Documente Academic
Documente Profesional
Documente Cultură
The aim of this paper is to clarify the meaning of comparative urban analysis, to
distinguish three approaches to it, and to argue for the importance of two of them
which has not been recognized in the past. Although the examples are drawn from
urban and regional studies the arguments made apply equally to comparative
anslysis in other fields.'
The paper is divided into five parts. First we consider the uses of comparative
studies, and distinguish between comparative research and comparative analysis.
We then present some conflicting views about the execution of comparative analysis
which reflect different theoretical positions and different ideas about underlying
causal relations in society. Section I11 uses examples to introduce the three ap-
proaches to comparative analysis. Two of these involve the idea of plural causation
and the importance of this concept and its implications in urban and regional
studies are stressed in the fourth section. We conclude by discussing the scope of
comparative analysis in the light of the earlier arguments.
This paper is a completely revised version of one first read at University College, London in
1980. It has benefited from commcnts made then and subsequently at the University of Kent,
University of Bristol (School of Advanccd Urban Studies), Institute of British Geographers
Conference, Leicester 1981, the ESRC/CNRS Symposium o n the Comparative Analysis of
Arab and Muslim Cities at Birkbeck College, London, 1984, and the Fifth Urban Change and
Conflict Conference, University of Susscx, 1985 1 would also like t o thank Nick Manning,
Bryan Roberts and John Walton for their helpful comments.
C.G.Pickvance 163
Firstly, comparative studies help one become aware of the diversity of social
phenomena and overcome ethnocentric assumptions about what is normal. For
example, in the housing field most British laypersons would assume that housing
policy is concerned with the distribution of housing, e.g. the provision of consump-
tion subsidies such as tax relief for owner occupiers, and housing benefits to tenants;
that council housing is the normal form of non-market housing provision, and that
employers have no role in housing provision.
However, exposure to the housing situation in other countries reveals that
housing policy may be more about maximizing the production of new houses, than
about consumption, as in the USA and Canada for example (Heidenheimer e f al.,
1983). Likewise it turns out that Britian is unusual in the large role of councils in
housing provision. All other advanced capitalist societies have a non-market housing
sector but this usually consists of non-government bodies - largely depen-
dent on government funding - like housing associations in Britain (Wynn, 1983).
Finally, the French experience reveals a much greater role of employers in housing
provision. All firms above a minimum size have to pay 0.9%(originally 1%) of their
payroll as a tax for housing. They also have nomination rights in cases where these
funds are subsequently used to build social housing (Pickvance, 1980, Appendix).
In Britain, employers only have a formalized influence in the allocation of new
town rented housing.
Awareness of this diversity produces a sort of culture shock. It makes one aware
of new and unsuspected connections. (The implications of this for comparative
analysis are dealt with later.) It also leads to a second reason for comparative
studies: to clarify ones theoretical interest in a topic. The act of comparison forces
one to consider what is essential and what is inessential among the observed differ-
ences between, say, forms of housing provision in two countries. How one draws
this line depends not on an inherent property of housing, but on ones theoretical
interest in it. For example, discovery that council housing in Britain is unusual in
being council-controlled makes one question whether this is its essential feature
or whether (more likely) ones interest in it is as an example of non-market housing
provision. If so then it can be placed in the same category as US public housing or
French social housing - and British housing association housing. This would reflect
a theoretical judgement that what is most important is its (non-market) economic
role and that the diversity of institutional forms can be abstracted from, i.e. ignored,
for the purpose at hand. In these ways, awareness of diversity through comparative
studies forces one to bring theoretical assumptions into the open.
Finally, comparative studies may be carried out in order to undertake compara-
tive analysis. This may seem an obscure statement to make. Are not comparative
studies or comparative research synonymous with comparative analysis? In my
view, no. Indeed I would see terms like comparative research and comparative
studies as misnomers. The fact that a study is based on data relating to two or more
societies is no guarantee fhat it is a comparative one. Such data are a necessary but
not sufficient condition of comparative analysis. It is quite possible to juxtapose
164 Comparative urban analysis and assumptions about causality
data from two societies, or as Walton says, give them serial treatment (1981,30).
For example, Duncans (1981) discussion of the differences between Swedish and
French housing is perfectly adequate to his objective of denying Castellss claims
about the generalityof the French-based housing model (1977, 156, 158). But it
involves only the serial treatment of the two countries, i.e. it goes as far as identi-
fying their similarities and differences. A comparative analysis on the other hand
goes beyond this and attempts to understand the two (or more) cases in terms of
one or more models. It is for this reason that it is only analysis which can be termed
comparative.2 The term comparative research is therefore an illegitimate one.
To clarify the meaning of comparative analysis let us return to the observation
of diversity in housing provision as mentioned above. At first sight it would appear
to mean that n o single explanatory model3 is applicable, that different processes
are at work in each country. More formally this amounts to saying that differences
between countries in the value of the dependent variable (whose variation we are
trying to explain) cannot be explained within a single explanatory model.
But hopefully the absurdity of this proposition is self-evident. It amounts to
saying that when two countries both have 60% owner occupation the same model
applies, but if one rises to 70% and the other remains at 60%, different models are
necessary. In fact, of course, explanatory models can explain variation in the
phenomenon of interest provided it takes certain forms. Let us consider, for example,
Harloes argument that owner occupation, at least vis-u-vis the mass production of
housing, only emerges when substantial proportions of the potential consumers
have long-term stable earnings (1981a, 30). Clearly this proposition is capable of
explaining differences between countries at one point in time, and within one
country over time, It leads us to expect countries to show a diversity of levels of
owner occupation (Y), in conformity with the proportions of their population
with long-term stable earnings (X), or Y = fi (X), where fi is a function. Before
we can establish whether crossnational data conform to this model or not, its
terms would have to be operationalized and the shape of the function linking
earnings and owner occupation specified, e.g. is it linear, and if so with what shape,
is it curvilinear (U-shaped, tapering, J-shaped, etc.)? Assuming this were done,
there are three possibilities, each of which corresponds to a type of comparative
analysis.
This point is clearly made by Zelditch who states that any explanatory generalizing research
involves comparison (1971, 27!). It is the act of analysis in which one seeks the causes of a
phenomenon which involves comparison. Thus even an analysis of a single case involves com-
parison since one compares the observed situation with an imagined situation in which the
suspected explanatory factorls) are absent (Smelser, 1976, 160-62). Comparative analysis, as
discussed here, assumes that data are gathered in two situations.
I make n o apology for using the term model. It has recently become popular t o decry notions
like models. levels, abstraction, etc. on the grounds that they involve simplication and artificial
separations. Since the complex totality of social reality is not graspable directly, simplifying
and necessarily artificial concepts and models are unavoidable. though argument over particular
concepts is always possible. Those, like some realists, who claim to be able t o d o without
them. have never revealed to what they owe their privileged access to the totality.
CG. Pickvance 165
Firstly the data might turn out to be exactly as postulated by the model. The
result would be a (two-variable) single model, Y = fi(X) which applied to all
countries: the rationale would be that owner occupation spreads as effective consumer
demands expands. This is referred to below as comparative analysis using a single
model. Secondly it might turn out that the relationship between earnings and owner
occupation took different forms in societies with different levels of industrialization,
e.g. Y = fz ( X ) and Y = f a Q This would amount to a reformulation of the original
model provided that the difference between f~ and f3 was interpretable within the
same market rationale. The two relationships could be rewritten as a single model
Y = f,(ZJ), where I is industrialization level. What is distinctive compared to the
first case is that we have had to add a societal characteristic, I , into the initial model.
This case is referred to below as comparative analysis using linked submodels, (I
describe fi and f3 as submodels because they share the same rationale, i.e. owner
occupation is related to stable earnings in some form).
A third possibility is that in two groups of societies the level of owner occupation
depends on quite different variables. For example that in advanced capitalist
societies it depends on the level of stable earnings, Y = f~( X ) , but in state socialist
societies it depends negatively on the scale of state house building (Z), Y =fa (2).
The latter case would represent a difference of rationale with the extent of owner
occupation reflecting the scale of state socialist housing production (gradually dis-
placing owner occupied prewar housing) rather than any market rationality. The
social process here is fundamentally different from the first two cases. This case is
referred to as comparative analysis using diverse models.
Thus it can be seen that diversity at the level of phenomena may be compatible
with a single model or with linked submodels, or may require two (or more) distinct
models to be formulated? All three cases will be illustrated in Section 111.
What do these three approaches to comparative analysis have in common? The
defining characteristic of comparative analysis, as understood here, is that it allows
the investigation of influences on the phenomenon or relation of interest which
are given or constant for one society at one point in time by studying cases in
which they vary. Its success depends on the range of variation naturally occurring
in the societal features in question - given that experimentation is excluded. Some
variation in societal features can be observed by studying one country over time,
e.g. presence and absence of democratic political systems, or changing position in
the world economy. But the range over which these features have varied may be
narrow, and the covariation between them makes it difficult to separate out their
The three approaches distinguished here are quite different from those identified by Skocpol
and Soniers ( 1 980). All three belong to their category of macrocausal analysis. (lheir
contrasturiented category is a form of serial treatment - it does not involve comparative
analysis. And their parallel comparative category has some resemblance to my single model
approach but is concerned to illustrate the theory rather than expose it to test and refinement.)
Comparative analysis may be searching for societal influences o n a phenomenon such as
owner occupation, or a relationship such as that between income and owner occupation
(Irtcworski and Teune, 1970, 39-46).
166 Comparative urban analysis and assumptions about causality
independent effects. (For this latter reason some writers advocate the study of the
influence of syndromes of social characteristics on the phenomena or relationship
of interest.)
The advantage of crossnational comparisons is that they allow a wider range of
variation to be observed in the societal characteristics of interest, and avoid some
of the covariation found in single societies. But the risks are twofold. Firstly the
phenomenon or relationship may have a quite different meaning in another
society. This raises the question of cross-societal equivalence of concepts which
is not discussed here (see Smelser, 1976). Secondly, our lesser familiarity with other
societies makes it more hkely that some feature unknown to us has an important
causal influence on the phenomenon or relationship of interest. The literature on
comparative analysis sets out two solutions t o this problem. One is t o make a
random choice from among societies showing great diversity of societal features in
an effort to randomize the effect of unknown societal features. This allows confi-
dence limits to be attached to the conclusions reached. This approach is favoured
in crossnational attitude or behavioural studies. A more common approach in urban
and regional studies is to restrict the societies included t o those with certain similari-
ties ( e g advanced capitalist, high real income, with democratic political systems).
This is the approach advocated by Walton (1973) and termed standarized case
comparison. The advantage is that while enabling one to study the phenomenon of
interest in the presence of diverse societal conditions the likelihood of unknown
conditions is reduced. These two approaches have been labelled the most different
systems and most similar systems approaches by Przeworski and Teune (1970).
They correspond to J.S. Mills Method of Difference and Method of Agreement.
There are two common reasons for conducting a comparative analysis: t o dis-
cover whether a theoreticallyderived model holds in empirical cases and t o see
whether an empirically-based relationship derived from the study of one society
holds in another. The former applies particularly to the recent popularity of com-
parative analysis in the urban studies field as part of the reaction to general theory.
The 1970s was a period of grand theory when theoretically-derived models - and
even single values of variables - which at best corresponded to realities in one
country, were advanced as having general applicability. For example, as mentioned
earlier, Castellss theoretical analysis of housing, formed with France in mind (1977,
145-69) was explicitly presented as having a wider validity, and Harloe (1979) and
Duncan (1981) quite rightly took issue with this claim. Comparative analysis is a
necessary complement to such theorizing since it gives priority to testing the model
over its whole scope of application (Willer, 1967, Chapter 6).
Having suggested three reasons for comparative studies, and argued that only
comparative analysis deserves its name, the rest of the paper is concerned with the
latter.
C.G. Pickvance 167
We now examine some conflicting views about the types of subject comparative
analysis can be applied to, and the types of research design appropriate to it. These
reflect basic differences of theoretical position and beliefs about underlying patterns
of causality .
In their important textbook on comparative analysis, Przeworski and Teune
(1970) show no hesitation in proposing its application to individual behaviour or
attitudes. More cautious souls would set their sights lower and apply comparative
analysis only to macrophenomena, e.g. structural features of societies such as level
or pattern or industrialization and urbanization, or aggregate behaviours such as
level of migration or residential mobility. Essentially this difference of attitude
concerns the sort of regularities one believes to exist in society. Behaviourists insist
that crossnational regularities are to be found even at the level of individual be-
haviour. In other words they are very optimistic about the levels of disaggregation
at which laws are likely to operate. Structuralists on the other hand expect that
regularities will only be found at the macrolevel, such as structural features of
society, and that the latter set limits to individual behaviour but do not determine it.
Przeworski and Teune distinguish between what I shall call universal and
relativist patterns of social determination.(Their terms are generalist and relativist.)
In the former case it is believed that the underlying causal model is sufficiently
general to warrant an Q priori abstraction from spatiotemporal parameters, whereas
j n the latter it is expected to be highly specific to particular social systems so that
the generality of statements concerning social reality is inherently limited (1970, 8).
They suggest that the universal pattern may apply to psychological processes and
the relativist pattern to social and cultural processes.
Relating this to our previous discussion, if a universal pattern prevails, then no
societal characteristics (beyond X and Y , the variables of interest) enters the model,
i.e. a given effect always has the same cause. This corresponds to comparative
analysis using a single model. If a relativist pattern prevails, different groups of
societies require different models. (Models may differ either because the same vari-
ables enter but with different weights, or because they contain different variables.)
Depending on the nature of the difference, the models may be a) linked submodels
(i.e. where one or more societal characteristics enter the original model) or b)
diverse models. These correspond to the second and third types of comparative
analysis. In these cases similar effects can have different causes in different places
(or times).
The universal/relativist distinction is an extremely important one, and has not
received the attention it deserves. Ones beliefs about the universality of underlying
causal patterns have major implications for the choice of research topic and the
design of research. The key issue in the first respect is that of the acceptable degree
of abstraction, i.e. how much of the social whole can be ignored when studying a
topic. Abstraction is inevitable in order to cope with complexity, but there are
168 Comparative urban analysis and assumptions about causality
This is a quite remarkable claim since it fails to specify any relationship explaining suicide.
The claim would only be valid if a) the three societies have similar positions on the variables
explaining suicide and b) a relativist model of causation was ruled out.
It might be objected that in practice the initial choice between most different systemsand
most similar systems research designs is irrelevant since each allows the actual pattern of
determination to emerge in the end. Whether this is the case is discussed at the end of the
next section.
170 Comparative urban analysis and assumptions about causality
His starting point is the Burgess theory of concentric zones and he makes its
claims quite clear: it is an ideal type representing the spatial structure which would
occur under conditions of radial growth and in the absence of distorting factors
such as topography. This is very different from the popular idea that it describes
actual spatial structure in every city (in America or worldwide). Schnore then
follows Quinn in listing a series of assumptions about cities which were implicit in
Burgesss theory as initial conditions and thereby limited its applicability: ethnic
and class heterogeneity, a commercial-industrial economic base, private property,
market allocation of land, and efficient transportation in all directions, a single-
centred city in which demand for land was highest in the centre, and upper-class
preference for peripheral residence. This list makes clear that Burgess had US cities
in mind. In drawing it up, Schnore is suggesting pointers to his own more general
theory. Schnore then examines studies of class residential patterns in Latin American
cities, and discovers that uniformly they reveal a pattern in which the highest
classes live centrally, though there is some evidence of a trend for them to live in
the periphery.
Faced with this divergence between US and Latin American spatial patterns, a
common reaction would be to conclude that the Latin American evidence disproved
the Burgess model. (This would involve a real misunderstanding of how to handle
diverse phenomena, since it implies that a model is constituted not by a relation
between X and Y , but by a particular value of Y . ) Where Schnore innovates is in
rejecting this obvious response, and opting for a single model which embraces
both the US and the Latin American evidence. To start with he notes that the
Latin American spatial pattern has been widely reported outside Latin America
in preindustrial cities. This comparison follows the most different systems design
and influences Schnores choice of explanatory factor, since it rules out culture
and brings into play industrial technology. (One suspects this was something of a
relief for Schnore since as a human ecologist he treats culture as epiphenomenal!)
Schnore then proposes an evolutionary model in which a new variable, technology,
takes the place of radial growth and explains the displacement of the Latin Ameri-
can pattern of residential stratification by the US pattern. (This is presented as a
universal process, and in empirical support he mentioned traces of the preindustrial
pattern in US cities.) More specifically technology influences class residential distri-
bution through a) competition for central land from emerging commercial and
industrial development which makes it less attractive for upper-class residents, and
b) new transportation technologies which allow the elite good access to central
locations, which they need to exercise economic and political power, without
actually living there. Thus the effect of Schnores analysis is to select technology as
the key hidden variable in Burgesss argument and make it the keystone in a new
model of residential stratification with universal application.
At best technology is a partial explanation, since its introduction depends on class
and other social interests, andits conditioningeffect on social life is subject to many
mediating influences. Human ecology explicitly discounts such processes and
Schnore admits that his model points to the necessary but not sufficient conditions
C.G.Pickuance 171
of spatial patterns (1965, 387). But whatever ones views of the substance of
Schnores model (see Yujnovsky, 1975; Gilbert, 1982) it can only be regarded as a
fine (and ambitious) illustration of how comparative analysis using a single explana-
tory model can comprehend divergent observations. But an evolutionary assumption
is not a necessary part of single models in comparative analysis.
comparisons of Australia, New Zealand and Iceland (as primarily mercantile econo-
mies) and European countries (as primarily industrial economies). Since the war he
expects some degree of convergence with the general spread of corporate capitalism.
The best evidence would be prewar, before the counteracting effect of corporate
capitalism had become pronounced. In fact the data he uses is exclusively postwar. It
shows a) a consistently higher level of owner occupation among the three mercantile
countries in 1970 compared with the west European countries (Table 1) and b) a
convergence between Britain and Australia since 1947: for Britain the rate of owner
occupation rose from 9% (1919) to26%(1947)and 50% in 1971 while for Australia
it rose from 50%(191 1) to 53% (1947) and 69% in 1971 (Mullins, 1981,70).
Iceland 71 France 43
Australia 69 Sweden 36
New Zealand 68 West Germany 34
Belgium 56 Switzerland 28
Britain 50
Secondly, Mullins contrasts the level of owner occupation and sewerage provision
among Australian state capitals over time. The justification for this is that the
argument developed for intercountry comparisons should also apply to intercity
comparisons since the Australian states had very different histories of development.
The data in Table 2 show that, prior to 1939, in the two more industrialized centres
(Sydney and Melbourne), levels of owner occupation were lower and levels of
sewerage provision higher than in the two less industrialized centres (Perth and
Brisbane), exactly as predicted. Since 1945 there has been a large degree of con-
vergence, again as Mullins's argument about corporate capitalism would predict.
l'able 2 Extent of owner occupation (0) and domestic sewerage (S) provision
in major Australian cities, 19 1 1-7 1
Source: Mullins (1981, Tables 1 and 2). n.b. Owner occupation data 1947: sewerage
data 1951.
C.G. Pickvance 173
Like Schnore, Mullins shows how comparative analysis can handle diverse ob-
servations. Mullinss three submodels are linked because all share an underlying
rationale in which capital accumulation is the shaping force - but he makes no
evolutionary assumption. His work is an excellent example of comparative analysis,
in particular because it tests an argument at crossnational and crosscity levels.
Mullins would, however, be the first to admit that he is theorizing in an area where
there is little previouswork-but see Szelenyi(l981,584-89),Berry(1983; 1984).
Before leaving Mullinss work it is worth examining the criticism levelled at it by Kemeny
(1983) and Mullinss response (1 983). Kemenys major objections concern the explanatory
model used, and the choice of data to be explained, Kemeny argues that Mullinss model of
stages of capitalist development is economistic and classifies but does not explain, and suggests
instead that class conflict should be made central. This is not a damaging criticism. For pur-
poses of exposition Mullins emphasizes stages, but it is clear that class conflict is the motor
of his model bringing about evolution between stages according to the particular internal and
external conditions of the country concerned (Mullins, 1983). At best Kemeny suggests the
need to bring out more clearly the mediations between economic structure and urban struc-
ture. At worst he implies that social classes are free-floating entities unrelated to economic
structure. (It is notable that by focusing on the postwar period Kemeny bypasses the issue
of stages.)
Kemenys criticisms of Mullinss choice of data relate exclusively to the crossnational
comparisons. He makes no comment on the crosscity comparisons, let alone advance an alter-
native analysis. Firstly Kemeny emphasizes the variation in recent home ownership rates in
European countries - see Table 1 - and the fact that in some cases they have fallen since the
war. These are not basic criticisms since a) the most relevant evidence on the Mullins thesis
would be prewar, before convergence under corporate capitalism set in, and b) he is now seek-
ing to explain the detailed evolution of owner occupation. but only broad trends. Secondly.
Kemeny points out that the USA and Canada have owner occupation rates of over 60%.
Several responses t o this are possible. The weakest would be to quibble about the precise
figure. More sensibly one might ask what capitalist development in these countries would lead
one to expect. In fact in both countries, despite regional variation, urbanization has been
C ~ O ~linked
Y to mercantile development centred around staples production (Roberts,
1978b. 611-19). This suggests that the US and Canadian data support Mullinss argument
since high owner occupation is a feature o f such developnient.
I t should be noted that Mullinss submodel of urbanization under industrial capitalism
assumes that population concentration, a strong working class, and state provision are found
together. Robertss ( 1981 ) article on comparative industrialization in Manchester, Barcelona
and Lima, however, shows that in the first two cities the need for collective goods and
services arising from industrialization was to some extent met by industrial firms themselves.
174 Comparative urban analysis and assumptions about causality
if more similarities were found. Nor d o I agree that it would be helpful to this
debate to study urban developments in societies with rather differing levels of
production to see whether the latter did in fact have a powerful influence on the
former factors (Harloe, 1981b, 187). Rather, the key point at issue concerns
theoretical assumptions regarding the processes by which empirical similarities are
produced. By proceeding from a universal model of determination - the idea that
similar phenomena musr have common causes - Pahl precludes the possibility that
they might have different causal patterns, and Szelenyis major step forward is to
show how this can happen.
But, one might ask, why is there a problem here? Surely causal processes are
observable and one should be able to resolve empirically whether level of technology,
level of production, bureaucracy, etc. or mode of production and distribution is
the key explanatory factor. Unfortunately the premise is invalid. the attribution of
cause is always a matter of inference relying on theoretical assumption, and while
empirical observation may lend support to our inferences it cannot do more (Pick-
vance, 1982). Moreover in the present case the nature of the potential causes makes
matters more difficult: concepts like technology and level of production refer to
broad background factors which cannot be observed in action. (In contrast one
can get closer to processes of housing allocation: a study of outcomes can in principle
be supported by a study of housing managers in action.) There are thus empirical
problems in identifying and measuring potential causes here, but the key one is of
theoretical assumption. Researchers committed to a universal model of determina-
tion will tend to be satisfied with an explanatory model which conforms to it and
will not look further. Researchers committed to a relativist model will tend only to
be satisfied when patterns of causation specific to a group of societies have been
uncovered.
The debate over east-west urban and regional similaritiesis not therefore resoluble
on the basis of the number of such similarities but involves different assumptions
about the universality of underlying causal patterns. The assumption of the most
different systems research design that similarities arise from a universal causal
pattern is very deeply rooted, and indeed is the basis of the classical experimental
design. But what is valid for natural science is not necessarily so for social science.
The common features shared by two societies may not be the cause of similar
phenomena occurring in them. In the next section I will argue that relativist patterns
of causality are widespread in social science.
The work of Schnore, Pahl, Mullins and Szelenyi demonstrates different approaches
to comparative analysis. Schnore and Pahl use single models to explain different
residential patterns in the US and Latin America, and similarities between east and
west in housing stratification and regional inequalities. Mullins uses linked sub-
models to explain similarities and differences in owner occupation and infrastructure.
176 Comparative urban analysis and assumptions about causality
And Szelenyi uses two models with different rationales to account for east-west
similarities. The argument here is not that any one of the three approaches to
comparative analysis is always appropriate, but that the linked submodels and
diverse models approaches need proper recognition. The reason why they have
lacked this recognition is that the relativist model of causation which they share
threatens deeply-held beliefs about causality which have long been central to social
science methodology, and because of the association of comparative analysis with
convergence theory.
We are faced with a paradox. On the one hand, Przeworski and Teune argue,
rightly I believe, that the relativist model of causation is predominant in social and
cultural processes, and Mullins and Szelenyi make good use of it in their different
approaches to comparative analysis. Yet on the other hand the conventional
approach to social science method generally and to comparative analysis in particular
is firmly based on the universal model.
In this section I shall illustrate the latter point by examining the fate of the
notion of plural causation, and then discuss its implications for comparative urban
analysis.
The essence of the relativist model of causation is that it acknowledges that the
same phenomenon can occur for different reasons or causes in different cases. What
is involved here is a notion of causality which has enjoyed a furtive existence since
its recognition by J.S. Mill in 1843.What he called the plurality of causes has been
referred to by others as plural causation, the term used here.
Mills discussion occurs in Chapter 10 of System of logic. He points out that for
simplicity his discussion of the Canons of Scientific Inquiry in Chapters 8 and 9 had
assumed that single causes were connected with single effects - both in the sense
that given causes had unique and thus distinct effects, and that given effects had
unique and thus distinct causes. But he now states that: [this] supposition does
not hold in either of its parts . . . it is not true that the same phenomenon is always
produced by the same causes; the effect a may sometimes arise from A, sometimes
from B . . . (1906,285). He goes on to point o u t that:
one of the principal consequences of this fact of Plurality of Causes, is, to render the first
of the inductive methods, that of Agreement, uncertain. To illustrate that method we
supposed two instances, A B C followed by a b c , and A D E followed by a d e. From
these appearances it might apparently be concluded that A is an invariable antecedent of
a . . . The moment, however we let in the possibility of a plurality of causes, the conclusion
fails. For it involves a tacit suggestion that a must have been produced in both instances by
fhc same cause. If there can possibly have been two causes, those two may, for example,
be C and E: the one may have been the cause of a in the former of the instances Iin the
above example], the other in the latter, A having no influence in either case (1906, 286.
emphasis added).
This is a quite distinct idea from that olmultiple causation whrre several causes act logether.
but it is always the same causcs and in the same weights. With plural causation, different
causes. o r the s3nic causes with different weights. act to produce a phenomenon in different
societics o r groups 01 societies.
C.G. Pickvunce 177
What are the implications of the prevalence of plural causation for comparative
urban analysis?
The most important is that it enables one t o develop more sophisticated and
richer explanations. Rather than be bound t o the idea that social phenomena can be
brought about in only one way, the diversity of causal processes becomes a legiti-
mate area of inquiry. This was clearly shown in Szelenyis argument on east-west
comparison. If Szelenyi had not implicitly been working with an idea of plural
causation, he would have been confined in a straitjacket: the universal model of
causation would have required him t o search for common causes for common
effects, and hence lead in the direction of a technocratic-organizational theory -
or possibly t o the reductio ad absurdum that Hungary was a capitalist society.
Another example is the differing patterns of causation operating o n British
regional policy at different times. Regional policy originated in the 1930s as a
limited response t o regional concentrations of unemployment. But in the 1960s
and early 1970s under the influence of Labour and Conservative governments
committed to modernization, regional policy was used t o bolster manufacturing
investment in the regions however small its effects in reducing unemployment
(Pickvance, 198 1). The reasons for channelling some investment incentives in this
form - when others were available nationally - was as a response t o the pressures
from local spatial coalitions (Pickvance, 1985b). But in the changes announced in
late 1984 regional policy has been cut in amount and partly subjected t o a cost/job
criterion, thereby reducing the grants given t o projects creating few jobs. I would
argue that this dramatic change reflects a new pattern of causes in which the
priority accorded t o modernization investment has declined and the need t o respond
t o local spatial coalitions has risen so as t o become a major factor sustaining regional
policy - as well as influencing the areas covered by it (Pickvance, 1986). In this
way plural causation helps one t o be aware of the changing causation of a given
phenomenon in one country over time, as well as the different causal patterns
underlying similar phenomena in different countries.
A second important implication of plural causation is that it discourages the
premature abandonment o f promising ideas. For example, if we find evidence that
contradicts our expectations, then working within a universal model of causation,
we are obliged to reject our working hypothesis. However, with a relativist model
(plural causation) we might be able t o imagine ways in which it can be retained. For
example in a recent study of local politics in Dartford and Chatham, Smith (1985)
was unable t o discover any evidence of business influence on local council policy.
He might therefore have jettisoned the idea as relevant in the American but not in
the British context. Instead, by pursuing it he found historical evidence t o support
the thesis that urban politics in the two towns had gone through three phases: a)
business people as philanthropists, b ) business control of council policy (as councils
expanded their functions and taxation powers), and c) council ailtononly of business
but frequently acting in support of present and potential interests). In this way the
initial insight could be retained and the changing patterns of causal influence of dif-
ferent groups in local politics over time and between countries could be explored.
C.3. Pickvilnce 179
These examples show that some past work has implicitly used plural causation -
but without any clear methodological basis. I would suggest that the traditional
caution in urban studies towards making comparisons of very different societies
reflects the same tacit recognition of the problems with methods based on universal
models of causation. Thus Castells argues against east-west comparisons and in
favour of comparisons within the same type of system (1975, 18). Similarly Abu-
Lughod (1 976) distinguishes between legitimate and illegitimate comparisons and
considers only those limited to countries of similar culture and technological level
t o be legitimate. The choice of common features in this method is a matter of
theoretical judgement. (Some idea of the evolution of the field can be seen by com-
paring Sjobergs (1959) and Waltons (1981) overviews of comparative urban
research.) By bringing plural causation out into the light hopefully the basis for
choosing similar cases can be clarified.
A third implication of plural causation concerns convergence theory. Compara-
tive analysis using a single model does not imply a convergence theory. If the two
occur together the connection is accidental. Thus while Schnores analysis did
assume that technology would always advance, Harloes model linking owner
occupation and stable earnings does not require that movement is in one direction
only. Indeed with continuing economic crisis it is evident that the recent growth in
mortgage defaults (fourfold between 1980 and 1984 in Britain) can reverse the so-
called natural upward trend in owner occupation without necessarily challenging
the relationship with stable earnings. The linked submodels and diverse models
approaches to comparative analysis are least likely to be associated with convergence
theory. Szelenyis work points out that there is a logic inherent in Hungarian state
socialism that leads t o housing inequality. For convergence to be a matter of pro-
cess rather than outcome it would have to be shown that this logic was unstable and
was giving way to market rationality (or, less likely, that the market principle was
declining in capitalist societies). Mullins points out how a marxist approach need
not imply convergent evolution. He does not claim that all societies pass through
each of his three stages, but only that certain consequences follow from being in
a given stage. Likewise I have argued that although urban movements are most
numerous and radical in specific societal conditions (absence of democratic politics
or long-term one party rule, rising state social intervention) there is no law that
state social intervention can only rise (as we are seeing at present) or that demo-
cratic politics once established are secure (as witness many Latin American countries)
(Pickvance, 1985a). To sum up, one can say that while convergence theory pre-
supposes a single model of comparative analysis (universal causation), none of the
three approaches to comparative analysis assumes convergence theory. Hence
comparative urban analysis can stay clear of debates over convergence theory if it
chooses. (See in this connection Harloe and Martenss (1984) attempt to rescue
that comparative analyses are simplistic are thus always valid. But that criticism
misses the point since no description can be complete. The real issue is what level of
abstraction/inclusion is necessary for the purpose in hand. In comparative analysis
that purpose is to uncover what is general and the extent of this cannot be known
beforehand. Some writers (e.g. Headey, 1978, Chapter 9) argue that only broad
societal conditions such as technological level, income level and political system
are likely to be the subject of such general propositions. But there is no obvious
reason why the role of agents such as classes as well as structures should not also be
included - as in the comparative analysis of strikes (Korpi and Shalev, 1980).
The main argument of this paper is that comparative analysis should employ all
three of the approaches illustrated in Section 111: single models, linked submodels
and diverse models. The connection between research design and assumptions about
underlying causal patterns needs to be recognized, and in particular the way that
natural science research designs assume the existence of universal patterns. Against
this I have argued that, in social science, relativist patterns of causation are wide-
spread and can only be discovered by the elaboration of linked submodels and
diverse models. It was shown that the notion of cause they assume, plural causation,
is a perfectly legitimate one which deserves recognition after such a long period of
marginalization. But I have also argued that the nature of underlying patterns of
causation can never be resolved on the basis of empirical observation. It is always
partly a matter of inference and hence theoretical assumption. By paying attention
to plural causation the fruitfulness of comparative urban analysis will be maximized.
VI References
Abu-Lughod,J. 1976: The legitimacy of comparisons in comparative urban studies:
a theoretical position and application to North African cities. In Walton, J. and
Masotti, L., editors, The city in comparative perspective, New York. Sage and
Wiley.
Berry, M. 1983: The Australian city in history: critique and renewal. In Sandercock,
L. and Berry, M. Urban political economy: The Australian case, Sydney:
George Allen and Unwin.
1984: Urbanisation and accumulation. In Williams, P., editor, Conflict and dev-
elopment, Sydney: George Allen and Unwin.
Castells, M. 1975: Contribution to discussion. In Harloe, M., editor, Proceedings of
the Conference on Urban Change and Conflict 1975, London: Centre for En-
vironmental Studies.
1977: The urban question. London: Edward Arnold.
Cohm, M.R. and Nagel, E. 1934: An introduction to logic and scientific method.
London: Routledge and Kegan Paul.
Cole, R.E. 1973: Functional alternatives and economic development. American
Sociological Review 38,424-38.
Dear, M. and Scott, A.J. editors, 198 1 : Urbanization and urban planning in capitalist
society. London and New York: Methuen.
182 Comparative urban analysis and assumptions about causality
Duncan, S.S. 198 1 : Housing policy, the methodology of levels and urban research;
the case of Castells. International Journal o f Urban and Regional Research
5 , 2 3 1-54.
Gilbert, A. editor, 1982: Urbanization in Latin America. Chichester: John Wiley.
Goode, W. J. and Hatt, P.K. 1952: Methods in social research. New York: McCraw
Hall.
Cordon, D. 1977: Capitalism and the roots of urban crisis. In Alcaly, R.E. and
Mermelstein, D., editors. The fiscal crisis o f American cities, New York:
Random House.
Harloe, M. 1979: Marxism, the state and the urban question: critical notes o n two
recent French theories. In Crouch, C., editor, State and economy in contem-
porary capitalism, London: Croom Helm.
1981a: The recommodification of housing. In Harloe, M. and Lebas, E., editors,
City, class and capital, London: Edward Arnold.
1981b: Notes on comparative urban research. In Dear, M. and Scott, A.J.,
editors, I98 1.
Harloe, M. and Martens, M. 1984: Comparative housing research. Journal of Social
Policy 13, 255-77.
Headey, B. 1918: Housing policy in the developed economy. London: Croom Helm.
Hegedus, J. and Tosics, I. 1983: Housing classes and housing policies: some changes
in the Budapest housing market. International Journal o f Urban and Regional
Research 7,467-94.
Heidenheimer, A., Heclo, H. and A d a m , C.T. 1983: Comparative public policy.
London: Macmillan.
Kemeny, J. 1983: Economism in the new urban sociology: a critique of Mullinss
Theoretical perspectives o n Australian urbanization. Australian and New Zeal-
and Journal of Sociology 19, 5 11-27.
Kobben A.J.F. 1968: The logic of cross-cultural analysis: why exceptions. In
Rokkan, S., editor, Comparative research across cultures and nations. Paris and
the Hague: Mouton.
Korpi, W. and Shalev, M. 1980: Strikes, power and politics in the western nations,
1900-1 976. Political Power and Social Theory 1, 301-304.
Mill, J.S. 1906: System of logic, eighth edition. (First edition 1843.) London:
Loiigman, Green.
Mullins, P. 198 1: Theoretical perspectives o n Australian urbanization. I: Material
components in the reproduction of Australian labour power. Australian and
New Zealand Journal of Sociology 17, 65-76.
1983: Theory and Australian urbanization: a comment on Kemenys remarks.
Australian and New Zealand Journal of Sociology 19, 528-33.
Murray, P. and Szelenyi, I. 1984: The city in the transition t o socialism.lnternationa1
Journal o f Urban and Regional Research 8 , 90-107.
Nagel, E. 1965: Types of causal explanation in science. In Lerner, D., editor, Cause
and effect, New York: Free Press.
Newton, K. 1974: Community decision-makers and community decision-making in
England and the United States. In Clark, T.N., editor, Comparative community
politics, New York: John Wiley.
Pahl, R.E. 1977a: Stratification, the relation between states and urban and regional
development. International Journal o f Urban and Regional Research 1, 6- 18.
C.G. Pickvance 183
1977b: Managers, technical experts and the state: forms of mediation, mani-
pulation and dominance in urban and regional development. In Harloe, M.,
editor, Captive cities, London: John Wiley.
1 9 7 7 ~ :Collective consumption and the state in capitalist and state socialist
societies. In Scase, R., editor, Industrial society: class, cleavage and control,
London: George Allen and Unwin.
Pickvance, C.G. 1980: The role of housing in the reproduction of labour power and
the analysis of state intervention in housing. CSE Political Economy of Housing
Workshop, Housing, Construction and the State, London: Conference of
Socialist Economists.
1981: Policies as chameleons: an interpretation of regional policy and office
policy in Britain. In Dear, M. and Scott, A.J., editors, 1981.
1982: The state and collective consumption (Course D202, Unit 24). Milton
Keynes: Open University Press.
1985a: The rise and fall of urban movements and the role of comparative analysis.
Society andspace 3,31-53.
1985b: Spatial policy as territorial politics. In Rees, G., editor, Political action
and social identity: class, locality and culture, London: Macmillan.
1986: Regional policy as social policy: a new direction in British regional policy.
In Brenton, M. and Ungerson, C., editors, The yearbook of social policy 198516,
London: Routledge and Kegan Paul.
Przeworski, A and Teune, H. 1 970: The logic o f comparative social inquiry. London:
John Wiley.
Roberts, B.R. 1978a: Agrarian organization and urban development. In Wirth, J.D.
and Jones, R.L., editors, Manchester and Sao Paulo, Stanford: Stanford Univ-
ersity Press.
1978b: Comparative perspectives on urbanization. In Street, D. et al., editors,
Handbook o f contemporary urban life, San Francisco: Jossey-Bass.
198 1: Migration and industrialising economies: a comparative perspective. In
Harloe, M., editor, New perspectives in urban change and conflict, London:
Heinemann.
Schnore, L.F. 1965: On the spatial structure of cities in the two Americas. In
Hauser, P. and Schnore, L.F., editors, The study of urbanization, New York:
John Wiley.
Sjoberg, G. 1959: Comparative urban sociology. In Merton, R.K., Broom, L. and
Cottrell, L.S., editors, Sociology today, New York: Basic Books.
Skocpol, T. and Somers, M. 1980: The uses of comparative history in macrosocial
inquiry. Contemporary Studies in Society and History 22, 174-97.
Smelser, N.J. 1976: Comparative methods in the social sciences. Englewood Cliffs,
Prentice-Hall.
Smith. S.L. 1985: A political economy of urbanization and state structure. Univer-
sity of Kent, Canterbury, unpublished PhD thesis.
Szelenyi, I. 1972: Housing system and social structure. Sociological Review Mono-
graph 17, 269-97.
1978: Social inequalities under state socialist redistributive economies, Znter-
national Journal of Comparative Sociology 19,63-87.
198 1 : The relative autonomy of the state or state mode of production. In Dear,
M. and Scott, A.J., editors, 1981.
184 Comparative urban analysis and assumptionsabout causality
1983 : Urban inequalities under state socialism. Oxford: Oxford University Press.
Walton, J. 1973: Standardized case comparison: observations on method in com-
parative sociology. In Armer, M. and Grimshaw, A.D., editors, Comparative
social research, New York: J o h n Wiley.
1977: Elites and economic development. Austin: University of Texas Press.
198 1: Comparative urban studies. International Journal of Comparative Sociology
22,22-39.
Willer, D.E. 1967: Scientific sociology. Englewood Cliffs: Prentice-Hall.
Wynn,M. editor, 1983: Housing in Europe. London: Croom Helm.
Yujnovsky, 0. 1975: Urban spatialstructure in Latin America. In Hardoy, J., editor,
Urbanization in Latin America, New York: Anchor.
Zelditch, M. 197 1 : Intelligible comparisons. In Vallier, I., editor, Comparative
methods in sociology, Berkeley ; University of California Press.