, Mitra
and Ok (1998) specify that the distance d,(x,y) between two distributions x and y
in D" is a strictly increasing, symmetric and continuous function of the individual
distances (dl(xl, yl),... , dl(x,,, y,)). This Decomposability axiom, along with some
other axioms, enables them to deduce that the mobility index is represented by the
distance function d:.9 where d,,(x,y) = y (n=
(+, xi-yla) for some
a > 1, y > 0.
Of course, there is some "overkill" here in the sense that the index of mobility will
be positive even when there is only a change in the scale of the income distributions
without any change in relative incomes. Perhaps, that is why Fields and Ok (1999)
refer to these as absolute measures of mobility.
Another descriptive approach towards the measurement of mobility is to use
transition matrices. A typical element pij of a transition matrix P = [pi] C 9+xm is
the probability of moving from income state (or class) i to income state j, and where
there are m states. Let P be the set of such transition matrices. Shorrocks (1978a)
defines an index of mobility as a continuous function M(.) on the set of transition
matrices. Shorrocks imposes the following axioms on the mobility index M:

Normalization (N):
0 < M(P) < 1 for all P e .

Monotonicity (M):
M(P) > M(P') if P >-P', where P - P' if
pi > p for all i j with at least one strict inequality.
Perfect Mobility (PM):
M(P) = I if P = ux' where u = (1,... ,1).

Thus, PM states that the mobility index should take the highest value when the
probability of moving to any income state is independent of the original state.
Although these axioms are all reasonable, Shorrocks uses a simple example to show
that they are mutually incompatible. Thus, let m = 2, and choose P = [pij] such that
Pij = for all i,j. Also, let P' = [p be such that pii = 0 and pij = 1 when i j.
Then, N and PM imply M(P)= 1 , but M(P') > M(P) from M.
618 B. Dutt

Fortunately, Shorrocks argues that transition matrices such as P' are unlikely to arise
in practice, because the higher values tend to cluster around the main diagonal in
observed transition matrices. Let P* be the subset of transition matrices such that the
probability of remaining in the same category is at least as high as that of transferring
to any other group. Then, the axioms N, PM and M are compatible on P*.
Apart from the literature analyzing income mobility from a descriptive perspective,
there is also a substantial volume of work studying the measurement of income mo-
bility in terms of its consequences for social welfare. Thus, Chakravarty, Dutta and
Weymark (1985) (henceforth CDW) advocate a concept of welfare which involves a
comparison between the time paths of incomes received over a number of periods
with the hypothetical-time paths of incomes obtained by supposing that starting
from the actual first-period income, the remaining income receipts exhibit complete
immobility. CDW consider distributions over a time interval [to, t,,) partitioned
into m equal subperiods [tk.,tk l). The income distribution in period k is denoted
yk = (y ...... ynk) Sequences of income distributions, denoted Y = (y',...,y')
are called income structures. Let y = lyi represent the aggregate income
y..d=
received by person i over the time period, and let ya = (ya, ... yn) be the aggregate
distribution. The benchmark structure Yb is the structure which results when income
shares are maintained through time starting from the actual first period distribution.
Thus, YRb =[ylyl ' ..I) Y ()
A mobility index assigns a numerical value to each income structure. CDW
assume the existence of an intertemporal social welfare function defined on all
income structures. The mobility index is obtained by comparing the levels of welfare
associated with Y and Yb. They also assume that the only features of the income
structures which are relevant for welfare comparisons are the corresponding aggregate
distributions ya and yb. This yields a general mobility index of the form

0 (Y(a)
M(Y) (3.27)
(y }'
where 0: 9++, 9 is a continuous, increasing function with ¢(1) = 0. This procedure
defines a class of mobility indices. Among members of this class, one index has a
simple interpretation. This is obtained by setting ¢(s) = s - 1, so that

E(ya)
M(Y)-= _(b ) (3.28)

This index measures mobility as the percentage change in EDEI of the actual aggregate
distribution compared with what it would have been with the immobile structure yb.
There have been other attempts to construct normative indices of mobility. For
instance, King (1983) constructs an index which measures changes in the rank orders of
the income distribution. King essentially constructs a hypothetical benchmark structure
Y, = (y*l,y*2) from the actual structure Y = (yi,y 2 ). He does so by specifying
Ch. 12: Inequality, Poverty and Welfare 619

y*~ = yl, while y*2 is the element in {y2,.. ,y,2} which person i would obtain if
his rank order in the income distribution did not change from y'. This hypothetical
structure is now used to define a scaled order statistic

Si = ly2 2 ) , i= 1,... ,n. (3.29)

(1(y )

King defines a social welfare function on the final distribution y 2 and the vector of
scaled order statistics s, F(y 2 , s), with F increasing in all of its arguments. Analogous
to the concept of the EDEI, he defines a "zero mobility equivalent proportion of
income P", defined implicitly by

F(/y2 , 0) = F(y 2, s) (3.30)

The increasingness of F implies that f > 1. King's index of mobility is given by

MK(Y) = 1 -. (3.31)

So, this index measures the "proportion of total income which from a position of zero
mobility, we would be prepared to sacrifice in order to achieve the degree of mobility
we observe ... " [King (1983), p. 109].
Chakravarty (1984) contains a generalization of King's index. Other papers which
also measure income mobility from a normative perspective include Atkinson (1981),
Markandya (1982, 1984), Kanbur and Stiglitz (1986), Slesnick (1986), Shorrocks
(1978b) and Dardanoni (1991, 1993, 1995).

4. Measurement of poverty

Sen (1976b) initiated the theoretical literature on the measurement of poverty. Sen
pointed out that the usual practice of simply counting the number of the poor or even
the use of measures which are based only on the aggregate incomes of the poor are
deficient because they ignore the income distribution of the poor 38. He went on to
derive a specific measure of poverty based on a set of axioms. Following the work
of Sen, subsequent authors have explored alternative approaches and evolved a large
number of different measures of poverty.
Sen (1976b) pointed out that there are two distinct stages in the measurement
of poverty. The first stage is the identification of who are the poor. Once the poor
have been identified, the second step consists in aggregating the incomes of the

38 Watts (1968) was actually the first to criticize the practice of simply counting the poor, remarking
that it had "little but its simplicity to recommend it".
620 B. Dutta

poor to arrive at a scalar measure of poverty. The identification exercise involves

the specification of a poverty-line level of income or consumption expenditure. The
obvious interpretation of this line is that anyone with an income or consumption below
the stipulated level is deemed to be poor. While the specification of the poverty line
is an essential ingredient in any attempt to measure the extent of poverty, there
is considerable scope for difference in views about the appropriate choice of the
poverty line3 9 . In the absence of any unambiguous objective criterion, an obvious
alternative is to conduct the measurement exercise using a range of values for the
poverty line. This is the approach explored by Atkinson (1987) and Foster and
Shorrocks (1988a,b). We will discuss these important papers later on.
Most of the literature has focused on the aggregation exercise. Perhaps the most
widely used measure of poverty is the Head-count ratio. Let z be the poverty line,
and recall that q(x; z) is the number of individual incomes in x which do not exceed z.
Then, the Head-count ratio H is simply the fraction of people whose incomes do not
exceed z:
q (4.1)
H(x;z) -
n

where q = q(x; z) and n is the number of people in the population. We will use Q(x) to
denote the set of people who are poor, dropping z from the notation since there will
be no confusion about the poverty line.
It is also common to measure poverty by the extent of shortfall of incomes of the
poor from the poverty-line income z. The income gap z - xi reflects the depth of a
person's poverty. One measure which reflects the depth of poverty in the population is
the average income gap. Thus, letting gi(x, z) = z xi, we have

I(x,z) = i CQ(x) gi(x) (4.2)

qz

However, both H and I are at best partial indicators of poverty. The Head-count ratio
ignores both the depth as well as the distribution of incomes amongst the poor, while
I ignores both the number of people who are poor as well as the distribution of incomes
of the poor.
The Principle of Transfers has come to be considered a minimal requirement which
must be satisfied by any sensible measure of inequality. Are there similar axioms which
can help to weed out poverty measures such as H and I? The following axioms have
come to be viewed as the minimal requirements for any sensible measure of poverty.
Definition. A poverty measure P satisfies Monotonicity if P(x; z) > P(y; z) whenever
xi < yi for some i C Q(x) and xi = yi for all other i E N.

39 See Ravallion (1994) for an illuminating discussion of various issues connected with the specification
of the poverty line. See also Kakwani (1984) and Zheng (1997).
Ch. 12: Inequality, Poverty and Welfare 621

Definition. A poverty measure P satisfies the Weak Transfer Axiom if P(x; z) > P(y; z)
whenever y is obtained from x through a progressive transfer involving only the poor.
The Monotonicity Axiom says that other things remaining the same, a reduction in
income of a person below the poverty line must increase the level of poverty. The
Weak Transfer Axiom requires the poverty measure to be sensitive to the distribution
of income amongst the poor. Other things remaining the same, poverty must increase if
the distribution of income amongst the poor worsens. Neither of the measures I and H
satisfies the Weak Transfer Axiom. As far as the Monotonicity Axiom is concerned,
H violates it since it is completely insensitive to anything other than the number of the
poor. Of course, I and other measures which are based on the income gap do satisfy
the Monotonicity Axiom.
Sen ( 1976b) had actually suggested a stronger version of the Weak Transfer Axiom.

Definition. A poverty measure P satisfies Sen s Transfer Axiom if other things

remaining the same, poverty goes up whenever there is a pure transfer of income from
a person below the poverty line to someone who is richer.
Although the two axioms are similar in spirit, Sen's Transfer Axiom is stronger because
it requires that poverty go up even if the number of poor come down as a result of
the regressive transfer.
Sen proposed the following general form for poverty measures:

P(x;z) = A(x;z) E gi(x;z) vi(x;z). (4.3)

i Q(x)

Equation (4.3) makes poverty a normalized weighted sum of individual income gaps.
When expressed in this form, the poverty measure will satisfy Monotonicity provided
all the weights are positive, while the Weak Transfer Axiom will be satisfied if
vi(x;z) > vj(x;z) whenever xi < xj. Out of the large number of weighting schemes
which satisfy these properties, Sen selected one where vi depended on individual i's
rank in the income distribution. Underlying this was the idea of relative deprivation -
the deprivation experienced by individual i is greater the larger is the number of poor
people with higher incomes. Thus, the poorest person gets a weight of q(x; z).

Definition. A poverty measure of the form given by Equation (4.3) satisfies

Ranked Relative Deprivation if vi(x;z) equals the rank order of individual i in the
income distribution of the poor.
In order to fix the normalization term A(x; z), Sen argued that the two measures H and I
capture all relevant aspects of poverty in the special case when all poor persons have
the same income - that is when there is absolutely no inequality in the distribution
622 B. Dutta

of income amongst the poor. In this case, he set P(x;z) = H(x;z)l(x;z) 4 0 .

These
specifications yield Sen's poverty measure:

S(X; Z) = 2 Ei c Q(x)g(x; z) ri(x; z)

(q +l)nz

where ri(x;z) is a ranking of the poor associated with the distribution x and the
poverty line z. When q, the number of the poor, is large, Sen shows that Equation (4.4)
can be written as

S(x; z) = H[I +(1 - I) G] (4.5)

where Gp is the Gini measure of inequality of the income distribution of the poor.
Equation (4.5) shows explicitly the precise way in which distributional sensitivity
is incorporated into Sen's measure of poverty.
Apart from the Monotonicity and Weak Transfer Axioms 41, S also satisfies a number
of other axioms, some of these being familiar from the earlier discussion on inequality.
It is clearly symmetric in individual incomes, and satisfies the Population Principle
since there will be no change in the level of poverty if a distribution is replicated.
S is also a relative index since the measure is homogeneous of degree zero in individual
incomes and the poverty line. It also satisfies the so-called Focus Axiom, which
requires that the level of poverty be completely insensitive to any change in the level
of incomes of the non-poor 42 .
Various alternative measures have been proposed in the subsequent literature. Anand
(1977, 1983) and Blackorby and Donaldson (1980b) noted that other measures of
inequality could be used to replace the Gini coefficient in the Sen index. Thus, let Ip
be any relative index of inequality, restricted to the income distribution of the poor.
Blackorby and Donaldson described a class of relative poverty indices P,.(x; z):

Pr(x;z)= H(I +( I) p). (4.6)

Blackorby and Donaldson provide an intuitive explanation of Pr. Since I = [z - (xp)]/z,

where Y(xp) is the mean income of the poor, Pr can be expressed in the alternate form

PH [Z - (p)] (4.7)

where Xp is the income distribution restricted to the poor. Hence, Pr is the product
of the Head-count ratio and an "average" income gap where the average is computed

40 Of course, there is no logical reason to single out the product form even in this special case.
41 Sen's measure does not satisfy the stronger version of the Transfer Axiom.
42 See Donaldson and Weymark (1986) for a discussion of the properties of a poverty index.
Ch. 12: Inequality, Poverty and Welfare 623

by using the EDEI of the poor rather than their mean income. Chakravarty (1983b)
introduces related measures which depend on "representative" income gaps.
Blackorby and Donaldson (1980b) also derive a class of absolute measures of
poverty by using absolute rather than relative inequality indices in Equation (4.6). An
absolute index of poverty remains invariant if the poverty line and incomes of the poor
are changed by the same absolute amount.
However, despite the intuitive appeal of the Blackorby and Donaldson approach,
some members of the class P, (as well as members of the corresponding class of
absolute poverty indices) may not satisfy the Monotonicity Axiom 4 3 .
An alternative method of obtaining poverty measures from inequality measures
was proposed by Takayama (1979), who defines the notion of censored income
distributions. These are obtained by replacing the incomes of the non-poor in a
distribution x by the poverty-line income z. Any relative inequality measure applied to
the censored distribution will yield a relative measure of poverty. However, Takayama
himself points out that the use of the Gini coefficient on the censored distribution will
yield a poverty measure which does not satisfy Monotonicity.
However, the censored income distribution has proved to be a useful device in the
construction of poverty measures. Chakravarty (1983a) proposes an interesting use
of censored income distributions 4 4 . Letting _(x) denote the EDEI corresponding to
the censored income distribution i, he defines the level of poverty to be equal to the
average income gap, where the average is computed by using the EDEI of the censored
income distribution. Formally,

Pc(x;z)= - ) (4.8)
z
Notice that Pc is very similar to the approach of Blackorby and Donaldson (1980b).
However, Pc always satisfies Monotonicity since the EDEI function is increasing in
individual incomes. This measure also satisfies the Weak Transfer Axiom since the
EDEI function is S-concave.
Some papers have also tried to derive "improved" versions of Sen's measure. For
instance, Thon (1979) views the failure of S to satisfy the Sen Transfer Axiom as
a serious shortcoming. Thon modifies the weighting scheme underlying S by adding
(n - q) to each weight ri(x; z). In other words, the weight attached to the income gap
of any poor person is his rank among all persons. Thon's measure may be written in
the following way:
q+l 2(n - q)
PT(X; Z)= q S(x;z)+ HI. (4.9)
PT now satisfies the stronger form of the Transfer Axiom. However, there has been
some debate 45 about whether this version of the Transfer Axiom is as compelling as the

43 See Foster (1984) for specific counterexamples.

44 See also Clark, Hemming and Ulph (1981).
45 See Sen (1979), Foster (1984) and Kundu and Smith (1983).
624 B. Dutla

weaker version. After all, a poverty measure cannot be completely unconcerned about
the number of the poor. If a regressive transfer reduces the number of the poor, then
there is a trade-off between the increase in poverty due to the worsening distribution
and a reduction in poverty due to the smaller number of the poor. Different measures
may well evaluate the trade-off in different ways 46 . In fact, Kundu and Smith (1983)
show that if a measure is sufficiently responsive to the number of poor and non-poor,
then it cannot satisfy the Sen Transfer Axiom.
Perhaps the most interesting modification of Sen's measure is due to Shorrocks
(1995). Shorrocks points out that S also violates continuity. He views the violation of
continuity as a serious problem since a discontinuous measure is particularly sensitive
to measurement errors, and proceeds to show that both shortcomings (the violations
of continuity and the Transfer Axiom) can be traced to the form of normalization
employed by Sen. Shorrocks requires the normalization employed by Sen to hold only
when everyone is poor. That is, Shorrocks employs the restriction that P(x, z) = I(x, z)
only when H(x,z) = I and Gp = 0. Formally, Shorrocks' measure is

1 g(x; z)
Ps(x;z)= 2 E (2n -2i+ )g (4.10)
i E Q(x)

Shorrocks also shows that PS has an attractive geometric interpretation. Let

x' = (x,...., x, ) be the vector of what Shorrocks calls normalizedpoverty gaps. That
is, x = max{ gi(x, z)/z, 0}. Then, Ps can be written as

Ps(x;z) = "(x*)[ + G(x*)] (4.11)

where G is the Gini coefficient.

Now, one can define the normalizedpoverty gap profile D(x*, ), a concept which
plays a role that is analogous to that of the Lorenz curve in inequality measurement 47.
D(x, p) is simply the sum of normalized poverty gaps of the I 00p% poorest individuals,
and Ps turns out to represent twice the area beneath the normalized poverty-gap profile
for x*. An alternative interpretation uses the line of maximum poverty, the latter being
obtained when all incomes are zero. Then, Ps is seen to be the area below the poverty-
gap profile expressed as a proportion of the area below the line of maximum poverty.
Chakravarty (1997) provides another interesting interpretation of Ps. He shows that
Ps is a particular member of Pc if the social evaluation is in terms of the Gini social
welfare function.

46 Of course, the Weak Transfer Axiom is compelling precisely because it is not confronted with any
such trade-off. See Hagenaars (1987) for an analysis of the nature of the trade-offs involved.
47 This issue is discussed later on page 627.
Ch. 12: Inequality, Poverty and Welfare 625

Kakwani (1980b) proposes a class of measures whose elements are obtained by

raising each weight ri(x;z) to a power k > 0, and renormalizing. The family of
measures is
(x;z)= q gi(x;z)(r(x;z)) (4.12)
nzk( i EQ(x)
where Ok (q) = e Q(x) ik
Ci
When k = 0, Kakwani's measure is simply HI. The Sen Measure is obtained when
k = 1. It satisfies the Monotonocity and Weak Transfer Axioms for all k > 0. The
interesting feature of the Kakwani class is that for k > 1, the measures satisfy a form
of transfer sensitivity. The form of transfer sensitivity described by Kakwani requires
that a transfer of a fixed amount of income between two poor persons must have a
larger effect on poverty the lower is the income ranking of the two persons.
Notice that this form of transfer sensitivity is different from that of Shorrocks and
Foster (1987) discussed earlier in the context of inequality measurement, although
the underlying motivation between the two forms is very similar. The counterpart of
the transfer sensitivity specified by Foster and Shorrocks essentially requires that a
transfer of a fixed amount of income between two poor persons must have a larger
effect the lower the incomes of the pair. Although Kakwani claims that there will
be some k for which the corresponding Pk will satisfy the latter type of transfer
sensitivity, this is not really satisfactory since the choice of k will depend upon the
specific income distribution. Clark et al. (1981) point out that any form of rank-order
weighting scheme will suffer from this defect, and suggest an alternative approach.
They advocate the use of deprivationfunctions which are social-evaluation functions
for individual poverty gaps. The inequality in poverty gaps is then assumed to
reflect the aggregate relative deprivation or poverty. Formally, they specify identical-
deprivation functions of the form

d(gi(x; z)) = [gi(x; z)]a. (4.13)

Assuming a social welfare function which is decreasing and additively separable in
the individual-deprivation functions, they derive the "equally distributed equivalent
poverty gap". This is the poverty gap which if shared by all the poor would give the
existing level of social welfare. This is given by

g*(x;z)= IQ(x)l E ) x ] (4.14)

The measure of poverty is then given by

qg*(x; z)
PCHU(X;Z) = (4.15)
nz
This measure clearly satisfies Monotonicity. Notice that when a > 1, the measure will
be strictly convex in incomes of the poor, and will exhibit transfer sensitivity.
626 B. Dutta

There have also been attempts to derive measures of poverty which are additively
decomposable using population weights. The principal contribution is by Foster, Greer
and Thorbecke (1984), who propose the following class of measures:

PFGT(X;Z) - n [g,(x;Z)] (4.16)

i Q(x)

for a > 0. Note that for a = 0, the measure becomes H. At a = , the measure
4
coincides with HI, while for a > 2, the measures satisfy transfer sensitivity .
Subgroup consistency is a condition which is weaker than decomposability.
It requires that overall poverty should fall when there is a reduction in the
poverty level of a subgroup while the poverty level of all other groups remains
unchanged 49. Foster and Shorrocks (1991) characterize the class of subgroup-
consistent poverty indices satisfying some basic properties such as Symmetry, the
Population Principle, Monotonicity, the Focus Axiom and a Restricted Continuity
condition which requires that the poverty index must be a continuous function of
poor incomes 50 .
Foster and Shorrocks show that weakening the consistency requirement to subgroup
consistency from decomposability does not really increase the set of available
poverty measures. A subgroup-consistent index is essentially a convex or lexical
combination of the Head-count ratio and a decomposable index. Of course, there are
measures which are subgroup-consistent but not decomposable 51 . Familiar measures
such as S, I and the Blackorby-Donaldson class are not subgroup-consistent.
The preceding discussion indicates that there is a rather large number of poverty
measures satisfying the two basic axioms of Monotonicity and Weak Transfer property.
Clearly, there is considerable scope for conflicting evaluations of the extent of poverty
in two different income distributions. Indeed, the conflict may be more severe for
poverty rankings than for inequality rankings because there are now two different
sources of disagreement. First, since the choice of the poverty line is largely subjective,
there may be differences in evaluation, euen if the same measure is used to aggregate
the incomes of the poor to yield a scalar measure of poverty. Second, even if there is
agreement about the poverty line, different functional forms for the poverty index will
also yield different evaluations.
Foster and Shorrocks (1988a) 5 2 derive conditions under which unambiguous
poverty judgements can be made when there is disagreement about the appropriate
level of the poverty line, assuming that the index of poverty belongs to the PFGT

4Y Note that these are the continuous versions of the PCIIL measures.
49 Note that the condition applies only when the subgroup poverty levels remain constant.
50 Notice that measures such as S and H, which change discontinuously when a person's income crosses
the poverty line, fail to satisfy continuity on the entire domain of incomes.
51 A family of indices proposed by Clark, Hemming and Ulph (1981) is an example of this class.
52 See also Foster and Shorrocks (1988b,c).
Ch. 12: Inequality, Poverty and Welfare 627

class. Let Z denote the range of possible values of the poverty line z. Let xP(Z)y
be true if P(x; z) < P(y; z) for all z E Z, with strict inequality for some z e Z. Foster
and Shorrocks show that if Z = 9, then for any positive integer a, xPFGT(Z)y iff
x a-stochastically dominates y. However, since it is reasonable to place some upper
bound on the poverty line, it is unrealistic to assume that Z = 9. On the other hand,
if Z = (0, z*) for some z*, then the shape of the distribution above z* cannot be relevant
for poverty judgements according to any poverty index satisfying the Focus Axiom.
This suggests that stochastic dominance of the censored distributions corresponding
to z* provides the appropriate condition for unambiguous poverty judgements. Foster
and Shorrocks (1988b) prove this equivalence for all positive integers a.
Atkinson (1987) extends the scope of the dominance result considerably. Let i and y
represent the censored income distributions corresponding to the poverty line z*, where
Z = (0,z*]. Atkinson shows that second-order stochastic dominance of i over is
equivalent to x P(Z)y for all continuous poverty measures satisfying Symmetry, Focus,
Monotonicity, the Sen Transfer Axiom and the Population Principle 53 .
Thus, Atkinson's dominance result allows for variations in both the specifications
of the poverty line as well as in how the poverty gaps (given a specific poverty line)
are aggregated to obtain a measure of poverty. The poverty-gap profile can be used
as a geometric representation of unambiguous poverty rankings, once a poverty line
is specified. Let us say that the distribution x P-dominates y if D(x,p) > D(y,p)
for all p C [0, 1], with strict inequality for some p. So, if x P-dominates y, then the
cumulative normalized poverty gap corresponding to x is higher than that ofy for every
value of p. Then, x will be judged to have more poverty than y by all poverty indices
that are symmetric additive functions of increasing, strictly convex transformations of
individual normalized poverty gaps 54. Correspondingly, if the poverty-gap profiles of
x and y intersect, then one can obtain a pair of poverty measures in this class which
differ in their rankings of the two distributions. The analogy with Lorenz domination
and unambiguous inequality rankings is obvious.

5. Concluding remarks

The main focus of this chapter has been on ethical measurement of inequality
and poverty. Section 3.1 describes the methodology underlying the construction of
inequality indices which are nornatiuely significant - these are indices for which
higher inequality is associated with lower social welfare whenever two distributions
have the same mean income. The two principal classes of normatively significant
indices are the AKS indices, which measure the percentage of total income which can

53 The Transfer Axiom can be dropped from this list in the case of first-order stochastic dominance.
54 This result and variations are due to Spencer and Fisher (1992), Jenkins and Lambert (1997, 1998)
and Shorrocks (1998).
628 B. Dutta

be discarded without reducing social welfare, and the Kolm indices, which measure
the absolute amount of income per person that can be saved without affecting social
welfare.
In general, unambiguous judgements on inequality cannot be made because the
rankings of income distributions produced by different inequality indices will not
coincide. Section 3.2 describes an alternative approach which declares one distribution
to be more unequal than another only when all "sensible" inequality indices agree
on the ranking of these distributions. This "dominance" approach relies on the
close connection between Lorenz curves, stochastic dominance and the principle of
progressive transfers which requires inequality to decrease if income is transferred
from a richer to a poorer person. Section 3.2 discusses several equivalence results
which provide the analytical foundations of the dominance approach.
Section 3.3 presents some results which axiomatically characterize different inequal-
ity indices. In particular, it focuses on the implications of decomposability, as well as
on characterization of the Gini index of inequality.
Section 3.4 is concerned with the measurement of income mobility, that is how
inequality moves over time. It discusses both the positive or descriptive measures, as
well as the recent substantial literature studying the measurement of income mobility
from the perspective of social welfare.
Finally, issues connected with the ethical measurement of poverty have been
discussed in Section 4. Assuming that the exercise of identifying the poor has been
completed, the section focuses on different ways of aggregating the incomes of the
poor so as to derive an index of poverty.

References

Anand, S. (1977), "Aspects of poverty in Malaysia", Review of Income and Wealth 23:1-16.
Anand, S. (1983), Inequality and Poverty in Malaysia (Oxford University Press, London).
Atkinson, A.B. (1970), "On the measurement of inequality", Journal of Economic Theory 2:244 263.
Atkinson, A.B. (1981), "The measurement of economic mobility", in: P.J. Eggelshoven and L.J. van
Gemerden, eds., Inkomens Verdeling en Openbare Financien (Het Spectrum, Leiden).
Atkinson, A.B. (1987), "On the measurement of poverty", Econometrica 55:749-764.
Atkinson, A.B., and E Bourguignon (1982), "The comparison of multi-dimensioned distributions of
economic status", Review of Economic Studies 49:183 201.
Basmann, R., and G. Rhodes, eds. (1984), Advances in Econometrics, Vol. 3 (JAI Press, Greenwich).
Bawa, V. (1976), "Optimal rules for ordering uncertain prospects", Journal of Financial Economics
2:95-121.
Bhattacharya, N., and B. Mahalanobis (967), "Regional disparities in household consumption in India",
Journal of the American Statistical Association 62:143-161.
Blackorby, C., and D. Donaldson (1978), "Measures of relative equality and their meaning in terms of
social welfare", Journal of Economic Theory 18:59-80.
Blackorby, C., and D. Donaldson (1980a), "A theoretical treatment of indices of absolute inequality",
International Economic Review 21:107-136.
Blackorby, C., and D. Donaldson (1980b), "Ethical indices for the measurement of poverty", Econometrica
48:1053-1060.
Ch. 12: Inequality, Pooerty and Welfare 629

Blackorby, C., and D. Donaldson (1984), "Ethically significant ordinal indexes of relative inequality",
in: R. Basmann and G. Rhodes, eds., Advances in Econometrics, Vol. 3 (JAI Press, Greenwich)
pp. 131-147.
Blackorby, C., D. Donaldson and M. Auersperg (1981), "A new procedure for the measurement of
inequality within and among populaiton subgroups", Canadian Journal of Economics 14:665-685.
Blackorby, C., W. Bossert and D. Donaldson (1999), "Income inequality measurement: the normative
approach", in: J. Silber, ed., Handbook of Income Inequality Measurement (Kluwer Academic
Publishers, Boston) pp. 133-157.
Bossert, W. (1990), "An axiomatization of the single-series Ginis", Journal of Economic Theory 50:
82-92.
Bossert, W., and A. Pfingsten (1990), "Intermediate inequality: concepts, indices and welfare
implications", Mathematical Social Sciences 19:117-134.
Bourguignon, E (1979), "Decomposable income inequality measures", Econometrica 47:901 920.
Chakravarty, S.R. (1983a), "Ethically flexible measures of poverty", Canadian Journal of Economics
16:74-85.
Chakravarty, S.R. (1983b), "Measures of poverty based on the representative income gap", Sankhy,
The Indian Journal of Statistics, Series B 45:69-74.
Chakravarty, S.R. (1984), "Normative indices for measuring social mobility", Economics Letters 15:
175-180.
Chakravarty, S.R. (1988), "Extended Gini indexes of inequality", International Economic Review 29:
147-156.
Chakravarty, S.R. (1990), Ethical Social Index Numbers (Springer, Berlin).
Chakravarty, S.R. (1997), "On Shorrocks' reinvestigation of the Sen poverty index", Econometrica
65:1241-1242.
Chakravarty, S.R., and B. Dutta (1987), "A note on measures of distance between income distributions",
Journal of Economic Theory 41:185-188.
Chakravarty, S.R., and D. Mukhejee (1999), "Measures of deprivation and their meaning in terms of
social satisfaction", Theory and Decision 47:89 100.
Chakravarty, S.R., B. Dutta and J. Weymark (1985), "Ethical indices of income mobility", Social Choice
and Welfare 2:1-21.
Clark, S., R. Hemming and D. Ulph (1981), "On indices for the measurement of poverty", Economic
Journal 91:515-526.
Cowell, EA. (1980), "On the structure of additive inequality measures", Review of Economic Studies
47:521 531.
Cowell, EA. (1988), "Inequality decomposition: three bad measures", Bulletin of Economic Research
40:309-312.
Cowell, EA. (1995), Measuring Inequality, 2nd Edition (Prentice-Hall/Harvester, London).
Cowell, EA., and K. Kuga (1981a), "Inequality measurement: an axiomatic approach", European
Economic Review 15:287-305.
Cowell, EA., and K. Kuga (1981b), "Additivity and the entropy concept: an axiomatic approach to
inequality measurement", Journal of Economic Theory 25:131 143.
Creedy, J. (1977), "The principle of transfers and the variance of logarithms", Oxford Bulletin of
Economics and Statistics 39:153 158.
Dagum, C. (1980), "Inequality measures between income distributions with applications", Econometrica
48:1791-1803.
Dalton, H. (1920), "The measurement of the inequality of incomes", Economic Journal 30:348-361.
Dardanoni, V (1991), " On the Lorenz curve ordering of discounted streams of income", Bulletin of
Economic Research 43:293-296.
Dardanoni, V. (1993), "Measuring social mobility", Journal of Economic Theory 61:372-394.
Dardanoni, V (1995), "Monotone mobility matrices and income distribution dynamics", Social Choice
and Welfare 12:181-192.
630 B. Dutita

Dasgupta, P., A.K. Sen and D. Starrett (1973), "Notes on the measurement of inequality", Journal of
Economic Theory 6:180-187.
Davies, J., and M. Hoy (1995), "Making inequality comparisons when lorenz curves intersect", American
Economic Review 85:980-986.
Donaldson, D., and J.A. Weymark (1980), "A single-parameter generalization of the Gini indices of
inequality", Journal of Economic Theory 22:67-86.
Donaldson, D., and J.A. Weymark (1986). "Properties of fixed population poverty indices", International
Economic Review 27:667 688.
Dutta, B., and J. Esteban (1992), "Social welfare and equality". Social Choice and Welfare 9:267-276.
Ebert, U. (1984), "Measures of distance between income distributions", Journal of Economic Theory
32:266-274.
Ebert, U. (1987). "Size and distribution of incomes as determinants of social welfare", Journal of
Economic Theory 41:23-33.
Ebert, U. (1988). "On the decomposition of inequality: partitions into non-overlapping subgroups", in:
W. Eichhorn, ed., Measurement in Economics (Physica-Verlag, New York) pp. 399-412.
Eichhom, W., ed. (1988), Measurement in Economics (Physica-Verlag, New York).
Eichhom, W., and W. Gehrig (1982), "Measurement of inequality in economics", in: B. Korte, ed..
Modern Applied Mathematics: Optimization and Operations Research (North-Holland, Amsterdam)
pp. 657-693.
Fei, J.C.S., G. Ranis and S.WY. Kuo (1978). 'Growth and the family distribution of income by factor
components", Quarterly Journal of Economics 92:17-53.
Fields, G.S. (1980), Poverty, Inequality and Development (Cambridge University Press. Cambridge).
Fields, G.S., and J.C.S. Fei (1978), "On inequality comparisons". Econometrica 46:303-316.
Fields, G.S., and E. Ok (1996), "The meaning and measurement of income mobility", Journal of
Economic Theory 71:349-377.
Fields, G.S., and E. Ok (1999), "The measurement of income mobility : an introduction to the literature",
in: J. Silber, ed., Handbook of Income Inequality Measurement (Kluwer Academic Publishers, Boston)
pp. 557-596.
Foster, J.E. (1983), "An axiomatic characterization of the Theil measure of income inequality", Journal
of Economic Theory 31:105-121.
Foster, J.E. (1984), "On economic poverty: a survey of aggregate measures", in: R. Basmann and
G. Rhodes, eds., Advances in Econometrics, Vol. 3 (JAI Press, Greenwich).
Foster, J.E. (1985), "Inequality measurement", in: H.P Young, ed., Fair Allocation (American
Mathematical Society, Providence, RI).
Foster. J.E. (1994), "Normative measurement: is theory relevant?", The American Economic Review,
AEA Papers and Proceedings 84:365-370.
Foster, J.E., and E.A. Ok (1999), "Lorenz dominance and the variance of logarithms", Econometrica
67:901-907.
Foster, J.E., and A.K. Sen (1997), "On economic inequality after a quarter century", in: A.K. Sen, ed.,
On Economic Inequality, 2nd Edition (Clarendon Press, Oxford) pp. 107 219.
Foster, J.E., and A.A. Shneyerov (2000), "Path independent inequality measures", Journal of Economic
Theory 91:199-222.
Foster, J.E., and A.E Shorrocks (1988a), "Poverty orderings", Econometrica 56:173-177.
Foster, J.E., and A.E Shorrocks (1988b), "Poverty orderings and welfare dominance". Social Choice and
Welfare 5.:179-198.
Foster, J.E., and A.E Shorrocks (1988c), "Inequality and poverty orderings", European Economic Review
32:654-662.
Foster, J.E., and A.F. Shorrocks (1991), "Subgroup consistent poverty indices", Econometrica
59:687-709.
Foster, J.E., J. Greer and E. Thorbecke (1984), "A class of decomposable poverty measures", Econometrica
52:761-766.
Ch. 12: Inequality, Poverty and Welfare 631

Foster, J.E., M. Majumdar and T. Mitra (1990), "Inequality and welfare in market economies", Journal
of Public Economics 41:351-367.
Gastwirth, J.I. (1971), "A general definition of the Lorenz curve", Econometrica 39:1037-1039.
Hadar, J., and W. Russell (1969), "Rules for ordering uncertain prospects", The American Economic
Review 59:25-34.
Hagenaars, A. (1987), "A class of poverty indices", International Economic Review 28:583-607.
Hardy, G., J. Littlewood and G. Polya (1934), Inequalities (Cambridge University Press, London).
Jenkins, S.P., and P.J. Lambert (1997), "Three ""s of poverty curves, with an analysis of UK poverty
trends", Oxford Economic Papers 49:317-327.
Jenkins, S.P., and P.J. Lambert (1998), "Ranking poverty gap distributions: further TIPs for poverty
analysis", Research on Economic Inequality 8 (JAI Press, Greenwich, CT).
Kakwani, N.C. (1980a), Income Inequality and Poverty (Oxford University Press, New York).
Kakwani, N.C. (1980b), "On a class of poverty measures", Econometrica 48:437-446.
Kakwani, N.C. (1984), "Issues in measuring poverty", in: R. Basmann and G. Rhodes, eds., Advances
in Econometrics, Vol. 3 (JAI Press, Greenwich).
Kakwani, N.C., and N. Podder (1973), "On the estimation of the Lorenz curve from grouped observations",
International Economic Review 14:278-292.
Kanbur, S.M.R., and J. Stiglitz (1986), "Intergenerational mobility and dynastic inequality", Mimeo
(Princeton University).
King, M. (1983), "An index of inequality with applications to horizontal equity and social mobility",
Econometrica 51:99 116.
Kolm, S.-Ch. (1969), "The optimal production of social justice", in: J. Margolis and H. Guitton, eds.,
Public Economics (Macmillan, London) pp. 145-200.
Kolm, S.-Ch. (1976a), "Unequal inequalities I", Journal of Economic Theory 12:416-442.
Kolm, S.-Ch. (1976b), "Unequal inequalities II", Journal of Economic Theory 13:82-111.
Kolm, S.-Ch. (1977), "Multidimensional egalitarianisms", Quarterly Journal of Economics 91:1-13.
Krelle, W., and A.E Shorrocks, eds. (1978), Personal Income Distribution (North-Holland, Amsterdam).
Kundu, A., and T.E. Smith (1983), "An impossibility theorem for poverty indices", International Economic
Review 24:423-434.
Lambert, P.J. (1989), The Distribution and Redistribution of Income: A Mathematical Analysis (Basil
Blackwell, Oxford).
Lambert, PJ., and J.R. Aronson (1993), "Inequality decomposition analysis and the Gini coefficient
revisited", Economic Journal 103:1221-1227.
Love, R., and M.C. Wolfson (1976), Income Inequality: Statistical Methodology and Canadian Illustrations
(Statistics Canada, Ottawa).
Maasoumi, E. (1986), "The measurement and decomposition of multi-dimensional inequality",
Econometrica 54:991-997.
Maasoumi, E. (1989), "Continuously distributed attributes and measures of multivariate inequality",
Journal of Econometrics 42:131-144.
Markandya, A. (1982), "Intergenerational exchange mobility and economic welfare", European Economic
Review 17:307-324.
Markandya, A. (1984), "The welfare measurement of changes in economic mobility", Economica
51:457-471.
Marshall, A.W., and I. Olkin (1979), Inequalities: Theory of Majorization and Its Applications (Academic
Press, New York).
Mehran, E (1976), "Linear measures of income inequality", Econometrica 44:805-809.
Mitra, T., and E.A. Ok (1998), "The measurement of income mobility : a partial ordering approach",
Economic Theory 12:77 102.
Moyes, P. (1999), "Stochastic dominance and the Lorenz curve", in: J. Silber, ed., Handbook of Income
Inequality Measurement (Kluwer Academic Publishers, Boston) pp. 199-222.
632 B. Dutta

Newbery, D.M.G. (1970), "A theorem on the measurement of inequality", Journal of Economic Theory
2:264-266.
Pfingsten, A. (1986), "Distributionally-neutral tax changes for different inequality concepts". Journal of
Public Economics 30:385-393.
Pollak, R.A. (1971), "Additive utility functions and linear Engel curves", Review of Economic Studies
38:401-414.
Pyatt, G. (1976), "On the interpretation and disaggregation of Gini coefficients", Economic Journal
86:243-255.
Pyatt, G., C. Chen and J. Fei (1980), "The distribution of income by factor components", Quarterly
Journal of Economics 95:451-473.
Rao, VM. (1969), "Two decompositions of concentration ratio", Journal of the Royal Statistical Society
132:418 425.
Ravallion, M. (1994), Poverty Comparisons (Harwood. Chur).
Rothschild, M., and J.E. Stiglitz (1970), "Increasing risk: 1. A definition", Journal of Economic Theory
2:225-243.
Rothschild, M., and J.E. Stiglitz (1973), "Some further results on the measurement of inequality", Journal
of Economic Theory 6:188-204.
Runciman, W.G. (1966), Relative Deprivation and Social Justice (Routledge, London).
Russell, R.R. (1985), "A note on decomposable inequality measures", Review of Economic Studies
52:347-352.
Sen, A.K. (1973), On Economic Inequality (Clarendon Press. Oxford). 2nd Edition 1997.
Sen, A.K. (1976a), "Real national income", Review of Economic Studies 43:19-39.
Sen, A.K. (1976b), "Poverty: an ordinal approach to measurement", Econometrica 44:219-231.
Sen, A.K. (1978), "Ethical measurement of inequality: some difficulties", in: W. Krelle and A.F. Shorrocks,
eds., Personal Income Distribution (North-Holland, Amsterdam) pp. 81-94.
Sen, A.K. (1979), "Issues in the measurement of poverty", Scandinavian Journal of Economics 81:
285-307.
Sheshinski, E. (1972), "Relation between a social welfare function and the Gini index of income
inequality", Journal of Economic Theory 4:98-100.
Shorrocks, A.F. (1978a), "The measurement of mobility", Econometrica 46:1013-1024.
Shorrocks, A.E (1978b), "Income inequality and income mobility", Journal of Economic Theory 19:
376-393.
Shorrocks, A.E (1980), "The class of additively decomposable inequality measures", Econometnca
48:613-625.
Shorrocks, A.E (1982a), "Inequality decomposition by factor components", Econometrica 50:193-211.
Shorrocks, A.E (1982b), "On the distance between income distributions", Econometrica 50:1337 1339.
Shorrocks, A.E (1983), "Ranking income distributions", Economica 50:3-17.
Shorrocks, A.E (1984), "Inequality decomposition by population subgroups". Econometrica
52:1369-1385.
Shorrocks, A.E (1995), "Revisiting the Sen poverty index", Econometrica 63:1225-1230.
Shorrocks, A.E (1998), "Deprivation profiles and deprivation indices", in: S.P Jenkins, A. Kapteyn and
B. van Praag, eds., The Distribution of Welfare and Household Production: International Perspectives
(Cambridge University Press, Cambridge).
Shorrocks, A.F, and J.E. Foster (1987), "Transfer sensitive inequality measures", Review of Economic
Studies 54:485-497.
Silber, J. (1989), "Factor components, population subgroups and the computation of the Gini index of
inequality", Review of Economics and Statistics 71:107-115.
Silber, J., ed. (1999), Handbook of Income Inequality Measurement (Kluwer Academic Publishers,
Boston).
Slesnick, D.T. (1986), "Welfare distributional change and the measurement of social mobility", Review
of Economics and Statistics 68(4):586-593.
Ch. 12: Inequality, Poverty and Welfare 633

Spencer, B.D., and S. Fisher (1992), "On comparing distributions of poverty gaps", Sankha: The Indian
Journal of Statistics, Series B 54:114-126.
Takayama, N. (1979), "Poverty, income inequality, and their measures: Professor Sen's axiomatic approach
reconsidered", Econometrica 47:747-760.
Theil, H. (1967), Economics and Information Theory (North Holland, Amsterdam).
Thon, D. (1979), "On measuring poverty", The Review of Income and Wealth 25:429-439.
Tsui, K.-Y. (1995), "Multidimensional generalizations of the relative and absolute indices: the Atkinson-
Kolm-Sen approach", Journal of Economic Theory 67:251-265.
Watts, H.W. (1968), "An economic definition of poverty", in: D.P. Moynihan, ed., On Understanding
Poverty (Basic Books, New York) pp. 316-329.
Weymark, J.A. (1981), "Generalized Gini inequality indices", Mathematical Social Sciences 1:409-430.
Weymark, J.A. (1999), "Comment on Blackorby, Bossert, and Donaldson", in: J. Silber, ed., Handbook
of Income Inequality Measurement (Kluwer Academic Publishers, Boston) pp. 157-161.
Yitzhaki, S. (1979), "Relative deprivation and the Gini coefficient", Quarterly Journal of Economics
93:321-324.
Yitzhaki, S. (1983), "On an extension of the Gini inequality index", International Economic Review
24:617-628.
Zheng, B. (1997), "Aggregate poverty measures", Journal of Economic Surveys 11:123 162.
AUTHOR INDEX

Abdou, J. 420 Barberi, S. 39, 50, 51, 79, 80, 82, 218, 316,
Abello, J.M. 205 420, 421, 532
Abreu, D. 243, 264, 265, 267, 269, 280 Barone, E. 7, 14, 15
Aczel, J. 561,567, 570 Barrett, C.R. 61, 80
Aizerman, M. 78, 100-102, 111, 115, 116 Barthelemy, J.P. 81
Aleskerov, E 78, 100, 110, 111, 115, 116, Basmann, R 630
124 Basu, K. 479, 482, 484, 494, 551
Aleskerov, E, see Aizerman, M. 78, 100-102, Batra, R.N. 142
111, 115, 116 Bawa, V 609
Amy, D.J. 225 Beer, S. 428, 449
Anand, S. 622 Beja, A. 40
Anckar, D. 226 Ben-Porath, E. 514, 521, 522, 525, 535
Anderson, E. 549 Benassy, J.-P. 295
Arian, A. 438 Bentham, J. 5, 470, 546
Armstrong, T.E. 83 Berga, D. 206
Arneson, R. 549 Bergin, J. 281
Aronson, JR., see Lambert, P.J.614 Bergson, A. 7, 8, 17, 462
Arrhenius, G. 576 Bhattacharya, N. 614
Arrow, K.J. 10, 12, 13, 19-23, 25, 37, 38, 50, Billera, L. 329, 343, 348
69, 71, 76, 98, 102, 110, 125, 133, 134, Black, D. 3, 4, 58, 68, 98, 110, 134, 136, 138,
136, 138, 161, 178, 179, 188, 206, 241, 177, 179, 206, 211, 241, 364, 374, 379,
364, 365, 375, 392, 428, 434, 462, 470, 380, 393, 428
474, 475, 510, 546, 553, 557, 582 Blackorby, C. 37, 38, 50, 84, 465, 476, 490,
Atkinson, A.B. 24, 26, 558, 599, 601, 602, 507, 520, 521, 527, 545, 547, 551, 556,
608, 609, 612, 619, 620, 627 559-563, 566, 571-578, 580-584, 589,
Auersperg, M., see Blackorby, C. 613 601, 604-606, 613, 622, 623
Aumann, G. 165 Blackwell, D. 508, 519, 569
Aumann, R.J. 102, 295, 302-304, 343 Blair, D.H. 20, 63, 75, 76, 124, 151, 152
Austen-Smith, D. 427, 433 Blau, J.H. 21, 24, 46, 50, 63, 74, 75, 144, 499,
555
Bock, H.-H. 218
Baigent, N. 37, 70, 162 Boehm, G.A.W. 190
Bailyn, B. 433 Bolger, E.M. 218, 219
Balasko, Y. 58 Borda (J.-C. de Borda) 3, 4, 70, 98, 134, 179,
Baliga, S. 275, 276, 281 362, 363, 392
Balinski, M.L. 184 Border, K. 582
Bandyopadhyay, T. 76, 79 Bordes, G.A. 65, 72, 147, 150, 555
Banerjee, A. 80 Bordes, G.A., see Blair, D.H. 20, 76, 124
Banker, R. 298 Bordley, R.E 207
Banks, J., see Austen-Smith, D. 427, 433 Borm, P., see Koster, M. 352
Banks, J.S. 58, 62, 211, 433, 435, 438, 444 Bossert, W. 38, 465, 478, 482, 483, 512, 514,
Banzhaf II, J.E 184 518, 531, 551, 560, 575, 603, 615, 616
Bar-Hillel, M. 464 Bossert, W., see Blackorby, C. 37, 38, 84,
1-1
I-2 Author Index

476, 507, 521, 547, 563, 566, 571-578, Corch6n, L., see Baliga, S. 275
580 582, 601 Corch6n, L., see Chakravorty, B. 275
Bourguignon, F. 613 Coughlin, P.J. 38, 80. 177, 179, 436
Bourguignon, F., see Atkinson, A.B. 612 Coulhon, T. 527, 582
Bowen, H. 241 Courant, P.N., see Chamberlin, J.R. 225
Brams, S.J. 126, 176, 179, 190, 192, 193, 197. Cowell, F.EA. 599, 613
199, 201-203, 208, 217 219, 221-224, Cox, G.W. 176, 225, 435-437, 444
226, 393, 407 Craven, J. 205
Brams, S.J., see Fishburn, P.C. 39, 71, 197, Creedy, J. 608
201, 202, 208, 221. 223 Crs, C.A.. see Balasko, Y. 58
Brams, S.J., see Potthoff, R.F. 225 Crouch, D. 449
Brewer, J. 427
Broome, J. 475, 548, 549, 554, 572, 581. 582, Dagan, N. 300, 302, 305
586 Dagum, C. 616
Brown, D.J. 49, 62, 111, 547 Dalton, H. 6. 558, 599, 607. 610
Brusco, S. 281 Danilov, V. 111, 251,253
Brusco, S., see Baliga, S. 275 Dardanoni, V 50, 619
Bryant, J. 345 Dasgupta, P. 22, 152, 153, 198, 245, 255-257,
Buchanan, J.M. 19 278, 575, 608 610
d'Aspremont, C. 20, 38, 51, 245, 420, 465,
Cabrales, A. 244 467, 493, 499, 505, 511, 519, 521, 525,
Cachon, G. 295 548, 551, 555, 557, 561, 569
Calvert, R. 430, 435 d'Aspremont, C., see Mongin, P. 465, 522,
Calvo, E. 339 549, 554, 582-585
Campbell, D.E. 21, 40, 53, 54, 57, 61, 70, 72, Davidson, R.R. 215
73, 75, 76, 78, 80, 83, 153 Davies. J. 608
Candeal, JC. 80 Davis, M. 335
Cantillon, E. 66 Davis, M.D., see Brams, S.J. 176
Caplin, A. 68, 69 Davis, O.A. 69
Carlson, E. 576 Day, H.E. 81
Carmen Sanchez, M., see Peris, J.E. 125 Day, WH.E., see Bock, H.-H. 218
Chakravarty, S.R. 599, 604, 611, 614, 616, de Frutos, M.A. 298, 327
618, 619, 623, 624 de Grazia, A. 179
Chakravorty, B. 250, 275 De Groot, M.H., see Davis, O.A. 69
Chamberlin, J.R 207, 225 De Meyer, B. 583
Charon, 1. 211 de Nouweland, A. 338
Chebotarev, P. 126 de Trenqualye, P. 244
Chen, C., see Pyatt, G. 614 Deb, R. 38, 75, 245
Chichilnisky, G. 160-164, 166 Deb, R., see Bandyopadhyay, T. 79
Ching, S. 206, 316 Deb, R., see Blau, J.H. 63
Chipman, IS. 7, 8 Debord, B. 71,218
Choi, J. 281 Debreu, G. 517, 533, 561, 562
Chun, Y. 298, 306, 322 Demers, A. 314
Chwe, M.S.-K. 218 DeMeyer, E 204, 551
Clark, S. 623, 625, 626 Denicolb, N. 512
Clarke, E.H. 241 Denicol6, V. 50, 78. 125
Condorcet (M.J.A.N. de Condorcet) 2-4, 57, Deschamps, R. 514, 517, 557, 558, 569, 571
98, 134, 179, 210, 211. 215, 363, 384, Deutsch, M. 291
427 Dhiilon, A. 528, 529, 534, 536, 580
Copeland, A.H. 210 Diamond, P. 84, 521, 522, 530, 531, 535,
Corch6n, L. 239, 281 589
Author Index I-3

Dixit, A.K. 133, 487, 551 Fei, J.C.S., see Fields, G.S. 607
Dobb, M. 13 Feiwel, G.R. 50
Dodgson, C.L. (Lewis Carroll) 4, 98, 134, Feld, S.L., see Miller, N.R. 211
211 Felsenthal, D.S. 184, 190, 207
Domotor, Z. 582, 583 Felsenthal, D.S., see Rapoport, A. 198
Donaldson, D. 560, 615, 622 Ferejohn, J.A. 64, 104
Donaldson, D., see Blackorby, C. 37, 38, 50, Fields, G.S. 607, 614, 616, 617
84, 465, 476, 490, 507, 520, 521, 527, 545, Filmer, R. 427
547, 551, 556, 559, 560, 562, 563, 566, Fine, B. 364, 367, 380, 388 390
571-578, 580-584, 589, 601, 604-606, Fine, K. 137, 185
613, 622, 623 Fine, K., see Fine, B. 364, 367, 380, 388
Doron, G. 201, 221 390
Dorussen, H., see Lin, T. 430, 444 Fiorina, M. 433
Downs, A. 430, 435 Fishburn, J.E, see Fishburn, P.C. 199
Dreze, J. 295 Fishburn, P.C. 37, 39, 43, 50, 58, 64, 71, 72,
Dubey, P. 184 83, 111, 136, 139, 140, 143, 185, 186, 188,
Duggan, J. 78, 280, 281, 428 190, 196, 197, 199, 201, 202, 205-208,
Duggan, J., see Aleskerov, E 115, 116 210 212, 214-216, 218, 221, 223, 225,
Duggan, J., see Banks, J.S. 435, 438, 444 226, 366, 374, 380, 387, 484, 526, 551,
Dummett, M. 144, 145, 210, 217, 241 582
Dutta, B. 37, 70, 80, 84, 210, 258, 259, 264, Fishburn, P.C., see Brams, S.J. 126, 190, 192,
280, 397, 468, 606 197, 199, 208, 218, 221, 223, 224, 226,
Dutta, B., see Chakravarty, S.R. 616, 618 393, 407
Duverger, M. 430, 437, 444, 449 Fishburn, PC., see Ferejohn, J.A. 64, 104
Dworkin, R. 24 Fishburn, P.C., see Gehrlein, W.V. 143, 204,
209, 216, 217, 226
Ebert, U. 516, 517, 520, 606, 614, 616 Fishburn, P.C., see Rubinstein, A. 50, 51
Eckmann, B. 165 Fisher, S., see Spencer, B.D. 627
Eichhorn, W. 567, 604 Fleming, M. 562
Eliaz, K. 281, 282 Fleurbaey, M. 71, 477, 479, 532, 534, 536,
El'kin, L., see Ilyunin, 0. 126 549
El'kin, L., see Popov, B. 116 Fleurbaey, M., see Blackorby, C. 576
Elster, J. 24, 295 Foley, D.K. 21
Enelow, J., see Lin, T. 430, 444 Foster, J.E. 549, 601, 608-610, 612-614, 620,
Enelow, J.M. 58, 176, 430, 436 623, 626, 627
Epstein, D. 210 Foster, J.E., see Shorrocks, A.E 608, 611,
Epstein, L.G. 525, 530 625
Erd6s, P. 58, 204 Fountain, J. 60, 64, 75
Esteban, J., see Dutta, B. 606 Friedman, E.J. 335, 342, 344, 347
Fudenberg, D. 244
Falmagne, J.-C. 551
Farkas, D. 385 Gaertner, W. 24, 65, 156, 158, 159, 165, 245,
Farquharson, R. 22, 177-179, 199, 200, 243, 490
264, 267, 421 Galton, E 4, 206
Farquharson, R., see Dummett, M. 144, 145, Garcia-Lapresta, J.L. 80
241 Gairdenfors, P. 24, 245, 378, 380
Farrell, J. 275 Gardner, R. 409
Featherston, E, see Chamberlin, J.R. 207 Garrett, G. 449
Feddersen, T. 427 Gaspart, F. 245
Fei, J., see Pyatt, G. 614 Gastwirth, J.l. 608
Fei, J.C.S. 614 Geanakoplos, J. 50
1-4 Author Index

Gehrig, W., see Eichhom, W. 604 Hammond, P.J. 20. 21, 24, 69. 245. 467. 475,
Gehrlein, W.V. 58, 143, 201, 204, 205, 209, 476, 502, 506, 521-523, 531, 549, 555,
210, 216-218, 226, 379 557, 575, 582, 583, 587
Gehrlein, W.V., see Fishburn, P.C. 58, 143, Hammond, P.J., see Dasgupta, P. 22, 198, 245,
205, 207. 380 255 257, 278
Gerard-Varet, L.A., see d'Aspremont, C. 245 Hansson, B. 49, 77, 78, 213, 377, 380
Gevers, L. 51, 516, 551. 555 Hardy, G. 608
Gevers, L., see d'Aspremont. C. 20, 38, 51, Hare, R. 547, 568
493, 499, 505, 511, 548, 551, 555, 557, Hare, T. 179
561, 569 Harris, M. 245, 278
Gevers, L., see Deschamps, R. 514, 517, 557, Harsanyi, J.C. 9, 10, 244, 475, 521, 526, 530,
558, 569, 571 548, 582, 583
Gibbard, A.E 21. 24, 59, 76, 110, 133, 178, Hart, S. 334
188, 240, 241,256, 413 Hayek, F. 239
Gilboa, I. 514 Heal, G., see Chichilnisky, G. 162 164
Gilboa, I., see Ben-Porath, E. 514, 521, 522, Heal, G.M. 14, 162, 164
525, 535 Heaney, J.P., see Straffin Jr, PD. 328
Gillen, R. 143 Heath, D., see Billera, L. 329, 343, 348
Girshick, M., see Blackwell, D. 569 Heinecke, A., see Gaertner, W 156
Girshick, M.A., see Blackwell, D. 508, 519 Hemming, R., see Clark, S. 623, 625, 626
Glasser, G.J. 218 Henriet, D. 210
Glazer, J. 264, 269, 282 Herrero, C. 300
Good, I.J. 211 Herrero, M. 269
Goodin, R. 546, 554 Herschbach, D.R., see Brams, S.J. 226
Goodman, L.A. 210. 364 Herzberger, H. 76
Gopinath, B., see Gehrlein, WV. 217 Hewitt, F., see McLean, I. 3, 427
Gorman, W.M. 8, 561 Hicks, J.R. 5, 7, 475, 491
Gotoh, R. 24 Hinich, M.J., see Davis, O.A. 69
Graaff, J. de V 7 Hinich, M.J., see Enelow, J.M. 176, 430, 436
Grandmont, J.M. 68, 69, 154, 155 Hoag, C.G. 211
Green, D. 102 Hobbes, T. 427
Green, J. 241 Hoffman, D.T. 197
Greenberg, J. 68 Hollard, G. 58
Greer, J., see Foster, J.E. 626 Holzman, R. 202, 397, 417-420
Grether, D.M. 76, 124 Hong, L. 258
Griffin, J. 545, 554 Horvath, Ch.D. 165
Grofman, B., see Miller, N.R. 211 Hougaard, J.L. 327, 353
Grofman, B., see Regenwetter, M. 207 Hoy, M., see Davies, J. 608
Grossman, S. 275 Hudry, O., see Charon, I. 211
Groves, T. 239, 241, 242 Hurka, T. 575
Guha, A.S. 59, 110, 555 Hurwicz, L. 14, 22, 239-242, 244, 245, 248,
Guilbaud, G.Th. 143, 144, 204, 205 258, 260, 276
Guinier, L. 218 Hylland, A., see Gibbard, A.E 76

Hadar, J. 609 Ichiishi, T. 351

Hagenaars, A. 624 llyunin, O. 126
Hagy, A., see Fishburn, P.C. 199 Inada, K. 13, 136-i40, 207
Haimanko, O. 342 Induriin, E., see Candeal, J.C. 80
Hallett, G.H., see Hoag, C.G. 211 Inglehart, R., see Rabier, J.-R. 445
Hammond, P., see Bordes, G.A. 555 ISEIUM 445
Author Index I-5

Jackson, M. 239, 244, 245, 262, 263, 265, Kundu, A. 623, 624
270, 277, 279, 280 Kuo, S.W.Y., see Fei, J.C.S. 614
Jackson, M.O., see Barbera, S. 82, 316, 532
Jain, S.K. 141 Ladha, K. 427
Janowitz, M.E, see Barthelemy, J.P. 81 Laffond, G. 211
Jenkins, S.P. 627 Laffont, J.J., see Green, J. 241
Johansen, L. 18, 23 Lagarias, J.C., see Gehrlein, W.V 217
Johnson, C.R., see Abello, J.M. 205 Laine, J. 80
Johnson, M.R. 40 Lake, M. 176
Jordan, J. 244 Lambert, P.J. 599, 614
Lambert, P.J., see Jenkins, S.P. 627
Kakwani, N.C. 599, 608, 615, 620, 625 Lange, O. 14-17
Kalai, E. 22, 41, 64, 74, 146-151 Laplace, P.-S. 4
Kaldor, N. 7 Lariviere, M., see Cachon, G. 295
Kaminski, M. 305, 311 Laslier, J.-E 66, 211
Kanbur, S.M.R. 619 Laslier, J.-E, see Laffond, G. 211
Kaneko, M. 137, 475 Lauwers, L. 83
Kaplow, L. 70, 80 Laver, M. 438
Kass, R. 446 Le Breton, M. 37, 38, 42, 63-65, 77, 147,
Keiding, H. 185, 186, 420, 421 210
Keiding, H., see Abdou, J. 420 Le Breton, M., see Banks, J.S. 435
Kelly, J. 111 Le Breton, M., see Bordes, G.A. 65, 147, 150,
Kelly, J.S. 50, 58, 64. 65, 83, 84, 137, 143, 555
188, 205, 220 Le Breton, M., see Hollard, G. 58
Kelly, J.S., see Blair, D.H. 20, 76, 124 Le Breton, M., see Lain6, J. 80
Kelly, J.S., see Campbell, D.E. 21, 53, 54, 57, Leclerc, B. 81
61, 70, 72, 73, 76, 78, 80, 153 Leclerc, B., see Barthelemy, J.PR 81
Kelly, J.S., see Fishburn, P.C. 64 Ledyard, J. 241
Kelsey, D. 24, 46, 63, 64, 501 Ledyard, J., see Groves, T. 239, 242
Kemeny, J. 211 Lepelley, D., see Gehrlein, W.V 379
Kemp, M.C. 69, 467 Lerner, A.P. 15
Keshav, S., see Demers, A. 314 Levchenkov, V 111
Keynes, J.M. 431 Levchenkov, VS. 66
Khmelnitskaya, A. 333 Levenglick, A., see Young, H.P. 211
Kilgour, D.M., see Brams, S.J. 201-203, 217 Levin, J. 210
Kim, K.H. 205, 420 Levine, D., see Fudenberg, D. 244
Kim, T., see Choi, J. 281 Lewis, A.A. 83
King, M. 618, 619 Lezina, Z., see Vol'skiy, V 98
Kirman, A.P. 37, 49, 83 Lin, T. 430, 444
Kolm, S.-Ch. 21, 24, 469, 475, 524, 535, 558, Lines, M. 225
599, 601, 603, 608, 612 Little, I.M.D. 12
Kolpin, V 322, 354 Little, J.D.C., see Fishburn, P.C. 226
Koray, S. 81 Littlechild, S. 345
Koster, M. 352, 353 Littlewood, J., see Hardy, G. 608
Koster, M., see Tijs, S. 328, 353 Llamazares, B., see Garcia-Lapresta, J.L. 80
Kramer, G.H. 68, 160, 434 Locke, J. 427
Krantz, D. 478, 551 Lockwood, B. 465
Krause, U. 50, 505, 511 Loeb, M., see Groves, T. 241
Krelle, W. 632 Loehman, E. 333
Kronick, R., see Doron, G. 201, 221 London, J., see McLean, I. 2, 134
Kuga, K., see Cowell, EA. 613 Love, R. 614
I-6
Lowell, L. 430
Luce, R.D. 511
Luce, R.D., see Krantz, D. 478, 551
Ludwig, WG. 217
Ma, B.K. 69
Maasoumi, E. 612, 613
Machover, M., see Felsenthal, D.S. 184
Madison, J. 428
Mahalanobis, B., see Bhattacharya, N. 614
Majumdar, M., see Foster, J.E. 612
Makarov, 1. 112
Mala, J. 58
Malawski, M. 53
Malinvaud, E. 14
Malishevski, A. 102
Malishevski, A., see Aizerman, M. 101, 102
Maniquet, F, see Fleurbaey, M. 71, 477, 479,
532, 534, 536, 549
Maoz, Z., see Felsenthal, D.S. 207
Marcus-Roberts, H., see Fishburn, P.C. 484,
551
Marglin, S.A. 133
Margush, T. 81
Markandya, A. 619
Markowitz, H., see Goodman, L.A. 364
Marshall, A.W. 608
Martin, A., see Quinn, K. 445
Martin, A., see Schofield, N. 445
Martinelli, C., see Duggan, J. 428
Mas-Colell, A. 20, 59, 64, 110, 136
Mas-Colell, A., see Hart, S. 334
Maschler, M. 304
Maschler, M., see Aumann, R.J. 295, 302-
304
Maschler, M., see Davis, M. 335
Maskin, E. 14, 20, 22, 117, 118, 120, 124,
148, 152, 239, 242-245, 249, 250, 252,
260, 264, 266, 267, 270-274, 282, 413,
518, 519, 561, 569
Maskin, E., see Dasgupta, P. 22, 152, 153,
198, 245, 255-257. 278
Maskin, E., see Fishburn, P.C. 58, 143, 205
Maskin, E., see Hurwicz, L. 245, 248, 258
Matsushima, H. 243, 264, 265, 280
Matsushima, H., see Abreu, D. 264, 265, 280
May, K.O. 37, 64, 126, 136, 184, 382
May, R.M. 204
McGarvey, D.C. 58, 204
McGuire, J.B. 81
McKelvey, R.D. 58, 67, 250, 433, 435
Author Index
McLean, 1. 2, 3. 134, 179, 211, 427
McLean, R. 354
McLennan, A. 50, 51, 79
McMorris, FR. 81
McMorris, F.R., see Barthelemy, J.P. 81
McMorris, ER., see Back, H.-H. 218
McMorris, ER., see Day, H.E. 81
McMorris, ER., see Margush, T. 81
Mehran, E 615
Mellen, S. 430
Merlin, V., see Tataru, M. 216
Merrill, S. 197
Merrill III, S. 190, 197, 207-210, 379
Merrill III, S., see Brams, S.J. 197, 226
Mertens, J.-F, see Dhillon, A. 528, 529, 534,
536
Michael, E. 442
Mihara, H.R. 83
Mill, J.S. 23, 220, 307, 427, 430, 546, 547
Miller, G., see Ladha, K. 427
Miller, N. 210, 434
Miller, N.R. 210, 211
Milnor, J. 511, 569
Mirkin, B. 81, 111
Mirman, L. 329, 343, 348
Mirrlees, J.A. 522
Mishan, E.J. 25
Mitra, T. 617
Mitra, T., see Foster, J.E. 612
Mongin, P. 465, 522, 525, 548, 549, 554.
582-585, 587
Mongin, P., see Coulhon, T. 527, 582
Mongin, P., see d'Aspremont, C. 467, 525
Mongin, P, see De Meyer, B. 583
Monjardet, B. 49, 111
Monjardet, B., see Barthelemy, J.P. 81
Monroe, B.L. 225
Mookherjee, D. 279
Moore, J. 239, 243, 247, 250, 251, 259, 267
269
Moore, J., see Maskin, E. 243, 267, 270-274
Moore, J.C., see Chipman, J.S. 7, 8
Morgenstern, O., see von Neumann, J. 22, 62,
548, 581
Moser, L., see Erd6s, P. 58, 204
Moulin, H. 39, 84, 102, 111, 124, 126, 179,
201, 202, 206, 243, 253, 264, 267, 298,
310, 311, 314-316, 321, 323, 326, 327,
329, 338 340, 350, 351, 397, 398, 409,
412, 420, 421, 465, 477, 479, 506, 518,
536
Author Index
Moulin, H., see Friedman, E.J. 344, 347
Mount, K. 242, 258
Moyes, P. 608
Mueller, D. 397, 421
Muench, T. 244
Mukherjee, D., see Chakravarty, S.R. 604
Muller, E. 122, 256, 406
Muller, E., see Blair, D.H. 151, 152
Muller, E., see Kalai, E. 41, 64, 146 149,
151
Murakami, Y. 53, 112, 185, 367, 370
Murphy, L. 545
Myerson, R. 245, 278
Myerson, R.B. 177, 214
Nagahisa, R. 64
Nagel, J. 226
Nagel, J., see Merrill III, S. 209
Nagel, J.H., see Brams, S.J. 226
Naiton, W. 110
Nakamura, K. 111,410
Nalebuff, B., see Caplin, A. 68, 69
Nalebuff, B., see Levin, J. 210
Nalebuff, B.J., see Dixit, A.K. 133
Nanson, E.J. 4, 98, 179, 211
Nash, J.E 22, 434, 472, 477, 485
Naumova, N.I. 311
Negishi, T. 16
Neme, A., see Barbera, S. 316
Nermuth, M. 76
Neumann, D., see McMorris, ER. 81
Newbery, D.M.G. 609
Ng, Y.-K. 576
Ng, Y.-K., see Kemp, M.C. 69, 467
Niemi, RG. 197, 201, 204, 205, 225
Nitzan, S. 213, 216, 380-384
Nitzan, S., see Farkas, D. 385
Nixon, D., see Schofield, N. 438, 444
Norberg, K. 427
North, D. 427
Nurrni, H. 58, 98, 201, 202, 207, 210, 211,
225
Odeh, R.E., see Davidson, R.R. 215
Ok, E., see Fields, G.S. 616, 617
Ok, E.A., see Foster, J.E. 608
Ok, E.A., see Mitra, T. 617
Olkin, I., see Marshall, A.W 608
O'Neill, B. 295, 298, 302, 304
Ordeshook, P.C., see Riker, W.H. 144
Ortufio-Ortin, I., see Corch6n, L. 281
I-7
Owen, G. 350, 351
Owen, G., see Littlechild, S. 345
Palfrey, T. 239, 243, 245, 261, 262, 277, 279,
280
Palfrey, T., see Jackson, M. 263, 265, 270
Pareto, V 7, 8
Parfit, D. 575
Parks, R. 115
Parks, R., see Schofield, N. 444
Parks, R.P. 69, 467
Parthasathy, KIR. 442
Pattanaik, PK. 10, 22, 24, 25, 39, 71, 133,
136, 137, 139, 140, 145
Pattanaik, P.K., see Bandyopadhyay, T. 79
Pattanaik, P.K., see Barrett, C.R. 61, 80
Pattanaik, P.K., see Batra, R.N. 142
Pattanaik, P.K., see Deb, R 245
Pattanaik, P.K., see Dutta, B. 397
Pattanaik, P.K., see Gaertner, W. 24, 245,
490
Pattanaik, P.K., see Sen, A.K. 13, 110, 137-
139, 141, 205
Pazner, E.A. 533
Peleg, B. 179, 257, 264, 397, 405, 408-412,
416-421, 432
Peleg, B., see Barberi, S. 420
Peleg, B., see d'Aspremont, C. 420
Peleg, B., see Keiding, H. 420, 421
Peleg, B., see Moulin, H. 124, 264, 420, 421
Peris, J.E. 78, 125
Peris, J.E., see Sanchez, M.C. 78
Perry, M., see Grossman, S. 275
Pesendorfer, W., see Feddersen, T. 427
Peters, H., see Peleg, B. 421
Pfingsten, A. 603
Pfingsten, A., see Bossert, W. 603
Pigou, A.C. 6, 25, 476
Plott, C.R. 20, 58, 67, 76, 102, 124, 433
Plott, C.R., see DeMeyer, E 204, 551
Plott, C.R., see Grether, D.M. 76, 124
Podder, N., see Kakwani, N.C. 608
Pollak, R.A. 69, 467, 603
Pollak, R.A., see Blair, D.H. 63, 75
Polya, G., see Hardy, G. 608
Ponti, G., see Cabrales, A. 244
Poole, K. 435
Popov, B. 116
Popov, B., see lIlyunin, O. 126
Popper, K. 430
Postlewaite, A. 245, 258, 277, 279, 280
1-8 Author Inldex

Postlewaite, A., see Hurwicz, L. 245, 248, Roberts, J., see Ledyard, J. 241
258 Roberts, K.W.S. 20, 38, 69, 256, 467, 475,
Potthoff, R.F 225 479, 501, 502, 535, 551, 561, 569
Pouzet, M. 50 Roemer, J.E. 465, 477, 536, 549, 589
Powers, R.C. 70, 81 Rosenthal, H., see Poole, K. 435
Powers, R.C., see Barthelemy, J.P 81 Rosenthal, R., see Glazer, J. 264
Powers, R.C., see McMorris, ER. 81 Rothschild, E. 3, 5
Primont, D., see Blackorby, C. 561 Rothschild, M. 609
Pyatt, G. 613-615 Roush, FEW, see Kim, K.H. 205, 420
Rubchinskiy, A., see Makarov, 1. 112
Quinn, K. 445 Rubinstein, A. 50, 51, 67, 69, 270
Quinn, K., see Schofield, N. 445 Rubinstein, A., see Glazer, J. 269, 282
Rubinstein, A., see Nitzan, S. 213, 380-384
Raanan, J., see Billera, L. 329 Runciman, W.G. 604
Rabier, J.-R. 445 Russell, R., see Blackorby, C. 561
Rabinovich, N. 295 Russell, R.R. 613
Rae, D. 427, 448 Russell, W., see Hadar, J. 609
Raftery, A., see Kass, R. 446
Raiffa, H., see Luce, R.D. 511 Saari, D.G. 50, 66, 71, 178, 179, 197, 201.
Rangel, A., see Cantillon, E. 66 215, 216, 226, 381, 386. 433
Ranis, G., see Fei, J.C.S. 614 Sahlquist, H., see Hansson, B. 213, 380
Rao, VM. 614 Saijo, T. 248-250. 258
Rapoport, A. 198 Salles, M. 38, 80, 140, 145, 146
Ravallion, M. 620 Salles, M., see Barrett, C.R. 61, 80
Rawls, J. 9, 24, 469, 475, 476, 545 Samuelson, P.A. 2, 7, 8, 11, 12, 21, 102, 241
Ray, D. 202 Sainchez, M.C. 78
Raynaud, H. 205 Sanchez, M.C., see Peris, J.E. 78
Raynaud, H., see Arrow, K.J. 37 Santos, J.C., see Calvo, E. 339
Raz, R. 206 Saposnik, R. 155, 156
Razzolini, L., see Deb, R. 245 Sargent, TJ., see Williamson, O.E. 143
Redekop, J. 64 Sarin, R. 522
Regenwetter, M. 207 Sato, F. 258
Reichelstein, S. 258 Satterthwaite, M.A. 21, 133, 178, 188, 241,
Reichelstein, S., see Mookherjee, D. 279 256, 258, 413
Reiter, S., see Mount, K. 242, 258 Satterthwaite, M.A., see Kalai. E. 41, 64, 146.
Reiter, S., see Reichelstein, S. 258 147
Reny, P.J. 50 Satterthwaite, M.A., see Muller, E. 122, 256,
Repullo, R. 249, 279 406
Repullo, R.. see Moore, J. 243, 247, 250, 251, Savage, L. 430
259, 267, 269 Scarf, H. 16
Rhodes, G., see Basmann, R. 630 Scarf, H., see Debreu. G. 533
Richelson, J.T. 190, 210 Scarsini, M. 201, 202
Riker, W.H. 144. 177, 201, 204, 210, 428-430. Schama, S. 427
433, 445 Schattschneider, E. 429
Riker, W.H., see Niemi, R.G. 201, 225 Schmeidler. D. 258, 264, 282, 349, 350
Ritz, Z. 151 Schmeidler, D., see Ben-Porath, E. 514, 521,
Ritz, Z., see Kalai, E. 150, 151 522, 525, 535
Robbins, L. 6 Schmeidler, D., see Gilboa, 1. 514
Roberts, E, see Fishburn, P.C. 484, 551 Schmeidler, D., see Hurwicz, L. 242, 248,
Roberts, ES. 190 260
Roberts, J., see Duggan, J. 281 Scluneidler, D., see Kalai, E. 74
Author Index 1-9

Schmeidler, D., see Pazner, E.A. 533 Shinotsuka, T. 84

Schmeidler, D., see Postlewaite, A. 245, 277, Shneyerov, A.A., see Foster, J.E. 613
279, 280 Sholomov, L. 112
Schofield, N. 58, 427-431, 433, 435, 438, Shorrocks, A.E 535, 608, 610, 611, 613, 614,
440-442, 444, 445, 449, 451 616, 617, 619, 624, 625, 627
Schofield, N., see Laver, M. 438 Shorrocks, A.F., see Foster, J.E. 620, 626,
Schofield, N., see McKelvey, R.D. 433 627
Schummer, J. 316 Shorrocks, A.E, see Krelle, W. 632
Schummer, J., see Chun, Y. 306 Shubik, M. 324, 338, 479
Schumpeter, J.A. 2 Shubik, M., see Shapley, L.S. 184, 479
Schwartz, T. 61, 111, 210, 211 Sidgwick, H. 545-547
Scitovsky, T. 7, 8 Sikora, R. 576
Segal, I. 272 Silber, J. 599, 614
Segal, U., see Epstein, L.G. 525, 530 Sj6str6m, T. 243, 244, 251, 252, 258, 259,
Selten, R., see Harsanyi, J.C. 244 262-265, 274
Sen, A.K. 2, 9, 10, 13, 20, 21, 23, 24, 37-39, Sj6str6m, T., see Baliga, S. 275, 276
46, 49, 50, 57, 62, 65, 66, 76, 78, 102, Slater, P. 211
107, 110, 133, 135 139, 141, 152, 205, Slesnick, D.T 619
257, 365, 385, 463, 465, 467, 471, 474, Slutsky, S. 156, 159
476, 479, 493, 527, 540, 545, 546, 548, Smith, A. 429
550, 551, 553, 555, 558, 589, 599, 601 Smith, J. 365, 374, 378, 381, 386-388, 390,
603, 606, 619, 621, 623 391
Sen, A.K., see Arrow, K.J. 25 Smith, J.H. 37, 212, 213, 216, 226
Sen, A.K., see Dasgupta, P. 608-610 Smith, T.E., see Kundu, A. 623, 624
Sen, A.K., see Foster, J.E. 601, 614 Smith, V 244
Sen, Arunava 248 Smorodinsky, M., see Kalai, E. 22
Sen, Arunava, see Abreu, D. 243, 264, 265, Snyder, J.M. 177
267, 269 Sobolev, A.I. 335
Sen, Arunava, see Bergin, J. 281 Sokolov, V., see Makarov, I. 112
Sen, Arunava, see Dutta, B. 37, 70, 258, 259, Sondermann, D., see Kirman, A.P. 37, 49, 83
264, 280 Sonnenschein, H., see Barbera, S. 50, 51, 79,
Sened, I., see Schofield, N. 438, 440, 442, 218, 421
444 Sonnenschein, H., see Mas-Colell, A. 20, 59,
Sengupta, M., see Pattanaik, P.K. 140 64, 110, 136
Serizawa, S., see Ching, S. 206 Sonnenschein, H., see Satterthwaite, M.A.
Serrano, R. 281 258
Sertel, M.R. 190, 217 Spafford, D. 224
Shamir, M., see Arian, A. 438 Spector, L., see Tabarrok, A. 226
Shapiro, I., see Green, D. 102 Spencer, B.D. 627
Shapley, L.S. 184, 291,332, 479 Sprumont, Y. 316, 339, 341, 352, 353, 420,
Shapley, L.S., see Aumann, R.J. 343 477, 479
Shapley, L.S., see Dubey, P. 184 Srivastava, S., see Herrero, M. 269
Sharkey, W., see McLean, R. 354 Srivastava, S., see Jackson, M. 263, 265
Shavell, S., see Kaplow, L. 70, 80 Srivastava, S., see Palfrey, T. 243, 245, 261,
Shenker, S. 314, 321 262, 277, 279, 280
Shenker, S., see Demers, A. 314 Staring, M. 218
Shenker, S., see Moulin, H. 310, 321, 323, Starrett, D., see Dasgupta, P. 608 610
326, 327 Stavely, E.S. 179
Shepsle, K.A. 166, 210, 433 Stearns, R. 58, 204
Shepsle, K.A., see Fiorina, M. 433 Stefanescu, A. 116
Sheshinski, E. 603 Stehling, E, see Bossert, W. 551
1-10
Stiglitz, J., see Kanbur, S.M.R. 619
Stiglitz, J.E., see Rothschild, M. 609
Stong, R., see Moulin, H. 314
Storcken, T., see Peleg, B. 421
Straffin Jr, PD. 210, 328
Strasnick, S. 531
Sugden, R. 24
Suh, S.C. 264, 327
Sumner, L. 545, 554
Suppes, P. 20, 503
Suppes, P., see Krantz, D. 478, 551
Suzumura, K. 8, 12, 20, 21, 24, 38, 40, 50,
102, 475, 490, 535, 549
Suzumura, K., see Arrow, K.J. 25
Suzumura, K., see Blair, D.H. 20, 76. 124
Suzumura, K., see Fountain, J. 60, 64, 75
Suzumura, K., see Gaertner, W 24, 245, 490
Suzumura, K., see Gotoh, R. 24
Suzumura, K., see Pattanaik, PK. 24
Sweezy, P.M. 16, 17
Tabarrok, A. 226
Takayama, N. 623
Tatamitani, Y., see Saijo, T. 258
Tataru, M. 216
Tauman, Y. 329
Tauman, Y., see Mirman, L. 329, 343, 348
Taylor, F.EM. 15
T6jedo, C. 354
Theil, H. 613
Thomas, A. 328
Thompson, C.J., see McGuire, J.B. 81
Thomson, W. 39, 53, 84, 247, 248, 258, 304,
316, 477, 536
Thomson, W. see Chun, Y. 306
Thomson, W, see Schummer, J. 316
Thon, D. 623
Thorbecke, E., see Foster, J.E. 626
Thorlund-Petersen, L., see Hougaard, J.L. 327.
353
Tian, G. 258
Tideman, T.N. 210, 220
Tideman, T.N., see Bordes. G.A. 72
Tijs, S. 328, 353
Tijs, S., see de Nouweland, A. 338
Tijs, S., see Koster, M. 352
Tirole, J., see Maskin, E. 282
Todhunter, 1. 4
Tovey, C.A. 58
Townsend, R., see Harris, M. 245, 278
Trannoy, A., see Laine, J. 80
Author Index
Truchon, M., see Le Breton, M. 63, 64, 210
Truchon, M., see T6j6do, C. 354
Tsui, K.-Y. 486, 612
Tullock, G. 69, 154, 225
Tungodden, B. 531, 580
Tversky, A., see Krantz, D. 478, 551
Ulph, D., see Clark, S. 623, 625, 626
Urken, A.B., see McLean, 1. 179
Uusi-Heikkila, Y., see Nurmi, H. 207
Valenciano, E, see Barbera, S. 80
Van Deemen, A. 223
Van Liederkerke, L., see Lauwers, L. 83
Van Newenhizen, J. 216, 226
Van Newenhizen, J., see Saari, D.G. 197
Vandercruyssen, D. 379
Varian, H.R. 21, 282
Vartiainen, H. 269
Vickrey, W.S. 9, 138, 241, 475
Vidu, L. 58
Villar, A., see Herrero, C. 300
Vinogradskaya, T., see Makarov, 1. 112
Vladimirov, A. 112
Vladimirov, A., see Aleskerov, E 110, 111
Vohra, R., see Dutta, B. 258
Vohra, R., see Serrano, R. 281
Volij, O., see Dagan, N. 305
Vol'skiy, V 98
von Hayek, FA. 7, 14, 15, 18
von Mises, L. 14
von Neumann, J. 22, 62, 548, 581
Wakker, P., see Sarin, R. 522
Walker, M. 81, 241, 258
Walker, M., see Hurwicz, L. 241
Walker, M., see Muench, T. 244
Wang, Y.T 337, 341, 348, 352
Wang, Y.T., see Sprumont, Y. 339
Ward, B. 138, 205
Watanabe, T., see Naiton, W. 110
Watts, H.W. 619
Weber, R. 332, 350
Weber, R.J. 190, 208
Weingast, B., see North, D. 427
Weingast, B.R. 166
Weingast, B.R., see Shepsle, K.A. 166, 210,
433
Weisberg, H.E, see Niemi, R.G. 204, 205
Weiss, J.H., see Ma, B.K. 69
Wettstein, D., see Postlewaite, A. 258
Author Index
Weymark, J., see Chakravarty, S.R. 618
Weymark, J.A. 80, 512, 526, 555, 560, 582,
583, 589, 606, 612, 615
Weymark, J.A., see Blackorby, C. 50, 465,
490, 527, 551, 556, 582-584, 589
Weymark, J.A., see Bossert, W. 465, 514, 531,
551
Weymark, J.A., see Donaldson, D. 560, 615,
622
Weymark, J.A., see Gibbard, A.E 76
Weymark, J.A., see Le Breton, M. 37, 38, 42,
65, 77, 147
Weymark, J.A., see Tsui, K.-Y. 486
Whinston, A., see Loehman, E. 333
Whinston, M., see Segal, I. 272
Whitford, A., see Quinn, K. 445
Whitford, A., see Schofield, N. 445
Wicksell, K. 21
Wilkie, S., see Chakravorty, B. 275
Williams, B., see Sen, A.K. 24, 474, 545,
546
Williams, S. 249, 250
Williamson, O.E. 143
Wilson, R.B. 21, 50, 51, 53, 54, 75, 76, 111,
115, 145, 511
Winslow, G.R. 295
Woirgard, E, see Charon, I. 211
1-11
Wolfson, M.C., see Love, R. 614
Wolinsky, A., see Rubinstein, A. 270
Xu, Y., see Pattanaik, P.K. 25
Xu, Y., see Suzumura, K. 24
Yaari, M., see Bar-Hillel, M. 464
Yaari, M.H. 477
Yamato, T. 254, 260, 262, 281
Yamato, T., see Saijo, T. 258
Yanovskaya, Y. 77
Yilmaz, B., see Sertel, M.R. 217
Yitzhaki, S. 604, 615
Yoshihara, N. 258
Young, H.P. 37, 71, 84, 179, 211-214, 218,
226, 295, 303, 306-309, 315, 333, 348,
350, 365, 374, 380-382, 384-386, 477,
530, 536
Young, H.P, see Balinski, M.L. 184
Zheng, B. 620
Zhou, L. 257, 583
Zhou, L., see Barberi, S. 218, 421
Zhou, L., see Malawski, M. 53
Zhu, D., see Wang, Y.T. 352
Zwicker, W.S. 215
Zwicker, W.S., see Brams, S.J. 201-203
SUBJECT INDEX

absolute Gini index 604 independence of irrelevant alternatives 376,

absolute index of inequality 603 377
absolute k-majority 107 Arrow's impossibility theorem 50, 188, 364,
absolute measures of mobility 617 375
accounting literature on cost-sharing 328 Arrow's paradox 98
acyclicity 61, 135, 159 Aumann-Shapley method 338, 343
acyclicity theorem for NIM rules 63 average cost 319
additively decomposable inequality measures for homogeneous good (ACH) 348
613 averaging rule 161, 164
additivity 323, 331, 336
admissible ballots 176 balanced set of reflexive binary relations 365
admissible individual preferences 366 ballot 176
admissible issues 366 ballot response profile 176
admissible preferences 385 bankruptcy 311
admissible strategy 193, 200 Banks's procedure 211
agenda 75 base relation 76
aggregative inequality measure 614 Bayesian implementation 276
airport landing game 345 Bayesian mechanism design 245
AKS index of inequality 601 Bayesian monotonicity 279
almost-decisive coalitions 153 BDR (Borda decision rule), see Borda decision
a-effective 408 rule (BDR)
a-effectivity function 416 belief aggregation 428
a-winning 408 benevolent dictator 293
alternative vote 391 Bernoulli hypothesis 548, 582
alternatives 39 i-effective 408
anonymity 113, 137, 152, 153, 161 163, 182, O/-effectivity function 416
387, 505, 557, 608 ,/-winning 408
social choice correspondence (SCC) 405 betweenness property 154
social decision rule (SDR) 381 binariness condition 103
social ranking rule (SRR) 382, 383, 388 binary independence of irrelevant alternatives
antagonistic preferences 140, 167 489, 553
antisymmetry 40 binary inversion 158, 159
approval voting 190, 194, 208, 407 binary inversion property 158
approval voting procedure 207 binary multistage voting 199
Archimedean property 387, 388 binary relation 100
Aristotle's proportionality principle 291 antisymmetric 40
Arrovian framework 36 complete 39
Arrow social welfare function 135, 146 148, continuous 57
150-153, 376 quasitransitive 59
Arrow-type impossibility theorems 188 bistochastic matrices 600
Arrowian social welfare function 376 Black decision rule 374
Arrow's choice axiom 76, 102 Black's procedure 211
Arrow's conditions 375, 376 bliss point 59
1-13
1-14 Subject Index

boolean function 103 collective rationality 375

Borda count 379, 386, 399 collegium 62
Borda decision rule (BDR) 364, 371-374, committee 144, 401, 418, 419, 432
377 382, 384-386 dictatorial 403
Bayesian justification 384 representations by game forms on economic
Borda efficiency indices 379 domains 420
Borda efficiency of different voting procedures voting by 421
379 comparability 478
Borda method 472 complete class 122
generalised 473 of closedness 114
Borda ranking rule (BRR) 364, 367, 371, 372, complete generating class 108
375, 376, 380-384, 386 composition 300, 315
Borda rule 153, 363, 364, 379-381, 393, 409 from minimal rights 300
Borda voting procedure 179 of rationing methods 309
Borda's procedure 216 upper 299
bottom layer 72 concerned individual 137, 138, 140, 142
bounded mechanism 263 concerned voter 190
Brown-Banks acyclicity theorem 62 concordance condition 101
BRR (Borda ranking rule), see Borda ranking Condorcet candidate 179, 203
rule (BRR) Condorcet combinatorics 203
Condorcet conditions 378
cancellation 381 Condorcet criterion 379
social decision rule (SDR) 381, 382 Condorcet efficiency
social ranking rule (SRR) 383 social decision rule (SDR) 379
candidates 176 Condorcet efficiency indices 379
canonical mechanism 250 Condorcet jury theorem 451
capacity Condorcet paradox 98, 133, 134
of agent in rationing problem 312 likelihood of 204
of committee 397, 419 Condorcet set 203
cardinal full comparability 552, 579 Condorcet voting procedure 210
cardinal measurability 579 Condorcet winner 139, 143
cardinal unit comparability 552 Condorcet's "other paradox" 215
censored income distributions 623 Condorcet's procedure 211
chain property 41 connectedness 147, 150
chaos 433 consensus functions 81
choice function 100 consistency 304, 332, 334, 381.387
choice problem 403 rationing problem 304
choice-set paradoxes 215 social decision rule (SDR) 381, 382
choose-I voting procedures 183 social ordering rule 384
choose-k procedures 218 social ranking rule (SRR) 383, 388
clockwise cycle 155, 156 consistency axiom 213
closed half spaces 155 consistent fixed path 340
coalition 43, 407 consistent path 337
a-effective 408 constant returns 317
a-winning 408 constrained equal awards 298
/f-effective 408 constrained equal losses 298
/3-winning 408 contagion lemma 44
coalitional oligarchy rule 123 Continental Congress 428
coalitional q-federation rule 119 continuity 154, 160, 161, 163, 166, 214, 305,
coalitionally undistorted social choice function 496, 557
414 contractibility 162-165
Subject Index
convex cone 163
convexity 162, 163
Copeland's procedure 210
core belief 431
core of E 416
core of the choice problem ((N, W), A)
cost function 317
cost-sharing method 317
counterclockwise cycle 155, 156
critical level of utility 575
cumulative voting 218
cycle 39
cyclical balance 155, 156, 159
cyclical indifferences 145, 146
cyclical majorities 134, 143, 179
decisive 43
inversely 43
significantly 61
decisive coalition 43
decomposability 612
decomposition of rationing method 298
decreasing marginal cost bounds 320
decreasing returns 318
decreasing serial cost sharing 327
demand in cost-sharing problem 317
demand monotonicity 339
deprivation functions 625
depth of poverty 620
dichotomous preferences 138, 140, 141, 153,
156, 167
dictator 54, 114, 146-148
dictator rule 123
dictatorial 406
dictatorial committee 403
dictatorial component 54
direct dictator 43
directly dictatorial 43
distance between corresponding income
distributions 616
distance function 604
distribution of preferences 153, 156
distributionally homothetic 604
distributivity 326
Dodgson's procedure 211
domain 40
dominance 607
dominance-based scoring rule (DSR) 364, 386,
388-390
dominance between strategies 191
dominance ranking 607
I-15
dominance solvable 421
dominant strategy mechanism 240
dominated candidate paradox 201
Dred Scott decision 429
DSR (dominance-based scoring rule), see
409 dominance-based scoring rule (DSR)
dual partition condition 185
dual serial method 325
dummy 330, 335
dummy consistency 332
echoic preferences 138
effectivity function 419
first 416
representations of 420
efficacy of a voting strategy 197
egalitarian method 315
election fraud 176
electoral college 431
equal-sacrifice methods 307
equal treatment of equals 296, 332
equality 599
equally-distributed equivalent income (EDEI)
600
equally-distributed equivalent poverty gap
625
equilibrium point 413
strong 414
equivalence relation 180
essential decomposability 152
ethical indices of inequality and poverty 600
Euclidean preference ordering 68
Euclidean space 134, 154, 160, 167
Euclidean utility 436
ex ante approach 524
ex-ante social-evaluation functional 583
ex-ante welfarism 581
ex post approach 523
exactly and strongly consistent social choice
function 414
exactly consistent social choice function 413
expansion equivalence 574
expected utility 608
expected-utility hypothesis 548, 581
extended continuity 573
extended Gini indices 611
extended independence of the utilities of
unconcerned individuals (EIIUUI) 574
extended sympathy 474
extensive form mechanisms 267
extremal restriction 137, 141, 153, 159, 160
1-16 Subject Index

fair division 293, 316 heart 435

fair queuing 314 heredity 101
faithful representation 410 hierarchical dictator rule 107
faithfulness 381 hierarchical federation rule 105
social decision rule (SDR) 381, 382 hierarchical oligarchy 106
favourable composite transfer (FACT) 611 hierarchical representation rule 106
feasible elimination procedure 397, 400, 417 hierarchical syndicate 107
feasible strategy 193 hierarchical veto rule 106
federation 113 high income aggregation property (HIGAP)
Fishburn's procedure 210 616
fixed-path cost-sharing methods 312, 339, hyperplane 155
346 hyperregressive method 308
focal voter 191 hypersaturating preference domain 150
formal welfarism 492
formally welfarist associate 492 impartial culture condition 143
fragmentation 448 implementation 239
free triple 41, 146, 147, 150 double 247
free triple property 41 fault tolerant 281
full domain 40 Nash 242, 247
functional voting rule 112 weak 247, 254
central class 113, 118 impossibility theorems 188
incentive compatibility 241, 278
game form 407 income distribution 599
multi-valued 406 income gap 620
giteau rules 72 income mobility 616
general impossibility theorem 98 increasing-committee-size paradox 218
generalised Borda methods 473 increasing marginal costs bounds 320
generalised Gini family 471 increasing returns 317
generalized entropy family 613 incremental cost sharing 324
generalized limited agreement 139 incremental equity 567
generalized Lorenz curve 610 independence of claim truncation 300
generalized point systems 386 independence of infeasible alternatives (IIF)
generalized value restriction 139 75
Gibbard-Satterthwaite theorem 397, 413, 420, independence of irrelevant alternatives (IIA)
421 43, 147, 148, 150. 372, 375, 387
Gini family, generalised 471 social ordering rule 384
Gini index 560, 602 social ranking rule (SRR) 376, 388
Gini social-evaluation function weak 70
generalized 615 independence of merging and splitting 297
single-series 615 independence of some alternative (ISA) 73
Gini welfare function 604 independence of the utilities of the dead (IUD)
global Borda rule 70 564
global positionality 367 independence of the utilities of unconcerned
grading principle 503 individuals (IUUI) 561
group decision function 141 independence theorem 57
independent decisiveness 78
Hansson independence condition 77 index of inequality 600
Hare rule 391 indifference relation 180
Hare system 220 individual evaluation functions 464
Harsanyi domain 526 individual evaluation profile 466
head-count ratio 620 individual good 545
Subject Index 1-17

individualistic social utility function 80 linear order 180

indivisible units 295 linear ordering 40
inequality 599 linear preference ordering 40
absolute index of 603 local positionality 367
aversion 602 locality 103, 112
Gini index of 602 locality condition 116
intermediate index of 603 Lorenz consistent 610
normative interpretation of indices 599 Lorenz curves 608
normatively significant index of 601 Lorenz-domination 608
relative index of 602 lower bound 301, 338
Theil index of 613 lower continuity 81
information invariance 551, 578, 579
information restriction theorem 73 majority candidate 179
informational basis 478 majority rule 43
informational coherence 341 semi-strict 141
inheritance problem 295 simple 57
intermediate index of inequality 603 strict 141, 144
interpersonal comparison of utilities 599 super 58
intersection principle 74 majority vector 156, 157
intertemporal anonymity 563 majority voting 473
intertemporal continuity 563 majority-voting principle 374
intertemporal independence of the utilities of majority winner 58
unconcerned individuals (IIUUI) 564 majorization 608
intertemporal strong Pareto (ISP) 563 marginalism 333
invariance transformation 479 maximin 559
inverse dictator 43 maximin index 603
inverse ordering 154, 156, 157 maximin principle 469
inversely decisive 43 meaningful statements 478
irrelevance of reallocations 297 means 163, 165
issues 364 measurability 478
measure of dispersion 599
jury theorem 427 measurement invariance 341
justice, theory of 545 mechanism 247
median voter 68
(kl,k 2)-majority 107 Miller's procedure 210
k-majority 114 min-of-means family 514
Kaplan normalisation 470 minimal equity 557
Kemeny's procedure 211 minimal individual goodness (MIG) 554
minimal individual symmetry 506
latin square 133, 153, 155, 158 mobility index 618
Leviathan 427 monotonic social choice function 183
lexicographic composition 105 monotonic transformation 248
lexicographic dictator 469 monotonicity 113, 248, 250, 273, 339, 374,
leximax 560 387, 405, 620
leximin 559 runoff systems 391
leximin rule 469 score-based decision rule (SCDR) 390
limited agreement 138, 142, 159, 160 score-based ranking rule (SCRR) 388 390
Lindahl correspondence 258 social decision rule (SDR) 381, 392
constrained 258 social ordering rule 384
Lindahl rule 240 social ranking rule (SRR) 383
line of maximum poverty 624 strict 308, 374
1-18 Subject Index

strong 251 norms 580

symmetric 349 not-strictly-best (NSB) value restriction 145
Moulin-Shenker cost-sharing method 353 not-strictly-worst (NSW) value restriction
multi-stage majority decision 141, 142 145
multi-valued game form 406 nucleolus 304
multinomial probit (MNP) 443 null component 54
multiple-districts paradox 202 null ordering 65
multiple-election paradox 202 null rule 42
multistage nonranked voting procedure 198 numerical full comparability (NFC) 553, 579

Nakamura number 109, 410 oligarchy 59, 114

Nanson runoff system 391 oligarchy theorem 59
Nanson's procedure 211 2 representation 103
Nash bargaining solution 472 one vote 113, 118
Nash equilibrium 441 one vote rule 105
trembling-hand perfect 262 ordering 39
undominated 260 ordinal full comparability (OFC) 579
neutrality 69, 113, 137, 152, 182, 378, 381, ordinal properties 38
387, 571 ordinality 348
social decision rule (SDR) 381 ordinally proportional rule 352
social ordering rule 384 outcast condition 102
social ranking rule (SRR) 382. 383, 388 outcomes 39
strong 493 overall comparisons 367
new-voter paradox 219
NIM (neutrality, independence, and monotonic- pairs
ity) 63 decisive coalition 43
no advantageous reallocation 297 fixed 55
no free lunch 319 independent 56
no-show paradox 202, 219 nontrivial 147, 149, 150
no subsidy principle 350 self-sufficient 74
no veto power 248, 250 trivial 147 150
non-dictatorial 43, 76 pairwise comparisons 367
non-imposition 113, 117 pairwise consistency 306
non-negative responsiveness 141 paradox of voting 57. 98, 179
non-null 43 parafinite CW complex 162, 164
non-null simple game 411 parametric rationing methods 305
nondictatorial decomposability 149 Paretian social choice correspondence (SCC)
nondictatorial functions 188 405
nondictatorship 148, 151, 161 Paretian welfare economics 599
nondictatorship condition 187 Pareto, strong 557
nonimposition 42 Pareto, weak 557
nonmanipulable voting procedure 151 Pareto condition 147, 148. 152, 161, 162
nonranked voting 189 Pareto criterion 42, 377
nonranked voting procedure 189 weak 375
nontrivial pair 147, 149, 150 Pareto decisiveness 64
normalized poverty gap 624 Pareto dominance 187, 499
normative conditions 113, 117 Pareto extension 49
normative interpretation of inequality indices Pareto extension rule 153, 154
599 Pareto indifference 489, 554
normatively significant index of inequality Pareto-optimal 404
601 Pareto optimality 76, 377
Subject Index 1-19

Pareto plus principle 576 to lower demands 312

Pareto weak preference 554 private alternatives 150, 151
partial order 206 probabilistic rationing 314
partial Pareto rule 123 profile 40
partial q-Pareto rule 119 profile-dependent welfarism 490
party-list voting 223 progressive transfer 607
path-generated cost-sharing method 336, 342 progressivity 299, 307
path independence 102 property of cancellation of a social ordering rule
PDR (plurality decision rule), see plurality 384
decision rule (PDR) property of strong positive association 405
perfect complements 345 proportional electoral rules 430
permissible individual preferences 375, 378 proportional electoral systems 445
Pigou-Dalton principle 506 proportional rationing 297
Plott's condition 102 proximity 158, 160
plurality 430 proximity preservation 162
plurality decision rule (PDR) 373, 374
plurality electoral systems 451 q-one vote rule 121
plurality voting 190, 399, 409 q-unanimity rule 121
plurality with a runoff 198 quandary 429
plurality with successive elimination 199 quasi-agreement 153
point distribution procedures 218 quasi-dictator 64
political equilibrium 443 quasi-transitivity 139-141, 159
Pollak domain 467 Quebec Act 428
population anonymity 572 quota 220
population binary independence of irrelevant
alternatives 572 random-order value 331, 348
population ethics 571 random-priority method 303
population Pareto indifference (PPI) 572 rank 109
population principle 610 rank of an alternative 367
population size 600 ranked relative deprivation 621
population substitution principle (PSP) 562 ranking 299, 345
population unrestricted domain (PUD) 572 rational 466
position of an alternative 367 trivial 466
positional scoring procedures 212 ranking paradoxes 215
positional scoring vector 212 rational ranking 466
positionalist voting rules 364 rationing 295
positive responsiveness 137, 152 rationing method 295
potential function 333 rationing problem 294
poverty 599 Rawls-Pazner-Schmeidler family 533
depth of 620 reduced profile 156, 159
poverty line 620 reduction procedure 157, 158
preference aggregation 428 regressivity 299
preference ordering 404 regular committee 418
preference plateau 206 relative index of inequality 602
principal of the party 442 relative k-majority 107
principle of dominance 388 renegotiation 269
principle of Pareto optimality 377 renegotiation function 271
principle of simple majority 375, 378 renegotiation monotonicity 273
principle of transfers 607 representation 114
priority 303 representation of G 410
to higher demands 312 representational imbalance 225
1-20 Subject index

representative systems 185 SCRR (score-based ranking rule), see score-

representative utility 558 based ranking rule (SCRR)
repugnant conclusion 575 SDR (social decision rule); see social decision
resource monotonicity 296 rule (SDR)
retract 162-164 second-order stochastic dominance 609
revelation mechanism 278 self-duality 302
revelation principle 240, 278 Sen’s majority decision theorem 66
reversibility of fairness 326 Sen’s transfer axiom 621
rich domains 254 separability axiom 496
robustness 152, 153 separable preference ordering 58
runoff 209 serial cost shares for homogeneous good 349
runoff procedure 386 serial cost sharing 321, 339, 344
based on the Borda ranking rule (BRR) serial dictatorship 5 1
3 78 Shapley value 332
runoff system 364, 390, 391 Shapley-Shubik cost sharing 324
based on the anti-plurality ranking rule Shapley-Shubik method 338
391 sign hnction 184
based on the Borda decision rule (BDR) simple game 144-146, 401; 402, 407, 408,
391 410, 411
based on the Borda ranking rule (BRR) 375, non-null 4 11
378, 390 proper 144, 145
based on the plurality ranking rule 391 properties 402
based on the plurality rule 391 strong 144, 145
Nanson 391 simple majority decision 133, 136, 141, 143,
156, 158, 159
satiation 162 simple majority function 184
saturating 41 simple majority principle 364
saturating domain 147, 150 simple majority rule 363, 378
saturating preference domain 147, 148 sincere strategy 200, 208
saturating preference ordering 147, 150 sincere voting 195
scale independence 603 single-caved preferences 138, 155
scale invariance 308 single-parameter Gini 560
scaled order statistic 619 single-peaked preferences 68, 138, 149; 155,
SCC (social choice correspondence), see social 167, 180, 198,206, 316
choice correspondence (SCC) single preference profile approach to formal
SCC (strong Condorcet condition), see strong welfansm 467
Condorcet condition (SCC) single transferable vote (STV) 179, 220, 391
SCDR (score-based decision rule), see score- singular committee 418
based decision rule (SCDR) Slater’s procedure 21 1
SCF (social choice fimction), see social choice Smith’s Condorcet principle 2 12
function (SCF) socia1 choice correspondence (SCC) 75, 404,
scheduling 3 13 405,407, 408, 410
Schwartz’s procedure 21 1 anonymous 405
5’-concave function 600 monotonic 405
5’-convex function 600 neutral 405
score-based decision rule (SCDR) 364, 373, Paretian 405
374, 377-379, 386-388 tight 408
score-based ranking rule (SCRR) 364, 373, social choice from a feasible set of alternatives
374, 377, 379, 386-388, 390, 392 366
score-based rules 386 social choice function (SCF) 8 1, 180, 181,
scoring vector for m-alternative issues 374 406, 413, 415
Subject Index
coalitionally undistorted 414
dictatorial 406
exactly and strongly consistent 414
exactly consistent 413
first effectivity function 416
nonmanipulable 4 13
strong representation of G 415
strongly neutral 185
undistorted 413
social choice rule (SCR) 239, 246
acyclic-valued 62
social choice set (SCS) 277
social decision function 133, 135-137, 139,
140, 146
social decision rule (SDR) 364, 366
based on painvise comparisons 370
local 103
positionalist 370, 379
satisfying independence of irrelevant
alternatives 370
strong positionality 370
weakly positionalist 370
social-evaluation functional 546, 550
social-evaluation ordering 555, 599
social good 545
social ordering 38
social ordering rule 366, 376
social ranking rule (SRR) 366
Archimedean property 387, 388
based on pairwise comparisons 368
positionalist 368
satisfying independence of irrelevant
alternatives 368
strongly positionalist 368, 369
weakly positionalist 368
social rankings of the alternatives 366
social utility function 80
individualistic 80
social welfare function 40, 133, 135-140,
146-153, 162, 600
component of 54, 55
computable 83
homothetic 604
quasitransitive-valued 59
transitive-valued 40
social welfare functional (SWFL) 467
imposed 468
social welfare ordering (SWO) 492, 555
sophisticated strategy 178, 200
sophisticated voting 200, 421
1-21
SRR (social ranking rule), see social ranking
rule (SRR)
stand-alone core 350
stand-alone cost 320
stand-alone lower bound 320
stand-alone upper bound 320
stochastic dominance 609
straightforward strategy 200
strategic manipulation 177, 393
strategic voting 177
strategy-proofness 177, 195, 254, 316
strict agreement 140
strict nonimposition 61
strict ranking 308
strict simple majority relation 203
strong aggregation invariance (SAI) 349
strong Condorcet condition (SCC) 378
strong connectedness 147
strong n-majority q-Pareto rule 121
strong neutrality 493, 555
strong Pareto 557
strong Pareto optimality 76
strong positive association (SPA) 405
strong positive responsiveness 64
strongly neutral social choice function 185
structural invariance 292
subadditive cost function 349
submodular cost function 351
successive majority voting 199
sufficient, definition 73
summation social choice functions 387
supersaturating preference domain 150
Snppes dominance 504
Suppes grading principle 503
surplus sharing 3 15
symmetric global mean of order Y 559
symmetrically central class 113, 118
of rules 105
symmetry 296,402,405, 608
symmetry axiom for population (SAP) 610
synmcate 114
taboo preferences 138
Talmudic method 303
r-system of ( k , , k 2 ) majorities 108
r-system of (t,k)-majoriv fanlilies 121
taxation 295, 307
Theil index of inequality 613
third-order stochastic dominance 609
top layer 72
topological structure 134, 154
1-22 Subject Index

tournament 66 Vandercruyssen findings 379

trade-off theorem 55 variable population 332
extreme 57 variable-population social-evaluation functional
quasitransitivity 61 572
tragedy of the commons 327, 344 vanable-population welfarism 572
transfer sensitivity 611, 625 veto power 59
transition matrices 617 vetoers 402
transitive relation 39 virtual implementation 264
translatability 604 VNM preference ordering 525
translation-scale measurability 552 voter preference profile 180
trivial alternatives 41 voter strategy 177
trivial ranking 466 voting by committees 421
voting by veto 421
ultrafilter 47 voting by veto rules 397
fixed 51 voting cycle 428
'
free 83 voting game associated with F and R 412
ultrafilter lemma 47 voting paradox 398
unanimity 113, 146, 148, 161-165 voting procedure 176
unanimity lower bound 321 voting schemes 421
unanimity rule 105, 118
unanimity upper bound 321 Walrasian correspondence 258
uncertainty 431 constrained 258
unconcerned individual 137, 156 Walrasian rule 239
unconcerned voter 190 WCC (weak Condorcet condition), see weak
uncovered set 435 Condorcet condition (WCC)
undistorted social choice function (SCF) 413 weak I-majority q-Pareto rule 121
uniform gains 298 weak Condorcet condition (WCC) 378, 379
uniform losses 298 weak dictatorship 469
unit invariance 341, 346 weak expansion equivalence (WEE) 574
universal bounds 322 weak order 180
unrestricted domain 553 weak Paretian rule 116
upper bound 301,349 weak Pareto 557
upper contour set 116 weak Pareto condition 148
US Constitutional design 451 weak positionality 367, 370
utilitarianism 545-591 social decision rule (SDR) 371
average 575 social ranking rule (SRR) 369, 370
classical 575 weak transfer axiom 621
critical-level 548, 575 weak unanimity 51
critical-level generalized 548 weak-welfarism 501
generalized 547, 559, 560 weakly positionalist
pure 470, 511 social decision rule (SDR) 371-374, 377,
relative 470 378
weak 558 social ordering rule 376
weighted 470, 508 social ranking rule (SRR) 376, 377
weighted-rank 471,512 weakly utilitarian 558
utility 599 weighted-gains method 310
weighted-losses method 310
value distinction 261,267 weighted majority function 184
value judgements 599 weighted majority game 403
value restriction 65, 138, 142, 146, 152, 153. weighted majority rules 153
159, 160 welfare rankings 599
Subject Index 1-23

welfarism 476, 490, 492, 501, 553 Wilson's theorem 53

ex ante 581 winner-turns-loser paradox 201
ex post 581 winning coalition 144, 145, 407
welfarism theorem 493, 555
yolk 434
welfarist 545
Young's procedure 212
well-being 545
Wilson's partition lemma 54 zero consistency 301