The Journal of Socio-Economics 38 (2009) 843–858

Contents lists available at ScienceDirect

The Journal of Socio-Economics

journal homepage: www.elsevier.com/locate/soceco

A simple theory of ‘meso’. On the co-evolution of institutions and

platform size—With an application to varieties of capitalism and
‘medium-sized’ countries
Wolfram Elsner ∗ , Torsten Heinrich
University of Bremen, iino - Institute of Institutional and Innovation Economics, Faculty of Economics and Business Studies,
Hochschulring 4, D-28359 Bremen, Germany1

a r t i c l e i n f o a b s t r a c t

Article history: This paper deals with institutional emergence in the well-known ‘evolution of cooperation’ framework
Received 4 April 2008 and focuses on its size dimension. It is argued that some ‘meso’ (rather than ‘macro’) level (to be numeri-
Received in revised form 9 April 2009 cally determined) is the proper level of cultural emergence, diffusion, and retention. Also Schumpeterian
Accepted 3 May 2009
economists (K. Dopfer et al.) have discussed institutions as ‘meso’ phenomena, and Schelling, Axelrod,
Arthur, Lindgren, and many others have dealt with ‘critical masses’ of coordinated agents, including
JEL classification:
related segregations of populations. However, the process and logic of emergent group size has rarely been
B52
explicitly explored so far. In this paper, ‘meso’ will be explained, in an evolutionary and game–theoretic
D02
D85
frame and a population perspective, in terms of a cooperating group smaller than the whole population
involved. Mechanisms such as memory, monitoring, reputation chains, and active partner selection will
Keywords: loosen the total connectivity of the deterministic ‘single-shot’ benchmark and thus allow for emergent
Evolution of cooperation ‘meso’-sized arenas, while expectations to meet a cooperative partner next round remain sufficiently
Emergence of institutions high. Applications of ‘meso-nomia’ include the deep structure of ‘general trust’ and the surprisingly high
Group size
macro-economic and macro-social performance in ‘small’ and ‘well-networked’ countries which helps to
Network size
explain persistent ‘varieties of capitalism’. A strategy for empirical application of the theoretical approach
Expectations
Social capital and some first empirical indications of its relevance are presented.
Trust © 2009 Elsevier Inc. All rights reserved.
‘Meso’-economics
Varieties of capitalism
Small countries

1. Introduction require the simultaneous constitution and size determination of the

This paper deals with the well-known issue of the emergence
of a social institution of cooperation. However, emergence here
will constitute some ‘meso’-sized group of institutionally coordi-
nated agents. Using the standard game–theoretic argument of the
‘evolution of cooperation’ in a Prisoners’ Dilemma (PD), the ‘folk the-
orem’, the dilemma can in principle be overcome by the emergence
of coordination through cooperation which in turn is achieved by
an (learned, informal) institution (see, e.g., Schelling, 1973, 1978;
Schotter, 1981; Axelrod, 1984/2006).2 However, this will basically
∗ Corresponding author. Tel.: +49 421 218 7535/6; fax: +49 421 218 4974.
E-mail address: welsner@uni-bremen.de (W. Elsner).
1
http://www.iino.de.
2
We refer to ‘coordination’ in a wide sense as an umbrella for ‘coordination’ in
a narrow sense (n.s.) and ‘cooperation’. Particularly, a ‘coordination problem’ (in a
n.s.) is solved by successful ‘coordination’ (n.s.) gained through some social ‘rule’. In
contrast, a ‘dilemma problem’ can only be solved by ‘cooperation’ (a coordination in
institution’s carrier group. This paper investigates the simple logic of
that group’s co-evolution and the determination of its size dimen-
sion, the numerical value of which, however, can be determined
only in specific models and simulations.
Critical here will be the (subjectively perceived) importance of
the common future, i.e. expectations either to meet the same inter-
action partner again next round or to meet any agent of a particular
type (cooperator or defector) next round which will determine the
individual’s decision on his behavior and in this way determine the
size of the emerging carrier group of the institution.
In a n-person setting, the fully deterministic ‘single-shot’ logic of
the ‘folk theorem’ implies a (weak) total connectivity among agents
where everyone may interact with everyone else with the same
a n.s. plus the habitual sacrifice of the short run extra pay-off), gained only through
an institution (i.e., a social rule plus endogenous sanction). For this use of terms, see,
e.g., Schotter (1981).

1053-5357/$ – see front matter © 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.socec.2009.05.001
844 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
probability p in every round. While ‘strong’ total connectivity is
defined here as p = 1.0, i.e. everyone definitely interacts with every-
one else in every round, ‘weak’ total connectivity is given if p < 1.0.
We will show that in order to explain a ‘meso’ carrier group we will
have to ‘loosen’ connectivity and in this way get beyond (weak) total
connectivity.
Particularly, in a population perspective, the individuals need to
experience the population shares of the types of strategies [be it
the shares of types of agents or of types of actions (when agents
may behave this or that way, depending on the constellations of
critical factors which are to be discussed)]. Expectations then will
have to be considered ‘contingent trust’, i.e. the expectation to meet
a cooperator (or a cooperating action) next round.
‘Agency’ capabilities and mechanisms such as memory, moni-
toring, reputation building and using reputation chains, as well as
active partner selection based on respectively gained knowledge
then ‘loosen’ connectivity, i.e. dissolving (weak) total connectiv-
ity. With random encounters given every round, agents will be
assumed to be able to form some ‘contingent trust’, based on some
empirically gained knowledge about some agents, and, within lim-
its, to reject the next random interaction partner. In this way,
agents will be able (within limits) to generate and ‘assort’ their
individual relevant groups of interaction partners, i.e. their peer
groups.
In this way, these mechanisms will allow for relevant group sizes
to increase from ‘very small’ (as with total connectivity) into ‘meso’
sizes while keeping expectations ‘to meet’ sufficiently high. Expec-
tation then still can be supportive to institutional emergence even
in a growing group, i.e. make cooperation still appear to be the
superior strategy for the individual even if group size increases.
However, if these agency mechanisms for some reason loose
power and efficacy or reach some limits, and cooperative payoffs
will approach some maximum (for instance reputation chains may
generate decreasing information gains) with growing numbers of
cooperators, while defectors increasingly profit from the growing
numbers of cooperators, the co-evolution of institutional emer-
gence and group size will yield some group smaller than the whole
population involved, in this way disconnecting the ‘whole population’
and the ‘maximum relevant group’, thus qualifying informal insti-
tutionalized coordination (i.e. cooperation) as a ‘meso’-economic
phenomenon.
However, some initial ‘minimum critical mass’ of cooperators
has to come into being stochastically through some motivation to
diversify behavior as we start from common defection. This may
first lead to the ‘takeover’ of the whole population by cooperators
but, if the relative success of common cooperation (as supported
by the agency mechanisms above) should become exhausted with
the growing share of cooperators as mentioned (a sigmiod curve or
a curve with constantly decreasing growth of cooperation payoffs),
the process will yield that ‘relevant cooperating group’ smaller than
the whole population (a ‘maximum critical mass’).
2. Size and ‘meso’ size in the literature so far
Some evolutionary economists have elaborated in recent years,
in a Neo-Schumpeterian perspective, on the ontology of ‘meso’ in
terms of ‘meso rules’ and the process of their generation, adoption,
diffusion and retention (see, e.g., Dopfer et al., 2004; for the respec-
tive groundwork, see Dopfer, 2001, 2005, 2007; Dopfer and Potts,
2008). They have argued that, and described how, origination, adop-
tion, diffusion and retention of a rule take place in a ‘meso’-sized
group of carriers with a ‘meso’-sized population of actualizations of
an ideal generic rule. However, they have not elaborated specific
causal mechanisms by which ‘meso’ comes into existence to solve
specific problems.
In addition to that approach, the present paper seeks to estab-
lish that the emergence, i.e. generation, adoption and diffusion of
an institution, can be traced back to a specific but general prob-
lem which agents have continuously to solve both individually and
collectively through that very process of institutional emergence.
Gibbons has recently advocated ‘to bring interests back into our
thinking about (. . .) routine production’ (Gibbons, 2006, p. 381;
italics added), referring to the ‘folk theorem’: ‘building an equilib-
rium means that interests creep in; one cannot analyze just the
evolution of beliefs’ (p. 385). In fact, the game–theoretic approach
is about a complex interest structure to be solved through mutual
adaptations of behaviors and expectations becoming consistent.
In the present paper, thus, we will explore a simple logic of
the co-evolution of (1) a complex incentive structure, (2) ‘experi-
enced’ expectations (‘to meet . . .’), indicative, in turn, and in varying
degrees, of (3) the group size, and of (4) the institution as such (as
both quest and outcome of the individuals’ efforts to improve their
well-being).
This might contribute to a general ‘meso’-economics wherein
‘meso’ groups, ‘meso arenas’, or ‘platforms’ in manifold socio-
economic areas (regional, industrial, or professional clusters,
networks, agglomerations, segregation and neighborhood struc-
tures, etc.) may become the theoretical locus of emergent structure.
Coordinated (and cooperative) systems of such ‘relevant’ sizes
may have a specific capability of learning, innovative collective
action, and thus eventually high macro performance, under different
parameter configurations (for the large field application of different
game structures in small-scale societies with different configura-
tions, including group size, see Henrich et al. (2004).3
However, a heroic implicit presumption of most game–theoretic
arguments is complete information. Agents are assumed to have a
direct observable connection between actions and outcomes and
thus intense incentives or pressure to learn. This transparency is rarely
the case in reality where the direct connection of action to feed-
back and thus the pressure to learn typically is much weaker. Real
societies, even ‘primitive’ and small-scale ones, display a surpris-
ing variety of degrees of learned and institutionalized cooperation
and reciprocity (see again Henrich et al., 2004) since mankind
is much more detached and freed from evolutionary pressure to
adapt and learn than the game–theoretic argument of complete
transparency presupposes. Thus, empirically, even ‘meso’-sized and
‘primitive’ groups sometimes may have low degrees of cooperation.
They can afford certain levels of non-cooperation and conflict. Typ-
ically, the ‘backup capacity’ of humans to improve their position
with low levels of cooperation is exploiting the commons of nature
and this does not immediately and transparently feedback to the
agent.
We will apply below the theoretical considerations of this
paper to ‘trust’ and macro-performance of some ‘small’ highly
industrialized countries being ‘well-structured’ into ‘meso’-sized
arenas. We will discuss a potential empirical research strategy
when ‘meso’-sized structures are already historically given and
their co-evolution with institutional emergence just means their
further reproduction, generating high levels of trust, cooperation,
social capital and, finally, high macro-performance. If ‘meso’ size
is relevant it could be used even for political design to generate
institutional emergence and high macro-performance.
3
Group size is there but one critical factor among others and interferes with
other factors to form different interaction conditions and trigger different resulting
degrees of institutionalized cooperation, although these real societies explored are
all ‘small-scale’ (ranging in size between 75 and 1219). This, however, does not imply
that relative smallness of interaction arenas, or ‘meso’ group size, as such would not
tend to be a favorable condition of institutional emergence. In fact, size was found in
that large cross-cultural field experiment to be a good predictor (similar and related
to anonymity and complexity) for payoff to cooperation.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
Also, institutionally coordinated systems may possibly also
become ‘petrified’, ‘sclerotic’ or ‘locked-in’, in the course of their life
cycles if collective action cannot be renewed to lock the system out
when necessary (for a formal model of merely technological ‘lock-
in’, see, e.g., Arthur, 1989; the classic for institutional lock-in, David,
1985; and for lock-in and -out recently, Dolfsma and Leydesdorff,
2009).
Many have paved the way for exploring critical size. For instance,
there has been a renewed interest in Schelling’s (1969, 1973, 1978)
early investigations in the emergence of coordination in attendance
problems and in emergent spatial segregation (see, e.g., Vinkovic
and Kirman, 2006; Aydinonat, 2007; see also already Elster, 1989).
Also, Axelrod’s 1980s approach to a quasi-evolutionary simulation
of emergent cooperation and evolutionarily stable critical mass (incl.
segregation) is still widely discussed (e.g., Axelrod, 1984, see the
statistics in Dawkins’ foreword to the 2nd ed., 2006), and the iter-
ated PD is still much applied and elaborated on (e.g., Knudsen,
2002; Devezas and Corredine, 2002; Eckert et al., 2005; Goyal,
2005; Traulsen and Nowak, 2006). Arthur’s ‘El Farol’ attendance
coordination problem (Arthur, 1994) also has triggered research
on coordination success and failure which implied a size issue.
Game theorists have found some confirmation of the relevance of
(‘meso’) group size also in lab experiments (see, e.g., Yamagishi,
1992). Generally, group (or ‘network’) formation processes have been
a continuing issue (e.g., Demange and Wooders, 2005; Page and
Wooders, 2007). The evolutionary dynamics in a PD, when control-
ling for a broad range of initial conditions and allowing for a variety
and an ongoing generation of ever more complex strategies, has
been developed far beyond standard PD-supergame equilibria or
well-defined attractor solutions such as a clear-cut Axelrodian TFT
dominance (e.g., Lindgren and Nordahl, 1994; Binmore, 1998; Foley,
1998; Watts, 1999).
Institutionally oriented game-theorists, such as Schotter, Field,
Hargreaves Heap or Ostrom, have built bridges between game
theory and evolutionary-institutional theorizing, and the size
dimension has often played some role (for an overview of the
issues, see, e.g., Dosi and Winter, 2000; Ostrom, 2007.) Also, many
evolutionary-institutionalist economists have elaborated on insti-
tutional emergence and group or ‘network’ conceptions of the
individual (e.g., Hodgson, 2000; Davis, 2007, 2008). And some par-
ticularly have contended that institutions are ‘meso’, would emerge
at some ‘intermediate’ level, being effective in ‘mid-sized’ groups,
etc. (e.g., van Staveren, 2001, 179f; Elsner, 2000).
Group size has also been an obvious issue of the collective good
problem since Olson’s (1965) Logic of Collective Action where the col-
lective good has a better chance of being produced the smaller the
relevant group which is constituted to generate the good (see, e.g.,
Dejean et al., 2008). Also, some have investigated ‘critical masses’
in collective action along Olsonian lines considering large contrib-
utors to the collective good. The latter can either produce the good
alone or mobilize a selected minimum producer group (see, e.g.,
Marwell and Oliver, 1993). This, however, is not done in an evolu-
tionary emergence perspective, thus what we pursue in this paper.
If game–theoretical arguments would be applied this would prob-
ably be a case for cooperative game theory.4
However, in all, the question as to why and how institutions
emerge and become effective at some proper ‘meso’ level has rarely
been explicitly delved into and the basic logic of the size dimension,
and particularly ‘meso’ emergence has rarely been focused on. This
is what this paper seeks to do.
Dosi and Winter (2000) have attempted an overview on evo-
lutionary theorizing, games, and emergent processes and have
4
We owe the idea of considering cooperative game–theoretical arguments in
different respects throughout this paper an anonymous referee.
845
concluded that there is no complex modeling without a proper
qualitative evolutionary ‘process story’ (see also Gruene-Yanoff and
Schweinzer, 2008). We will ‘embed’ a simple logic in such a frame
below. We confine ourselves to some simple formalism and graph-
ical illustration along Schelling’s and Axelrod’s lines.
It has often been suggested that the ‘macro’-level, convention-
ally understood here as the national level of formal organization
and public agency, has become less relevant in a (global) cul-
tural emergence perspective. However, it is disputable whether it
still is more appropriate to consider informal institutional emer-
gence still under a ‘micro-to-macro’ perspective and terminology
rather than to conceptualize ‘meso’ as a socio-economic level of its
own (for a ‘micro-to-macro’ terminology, see, e.g., Hodgson, 2000;
Ayres and Martinás, 2005). We will argue here that there are con-
siderable theoretical and empirical reasons to envisage a specific
level of informal cultural emergence below, and perhaps across,
conventional ‘macro’ jurisdictions (which typically are the loci of
enculturation through more formal and often ceremonial institu-
tions). And since emergent structure is not reducible to its initial
micro-components (agents of different kinds) it is of course above
the ‘micro’ level. In all, what complex evolutionary-institutional
theorizing, modeling or simulation, and real-world clusters, net-
works and all kinds of group cultures are all about may require a
theoretical space of its own—‘meso’ (see, e.g., Chen, 2008, p. 121).
Thus, we define micro as the level of individual agents and their
interactions. As soon as some ‘structure’ (institution) has emerged
that exists independently from any individual agent’s action, we
understand this to belong to the meso level if also a ‘meso’-sized
relevant interaction arena or ‘carrier group’ has been co-evolving
with that structure or can be identified. A meso-sized group, in turn,
is defined as any ‘relevant’ (i.e. actually emerged as cooperating)
group of a size equal to or smaller than the larger whole population
involved. Finally, if the ‘relevant’ group can be shown to be smaller
than the whole population the latter may be considered to belong
to the macro level, mirroring perhaps the real-world ‘national’ level
as mentioned.
Note that we will consider a population later that is ‘structured’
only between (TFT-) ‘cooperators’ and ‘defectors’ (where the par-
ticular division structure will change with ‘emergent structure’),
with no opportunity to leave this arena and population and migrate
among several populations, i.e., no population of populations,
which may be an important framework for other game–theoretic
settings (e.g., Knudsen, 2002; Traulsen and Nowak, 2006).
In the following, we will first present our argument in the frame
of a simple static and deterministic ‘single-shot’ logic. Then we
will consider group size in a co-evolutionary process of institutional
emergence and group constitution. In the context of ‘embedding’
this logic in a ‘process story’, we will adopt a population perspec-
tive based on some motivational considerations where the portions
of cooperating and defecting agents in the population become a
stochastic variable at the outset. The evolutionary stable ‘mini-
mum critical mass’ and the (maximum) ‘relevant cooperating group’
are identified, the latter immediately being the whole population
at first. Memory, monitoring, reputation chains and (some) active
partner selection then allow for a smaller minimum critical mass
through faster increasing cooperative payoffs. They also allow for
the ‘relevant’ group size to increase above a ‘very small’ group
through loosening connectivity, i.e. ‘contingent trust’ can be kept
high then while the ‘relevant group’ may increase, through better
assorted individual peer groups. However, when those mecha-
nisms wear out with increasing portions of cooperators the relevant
group size may become a ‘maximum critical mass’ smaller than the
whole population. Finally, we will consider some recent applica-
tion, namely persistent national differences in ‘general trust’ and
corresponding macro-performance, an interesting dimension of
the topical ‘varieties of capitalism’ issue. A strategy for empirical
846 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
application of the theoretical approach and some first empirical
indications of its relevance will be discussed.
3. The ubiquity of the dilemma problem and ‘emergent
structure’
The prominent relevance of the collective-good/social dilemma/
PD problem requires a most careful explication. Its theoretical
prominence in fact stems from its ubiquity as an (actual or latent)
practical everyday problem. There is in fact a collective-good prob-
lem involved in every single economic decision, even in the most
simple supermarket purchase, but also in any more demanding
technological coordination problem in the fragmented value-added
chain. If a full-fledged institution already exists with zero defection,
then typically any agent actively contributes to the reproduction of
the institution, and of the corresponding expectations of others,
through cooperative behavior. However, if this is not the case and
if an agent may expect another agent to behave in a cooperative
way next interaction there is, under certain conditions a dominant
incentive for him not to contribute. These conditions include, besides
the payoff structure, the expectations ‘to meet’, memory, monitoring,
reputation, related sanctioning and the danger of being rejected and
excluded from many interactions. By not contributing he may take
the opportunity of a potential short-run one-shot extra gain, for
instance, by running away without paying, by somehow cheating,
avoiding own costs, exploiting some commons, exploiting positive
externalities from others, etc. For instance, under certain circum-
stances in the fragmented value-added chain, the incentive to free
ride, by saving R&D expenses and profit from incoming knowledge
spillovers which are ‘inappropriable’ by their creators, may become
virulent. Similarly, even in a purely technological Arthurian random
net-technology choice problem, agents may be dominantly incited
to free ride by waiting until others have made their decisions, in
this way avoiding later regret (if they can afford to wait).
Generally, agents in a more or less individualistic culture may
be incited to defect in manifold ways, and will do so insofar as
the situation is not fully governed by institutions (not considering
formal hierarchical control). So any socio-economic (trans-) action
is embedded in a larger dilemma problem and will, or will not,
contribute to the production or reproduction of the general frame
of expectations which in turn allow for, or undermine, institutional
emergence to overcome that basic dilemma.
Along these lines, and according to a large literature of applied
game–theoretic argument, we have argued elsewhere that any
production, information and innovation system, under conditions
of fragmented value-added chains (in face of complex integrated
products), of (competing) net-technologies, and the collective-good
character of basic information, can be modeled as a system of
mutual externalities, collectivities and cumulative actions, such
that it can be reconstructed as a PD where any transaction is embed-
ded in (Elsner, 2005).
However, the PD structure often exists only ‘in the background’,
while the observable social surface is dominated by some solutions,
i.e. some institutionalized arrangements. These may be ‘instrumen-
tal’ (i.e. proper for ongoing problem-solving, which is the subject of
this paper) or ‘locked-in’ on an inferior technology (being also an
inferior institution), or even completely mutually blocked through
general free-riding and non-action.
In a life-cycle story of institutions, with emerging lock-in of coor-
dination and with improper collective action capability, improper
institutionalized coordination or complete mutual blockage of
action may easily be traced back to power and status positions
established in the course of the repeated application of an institu-
tion beyond the point of its due change (to yield further economies
of scale from the institution). This would be the Veblenian ‘ceremo-
nial’ dimension of the institution (e.g., Bush, 1987). A ‘ceremonially
dominated’ institution would appear in reality as an ‘abstract norm’.
The existence of proper and improper ‘solutions’ may not even
be realized as such by the agents who perhaps do not know bet-
ter. The ‘surface’ of institutionalized everyday ‘solutions’ (including
institutional lock-in and ‘non-action’) typically is more easily visi-
ble than the complex problem structure in the background and its
alternative and perhaps more adequate potentialities.
Given ubiquitous dilemma problems, individualist decision-
making may lead to inferior results, and ideal markets, or
close-to-ideal de-regulated real markets, and the prices resulting
in them may fail to generate and diffuse the shared knowledge and
expectations required for some ‘reasonable’ behaviors and outcome.
A solution (superior to mutual individualist blockage or lock-in)
then may require a ‘higher’ form of rationality than a dis-embedded
market can provide by itself.
That ‘higher’ form of rationality allowing for shared knowl-
edge and informal coordination cannot be comprehended other
than an institutionalization of cooperation through a learned and
habituated social institution, in the face of a dominant incentive
‘rationally’ to defect. This is because coordination has to assume
the specific form of cooperation, i.e. coordination plus sacrifice.
Rational behavior in a short-run (‘one-shot’) perspective will be
unable to overcome the dilemma. In a dilemma-prone, decentral-
ized system, the dilemma problem can be overcome, if not through
formal and authoritarian mechanisms, only by ‘habitualized’ and
‘semi-conscious’ behavior. Consequently, any proper (‘instrumen-
tal’) solution requires recurrent interaction, in historical time, to
open up room for joint learning, long-run and mutually adapting
expectations, and path-dependent cumulative behavior to solve the
dilemma (or not).
There is a fallacy of aggregation involved in dilemma-prone
direct interdependence, reflecting the fact that decisions which
may appear, at first glance, ‘right’ and ‘optimal’ from an isolated
individual(ist) point of view may unintentionally turn out to be
collectively, and often even individually, ‘wrong’ and inferior (‘unin-
tended consequences’).5 Since the ‘aggregate’ here results from a
more complex decision structure and process it is ‘non-additive’.
Evolutionary-institutional economics conceives the outcome of
a complex system and process as ‘emergent structure’, specifically
as institutionalization (e.g., Hodgson, 2000, 113ff.). Emergence of
structure implies an ‘entity (that) has properties which cannot
be deduced from prior knowledge of the elements’ (113f.). The
micro–macro levels’ interdependence contributes to the ‘micro-
foundation of macro’ and to a ‘reconstitutive downward causation’
of individual behavior through institutions, i.e. a ‘macro-foundation
of micro’ (see also Foster, 2005, on complex systems).
This is, of course, not to suggest that a formal mechanism, mod-
eling, and simulation of emergent process can generate ‘something’
from ‘nothing’. For instance, some ‘common culture’, ‘cultural’
capabilities, and social ‘agency’ competence (such as searching
and learning, risk-taking, being not perfectly envious, monitoring,
memorizing, and identifying others, expecting and anticipating,
reputation building, and partner selection—see below) may have
to be assumed at the outset. Nevertheless, the complex process
is open-ended and path-dependent, i.e. its specific collective, or
‘social’, outcome at some point of time is not determined by the
individuals and may sometimes even be surprising. In this way, the
emergent process may increase our understanding by directing our
attention back to crucial properties of the initial structure, of the
5
The current financial meltdown and different real-economic crises (business
cycle, food, resources, climate) have provided ample evidence of processes of
unintended consequences because of improper incentives, false expectations and
improper coordination, e.g., ‘herd behavior’.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
process and of the micro units that may have been presupposed
only tacitly so far.
Agents will start in our as the same short-run maximizers which
may be considered a ‘worst-case’ condition. However, the solution
of the problematic decision setting requires a change of behavior.
If those initial ‘worst-case’ maximizers can change their behaviors
and can create cooperation through establishing a social institution
and a social carrier group something seems to be gained.
This is prototypical of non-cooperative game theory compared
to other approaches such as cooperative game theory or exper-
imental behavioral game theory that require or bring in more
assumptions such as preexisting norms (e.g., ‘inequality aversion’),
perceptional framings, and other sorts of ‘conditional cooperation’
(e.g., caring about others’ intentions).6
4. Population size in a ‘single-shot’ perspective
4.1. A formal sketch
The simplest formal illustration of the static ‘single-shot’ solu-
tion provides the logical condition for the superiority of cooperation
over defection. Assume
a, a d, b
b, d c, c
with b > a > c > d, and a > (d + b)/2. As is well-known the payoffs P in
a supergame for the TFT player always encountering another TFT
player, and for an ALL D player encountering a TFT player, with ı
being the common discount factor, are
PTFT/TFT = a + ıa + ı2 a + · · ·
a (n − 1)
=
1−ı
and
PALL/TFT = b + ıc + ı2 c + · · ·
c A discount factor for a series of rounds, including the probability
= + b − c,
1−ı
respectively. Cooperation then pays if
(b − a)
PTFT/TFT > PALL D/TFT , → ı >! , (1)
(b − c)
as popularized for instance by Axelrod (1984/2006).
According to inequality (1), cooperation (as an institution) may
become logically possible in a single-shot decision, as an equi-
librium of a PD supergame different from the ‘one-shot’ Nash
equilibrium. It is the perspective of an individual agent to decide
before the very beginning of interactions which of the two strategies
possible to choose.
This, however, does not tell much by itself about the individual
motivations to act and the process of emergence of a cooperative
equilibrium. Therefore, we will consider some prerequisites and
implications of this logical condition of institutional emergence
often presumed only tacitly.
The critical factors are the given quantitative dilemma-prone
incentive structure, i.e. the quantitative strength or weakness of the
collective-good problem, a, b and c, relative to the common discount
factor, also interpreted in a supergame as the ‘probability to meet the
same interaction partner again next interaction’, i.e. the importance
of the common future (ı). Cooperation will be the more superior the
smaller the ‘opportunity costs of common cooperation’ (b − a) in
6
An overview of research on pre-existing norms that real-world agents, i.e. the
participants in laboratory experiments, carry into the experiments, in the frame
of applied experimental game theory and behavioral economics, is given, e.g., by
Camerer and Fehr in Henrich et al., 2004, Chapter 3, 55ff.
847
relation to the ‘opportunity costs of common defection’ (b − c), and
the larger the importance of the future.
At first, note that ı strictly applies within a supergame only. The
supergame has an indefinite length, i.e. an indefinite number of inter-
actions, and its end is surprising for the agents so that the infinite
calculus above applies.
However, this implies that within the supergame the two inter-
action partners will not change. Thus the classical discount factor
applies where p1 is the probability that the next interaction takes
place with the same agents and r is the usual time preference (‘inter-
est rate’). Let the discount factor applying within a supergame be
ı1 :
p1
ı1 = . (2)
(1 + r)
In order to introduce partner change, the probability to meet the
same again next round and, with this, population size we assume
a ‘structured’ supergame where, after an indefinite number of inter-
actions, a round will (randomly) end and a new round will for each
individual begin with random partner change (week total connec-
tivity). That is we suppose a uniform periodization over time both
within and between rounds. With uniform periodization, the first
interaction of the next round takes place in the very next time unit
so that a TFT player who has a memory length of one time unit (one
interaction) will remember if he meets the same interaction partner
again in the new round.
This is where group size comes in. The agent will meet the same
interaction partner again next round in a population of size n, with
probability
1
p2 = . (3)
Obviously, the probability to meet the same agent again next
round is rather small and decreases with an increasing population
size.7
‘to meet the same again next round’ after random partner change,
then can be defined as ı2 8 :
 p + (1 − p )p  p1 (1 − p1 )
1 1 2
ı2 = = + . (4)
(1 + r) (1 + r) [(1 + r)(n − 1)]
Again, p1 is the probability to meet the same again within a
round, i.e. that interactions within the round will go on (with the
same), (1 − p1 ) thus the probability that this will not be the case,
7
A note on memory and its implications may be in order here. As mentioned,
the cooperative (TFT) agent is supposed to have a memory length of one time unit
(t = 1) which may considered to be equivalent to a required subjective ‘one-time-
unit probability to meet again’ perceived by agents. However, if cooperative agents
were assumed to remember longer time spans, e.g., a memory of t = T, the maximum
population size allowed for the institutionalization of cooperation c.p. will increase,
since the probability of meeting a certain agent again in any of the future rounds
within their memory period T would considerably increase, given the population
size. Again, by contrast, population size would increase with increasing memory
periods while the extended ‘probability to meet again’, or ‘cumulated probability’,
may remain constant. From n = 1/p2, t = 1 +1 (see Eq. (3)), for instance, would follow
the max. population size n1/[1 − (1 − p2, t = T )1/T ] + 1. Obviously, n increases with T.
For example, for b = 4, a = 3, c = 2, p2, t = 1 = 0.5, according to the single shot inequality
(1) the related maximum population size for t = 1 would be 3. An increase from
t = 1 to t = 2 then increases n from 3 to about n = 4.4, for t = 3 to about n = 5.9, etc.
It is obvious that memory length is a most critical factor for structural emergence.
However, in this paper, we will deal with memory not in the frame of the single-
shot logic but within the population perspective below when connectivity will be
loosened through certain mechanisms. Memory will be one of those.
8
We are aware that the two logics of intra- and inter-round or inter-supergame
calculations should be explicitly modeled, not only under the restriction of ‘meet-
ing the same again’ but also with its various potential outcomes. The pay-offs of
a supergame of supergames would be the various potential capitalized pay-offs of
individual supergames.
848 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
i.e. a round will (randomly) end, and thus (1 − p1 )p2 the probability
that, in case a round has ended, the agent will meet the same again
next round.
Applying the discount factor for a structured supergame (which
mirrors the condition ‘to meet the same again next interaction and
next round’)9 to both the cooperator’s and defector’s payoffs, we
yield the following formulation of the single-shot condition:
a c
>! +b−c
(1 − ı2 ) (1 − ı2 )
or
a c
>!
{1 − p1 /(1 + r) − (1 − p1 )[(1 + r)(n − 1)]} {1 − p1 /(1 + r) − (1 − p1 )/[(1 + r)(n − 1)]} + b − c
Focusing on population size we can determine the critical popu-
lation size for the agents to be indifferent between cooperation and
defection, thus marking the border (maximum or minimum val-
ues) of the variable settings (especially of the group size n) leading
to cooperation:
(1 − p1 )
ncrit = (6)
{(1 + r)[1 − p1 /(1 + r) − (a − c)/(b − c)]} + 1.
As will be shown below, ncrit is a maximum critical mass for the
relevant cases that may not be exceeded for cooperation to emerge.
This condition holds the more (and population size may increase
the more) the larger the payoff advantage of the cooperator over
the defector (a − c) relative to the initial advantage of the defector
over his later payoff (b − c).
4.2. Co-evolutionary adaptation of size and expectations
Population size now may be considered a condition of the supe-
riority of cooperation determining probabilities and expectations
‘to meet again’; but also the relevant cooperating population may
be constituted, in turn, and its size explained and determined, under
certain expectations, given the incentive structure and the striving
of the agents to solve the problem (and to improve their long-run
payoffs).
Considering p1 , n and the incentive structure as parameters,
three out of four complexes of conditions will have to be determined
in a co-evolutionary process: (i) population size, (ii) consistent with
given or emerging expectations, (iii) successful institutional emer-
gence and (iv) the quantitative incentive structure which may be
parametrically changed, however, without dissolving the dilemma
structure as such. For instance, if the problem structure (a, b, c) is
given the maximum population size nmax , and minimum expectations
ı2min (or p1min , according to Eq. (6)) that still allow for successful
institutional emergence will have to be determined in the very pro-
cess of institutional emergence. Similarly, if the population size n
and expectations are given, the dilemma structure (a, b, c) that still
is consistent with successful institutional emergence will have to
be determined in the process of institutional emergence. For exam-
ple, if b and c are given, amin is to be determined; if a and c are given,
bmax to be determined, etc.
For instance, agents may determine the population size (e.g.,
through active partner selection as explained below) and some-
how experience this population size, while recurrently solving
the dilemma, at a given incentive structure. They may adapt the
maximum number of partners with whom they are capable of inter-
acting (given the incentive structure) in order to solve the problem,
9
The condition to meet the same again seems to be the perspective of the cooper-
ator. At least, the defector is indifferent in his behavior to whom he will meet (he is
assumed to always defect). However, he is not really indifferent since he would not
wish to meet the same again but to meet a new cooperator every round whom he can
exploit initially. Clearly the defector is interested in a small p2 or large population
size n while the cooperator is interested in the contrary.
according to their subjective minimum requirements of ‘expecting
to meet the same partner again’, in their efforts to create a solution
superior to common defection.
In the same vein, we may even consider the incentive struc-
ture being adapted according to the other conditions given,
i.e. population size, expectations and the effort of problem-
solving. For instance, incentives to cooperate (payments, rewards,
awards, reputations, etc.) may be somehow increased or decreased
(5)
by the agents in a myriad of interactions. Agents in this way may co-
determine payoffs in the very process of their interactions, through
contributions of various kinds. In this way, even the incentive struc-
ture may be considered endogenous.
4.3. The cooperative area in numerical results and the limits of
the ‘single-shot’ logic
The condition for the emergence of cooperation in the fully
deterministic single-shot logic (agents play repeatedly but choose
their strategy only once in the beginning) is that a neutrally stable
cooperative strategy (TFT) is available. ‘Neutrally stable’ is a strat-
egy that performs best against itself. It is not possible to displace
an established ‘neutrally stable’ strategy. The advantage of TFT over
ALL D (both against TFT) is (a − c)/(1 − ı2 ), the disadvantage of TFT
is (b − c).
The condition for ‘neutral stability’ of TFT as given in Eq. (5)
(a−c)
therefore can also be written as(5a)b − c =! (1−d2) .
The crucial variables here are the relation of the payoffs
(a − c)/(b − c), with 0 ≤ (a − c)/(b − c) ≤ 1, and the population size n,
with the relevant sizes 2 ≤ n. For each (a − c)/(b − c) there is a crit-
ical minimum or maximum population size n that follows from Eq.
(6). Whether ncrit is a maximum or minimum n depends on the
positivity or negativity of three expressions in Eq. (6).
If we lift for a moment the restrictions on n, ı2 , and (a − c)/(b − c),
we may get a continuous 2-dimensional space of n and (a − c)/(b − c)
values (for given r and p1 ) that displays areas of defection and
areas of cooperation (see Fig. 1 for r = 0.05 and p1 = 0.1).10 A sub-
space – marked in Fig. 1 – contains the permitted values for n and
(a − c)/(b − c), while ı2 increases with decreasing n, even above 1,
which in turn marks a non-permitted area. Particularly, the val-
ues between n = 1 and n = (1 − p1 )/(1 + r − p1 ) + 1 (which is the n-axis
intercept of ncrit ) involve ı2 > 1 and therefore are not part of the
permitted subspace.
The graph of ncrit [(a − c)/(b − c)] has two asymptotes, a horizon-
tal one at n = 1 and a vertical one at (a − c)/(b − c) = 1 − p1 /(1 + r).
The maximum critical mass starts right above ncrit = (1 − p1 )/(1 + r −
p1 ) + 1 for very small (a − c)/(b − c) (for (a − c)/(b − c) = 0, there is no
cooperating population) and increases quickly towards infinity for
rising (a − c)/(b − c) (given p1 = 0.1 as said). Also, to be sure, to the
right of the vertical asymptote, the area of cooperation continues
for permitted values until (a − c)/(b − c) → 1.
The size of the area of cooperation obviously depends on the
position of the vertical asymptote. The more this asymptote shifts
10
Note that p1 is to be interpreted as the probability that a round in the structured
supergame explained above is extended by at least one more interaction, and px1 there-
fore the probability that it is extended by at least x more interactions. The expected
value of the number of interactions per round therefore is x = logp1 0.5. For p1 = 0.1, the
expected value of future interactions is about 0.3. This is obviously a small number
and a rather adverse condition for cooperation. Axelrod (1984/2006), for example, set
some 200 future interactions implying a p1 of about 0.9965 as mentioned.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
Fig. 1. Area of cooperation and area with relevant values (i.e. n > 2) in a single-shot
supergame model with p1 = 0.1 and r = 0.05. Note: The graph of the ncrit function is
drawn black, the asymptotes are fat grey lines, and the borders of the area of relevant
values (n > 2) are thin grey lines.
to the left, the larger the area of dominant TFT will get. The posi-
tion of the asymptote is (a − c)/(b − c) = 1 − p1 /(1 + r) and is governed
by p1 (which is the parameter for the expected remaining num-
ber of iterations in the current round as mentioned in Footnote
9). In Fig. 1, the area of cooperation is relatively small, because
p1 is set to p1 = 0.1, a condition rather hostile to the emergence
of cooperation. Nevertheless, we can see that for an arbitrarily
set ‘meso’-sized population of, say, n = 5000, cooperation is feasi-
ble. Eq. (6) yields the condition that (a − c)/(b − c) = 0.9045. Thus, a
rather weak dilemma in the incentive structure is required, given
the rounds of continuous interaction (the degree of continuity in
the social relations) are rather short, slightly above one interaction
(again: p1 = 0.1). The value 0.9045 would be given, e.g., with a = 1810,
b = 2001, c = 1 (d < 1). Obviously, as soon as we increase the length
of round (increase continuity in the social relations) agents can c.p.
deal with stronger dilemmas or larger population sizes to generate
cooperation.
Over all, the result is rather optimistic. Large areas of the per-
mitted and relevant value combinations are areas with ‘neutral
stability’ of TFT, i.e. population sizes that allow for the emergence
of cooperation, under given conditions.
However, these positive findings are subject to two limitations:
(1) As said, neutral stability means that TFT cannot be driven out
of a population or, in our case, is statically the best ex-ante
response—to a population that is expected to consist of cooper-
ators. This population is therefore not exploitable by defectors.
Nevertheless, this does not explain, how a cooperative popula-
tion, or, in the single-shot case, the expectation of a cooperative
population may come into existence.
(2) The critical population size still is derived from a simple for-
mulation of the ‘probability to meet again’, a fully deterministic
perspective.
Therefore we need to modify the model to include a stochastic
population perspective, implying the possibility to engage in agency,
and to investigate the possibilities of the initial development of a
cooperative ‘culture’.
849
5. A process story of institutional emergence and size
The logic of expecting ‘to meet the same again’ obviously is a
theoretical straightjacket that restricts the process of institutional
emergence to a narrow determinism based on an ex ante and
once-for-all decision (‘single-shot’) on strategy choice according
to the probability to meet again in order to sanction and sanc-
tify earlier behavior of others. The deterministic ex ante decision
applies simultaneously to all agents in a population; a relevant
cooperating group is always identical with the relevant popula-
tion, i.e. no group within a population, no population ‘structured’
by types of agents (strategies). Finally, it excludes agency. Thus,
both some stochastic process, i.e. a population perspective, and
agency need to be considered. The latter has to be overcome by
‘loosening’ the total connectivity among agents. In a first step, we
will lay down the theoretical frame in a short qualitative ‘process
story’.
Generally, it will be assumed that probabilities ‘to meet’ and thus
population structure and size can be experienced by the agents, as
has been assumed so far, so that agents forms subjective expec-
tations, i.e. probabilities for the future mirrored in the discount
factor.
First, considering the solution above as a sequence or process, the
institutional solution cannot come about through narrowly ratio-
nal agents, i.e. short-run maximizers. It describes no process or
mechanism to achieve this result with short-run maximizers. These
would, in a process, only be capable of generating a series of one-
shot Nash solutions. Thus, as mentioned, an institution can only
emerge through habituation. The institution will have to be a ‘semi-
conscious’ phenomenon, and may be so as long as expectations of
conformity with it are met, supported, e.g., by a favorable numer-
ical result of the inequality (5a), and the incentive structure and
importance of expectations of a common future (to meet again)
remain unchanged. Therefore, institutional emergence has to follow
a broader and long-run rationality. [In contrast, the institution may
be abandoned through a more or less deliberate consideration when
a new (‘rational’) single-shot calculation (after some condition has
changed) no longer justifies rule conformity, when ‘surprise’, ‘dis-
appointment’, ‘frustration’ or ‘getting exploited’ has occurred.] This
broader rationality has also been elaborated under the perspective
of a ‘horizonal’ effect, i.e., individual (cognitive, planning) horizons
need to be extended, in a world of complementarity, cumulativity,
etc., if effective agency is to be gained (cf., e.g., Jennings, 2005).
Second, the institution, particularly an initial ‘minimum critical
mass’ of cooperators, may emerge on the basis of the individual
motivations (1) to escape repeated frustration from common defec-
tion (from aspiring b and receiving only c), and (2) to learn and
to increase knowledge, and particularly to explore what a differ-
ent behavior, namely common cooperation, may bring about (‘idle
curiosity’ or ‘instinct of workmanship’ as Veblen would have coined
it), to find a way to improve one’s economic situation (to gain com-
mon a’s rather than c’s). The payoffs for common cooperation may
not even be known (incomplete information) and may then get
explored by searching agents. The institution thus may emerge just
out of an agent’s vision that there is more to be gained than repeated
frustration. Agents who then make contributions to cooperation
need to be imaginative, ‘explorative’, and creative. Therefore, agency
needs to be carefully defined in an evolutionary process story.
Third, the individual who then starts to search and experiment
with different behavior will have to contribute repeatedly to the
change of expectations in favor of cooperation. The process, thus, is
cumulative in the sense that all agents must repeatedly and interac-
tively (sequentially) contribute (or will have to cumulatively punish
each other).
Fourth, these agents have also to be risk-taking and not be too
envious. The first to send a signal for a potential better common
850 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
future will have to take the risk of being exploited, at least once. He
also may never be able to compensate for this, as compared to the
other, even if common cooperation starts in immediate response
to his action. This agent thus needs to be mainly focused on his
own net gain which he has to compare only with his payoff under
continued common defection. Compared to this, he clearly will be
better off over time.
Fifth, with agents starting to learn, search, experiment, and
diversify behavior (in our two-strategies world, this means start-
ing cooperation), we may introduce a population perspective. Agents
then can no longer exactly tell the strategy of another agent whom
they will meet next. Behavior thus may be considered random,
and agents will have to experience the ‘true’ population shares.
The ‘pure’ expectation ‘to meet’ will be replaced by the expected
‘probability to meet a cooperative agent next round’ that we call
‘contingent trust’.11
Sixth, the evolutionarily stable initial ‘minimum critical mass’ of
cooperators becomes crucial. With a ‘minimum critical mass’, insti-
tutionalized cooperation may expand into a population initially
consisting only of defectors.
Seventh, while agents will have to experience ‘contingent trust’,
they no longer remain focused on just (the ‘probability to meet’)
the same agent. They will have to know about many agents as
possible. Thus, more elaborated capabilities of agents have to be
considered. Instances of such agency will be memory, monitor-
ing, building and transmitting reputation, and some active partner
selection based on the knowledge generated by these mecha-
nisms.
Eighth, group size of the relevant group then may be determined
by the individuals through some active partner selection. Agents
may in this way affect both the size and agent-type composition
of their individual ‘peer group’ of interaction partners. We may con-
sider the ‘meso’ group size then to be co-determined by the agents
who actively adapt their ‘relevant group’ to a maximum size and a
composition which still allows them to contribute to institutional
emergence, according to a given incentive structure and their con-
tingent trust. Specifically, they may even enlarge their relevant ‘peer
group’ when they can increase their contingent trust sufficiently
through some partner selection.
Real life has properties that may serve as criteria of partner
selection, i.e. proximity, or neighborhood, as either spatially, socially,
or professionally defined. Agents then may confine themselves to
some group of interaction partners through mobility, choice of
localization, ‘social exclusion’, or just actively picking out coopera-
tors, etc., in an effort to keep the expectation of cooperation high
while solving the dilemma problem. It is a rationality of ‘smallness’
of the peer group of an individual (sometimes with the danger, in
turn, of too great a ‘cliquishness’, petrifaction or sclerosis of institu-
tions attained, and subsequent institutional lock-in on an inferior
path).
Technically speaking, we then will ‘loosen’ total connectivity of
agents in favor of some ‘weaker’ connectivity in the sense that the
scale of interaction possibilities among agents becomes smaller (cf.,
e.g., Batten, 2001, 89ff.; Watts, 1999, 204ff.).
Ninth, the system then adopts an endogenous dynamic with dif-
ferent equilibria (attractors). Not only relatively small ‘minimum
critical masses’ will be illustrated but also some specific dynamics
of cooperators’ and defectors’ payoffs. This includes some ten-
dency to destabilize the rule once it is established and has diffused
among agents beyond the equilibrium size of the carrier group.
The probability to gain more through unilateral deviation from the
11
Note, however, that as soon as the minimum critical mass has been established
this way, our system below will still behave deterministically when moving to its
new equilibrium.
institution and exploitation, i.e. the incentive to defect, then may
increase for individuals in a growing population of cooperators.
These defectors, though, would not want to ‘kill the goose that is
laying golden eggs’ (Axelrod, 1984/2006, p. 221). This may lead to an
equilibrium ‘maximum critical mass’, being smaller than the whole
population involved, with a certain number of rule breakers that
the institution and its carrier group can afford without breaking
down.
An ‘embedding process story’, of course, does not sufficiently
explain the logic of the process of emergence of ‘meso’ size. Some
further implications of the single-shot logic combined with its
evolutionary embedding shall be explored in two steps, first the
population approach, second agency.
6. A population perspective and the ‘minimum critical
mass’—a graphical display
Behaviors may diversify, i.e. cooperation may occur now side
by side with defection, motivated (in frustration experiencing,
searching agents) as explained above. The number k of cooper-
ative agents in a population n now becomes a critical variable,
a stochastic variable causing a stochastic process at the outset,
i.e. until a ‘minimum critical mass’ is established, from where on
the new equilibrium will be reached deterministically as will be
seen. A single-shot calculation still applies. (Note that k has been
zero or n in the fully deterministic logic as discussed so far.) In
the following, the average payoffs per round for cooperation and
defection will be mapped over k. Some k then will indicate the
‘minimum critical mass’ of initial cooperators required for insti-
tutional emergence, i.e. the ‘minimum size of any coalition that can
gain by abstaining from the preferred choice’ (Schelling, 1978, p.
218).
Schelling’s illustration (1978, 217ff.; also applied, e.g., by Elster,
1989, 27-44) of the case ALL C vs. ALL D illustrated the usual out-
come of a PD one-shot Nash equilibrium: no individual incentive to
cooperate and thus no mechanism for a viable cooperating group to
emerge, at any k. This is equivalent to a very large population since,
in both the cooperator’s and the defector’s perspectives, there is a
zero ‘expectation to meet the same again’ in the very large fully
anonymous population and, thus, a zero expectation to be able to
sanction or to be sanctioned. Therefore, a zero commitment result-
ing. It follows ı1 , ı2 → 0, since p1 , p2 → 0 and, therefore, n → ∞. This
will serve as a benchmark in Fig. 2 below.
Obviously, for k = 0, the function of average payoffs for defec-
tors will be fALL D (k) = c, and for k = n, the cooperator’s function will
be gALL C (k) = a. If all but one cooperate (k = n-1), the last defector
left in this large population will always (and thus on average) gain
Fig. 2. Illustration of the payoff functions for TFT vs. ALL D, depending on the popula-
tion size (ı2 ), yielding, in a ‘small group’, a ‘minimum critical mass’ for institutional
∗ ∗
emergence, kSG , and the ‘relevant cooperating group’ at kmax = n.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858 851

f(k) = b, while at k = 1, the remaining cooperator will always (and is the only interaction these particular two will have in that very
on average) gain g(k) = d. Note that f is not defined for k = n and g large population).
not defined for k = 0. In all, the payoff of a defector always exceeds • (1b) Accordingly, the first to cooperate in the very large popula-
that of an ALL C cooperator for any given k, hence no ‘minimum tion when all others defect (k = 1) will continuously receive d, i.e.
critical mass’ beyond which cooperation will pay for an individual 0a + [(n − 1)/(n − 1)]d, on average per round since he will always
agent. This has also been Schelling’s (1978) benchmark configura- be exploited in the first interaction with any defector where no
tion. further interactions with the same agent will follow in a round.15
In contrast, TFT is considered the prototype of a more realistic In contrast, if everyone cooperates (k = n), each will continuously
and less ‘pathological’ (less ‘masochistic’) type of cooperation than receive a in every period.
ALL C. It sanctions and thus makes future expectation relevant, as
reflected in the single-shot inequality. Using the general ı2 of Eq.
(4), the payoff functions will yield average payoffs according to12 It follows that f(ı2 → 0, k = [0, n − 1]) will always be above g(ı2 → 0,
k = [1, n]) in their common domain of k = [1, n − 1].
k k c
 
(n − k)
 c

fALLD = +b−c + (7a)
n n (1 − ı2 ) n (1 − ı2 )
• (2a) The situation changes as the population gets smaller. In a very
and small population, when ı2 → 1, defectors will receive c on average
k k a
 
(n − k)
 c
 per round when k = 0, as was the case in the very large popula-
gTFT = + +d−c . (7b) tion. However, if all but one agents cooperate (k = n − 1), this last
n n (1 − ı2 ) n (1 − ı2 )
defector will receive only slightly above c on average, since he will
This will explain the conditions of evolutionary stability of TFT in be able to exploit a TFT cooperator only once, then getting b, i.e.
an ALL D population. Note that this includes the earlier fully deter- [(n − 1)/(n − 1)][c/(1 − ı2 ) + b − c] + 0[c/(1 − ı2 )], and from then on
ministic approach (the single-shot of a ‘structured supergame’) will get c with him (over all rounds in the supergame. Note that, in
through ı2 in a partially stochastic approach. the smallest group possible (two), he will encounter the same TFT
According to Axelrod’s (1984/2006, Chapter 3) illustration, with agent again and again over all rounds, and the TFT will remem-
ı = 0.9, b = 5, a = 3, c = 1, d = 0, this yields fALL D = 14(k/n) + 10(n − k)/n ber him from one interaction to another and thus also between
and gTFT = 30(k/n) + 9(n − k)/n, i.e. for gTFT > ! fALL D , a minimum of rounds.)
k/n > !1/17 or about 5.9% TFT-cooperators would be sufficient to • (2b) If only one cooperates in the very small population (k = 1),
allow for the institution of cooperation to emerge. the TFT cooperator will receive zero from common cooperation,
Generally, this simple system has two stable equilibria and one i.e. 0a, but somewhat slightly below c, since he will be exploited
unstable equilibrium: k/n will assume either zero or one or the once and then will continue with virtually the same interaction
‘minimum critical mass’ kmin ∗ (given n) that supposedly has been partner with c on average per interaction for both. And again, if
attained through ‘stochastic’ behavioral diversification (search, all cooperate (k = n), they will receive a, as in the case of the large
experimentation, etc.)13 : population.

(i) 0 if f (kt ) > g(kt )
kt+1 As can be easily seen, the large population again yields the one-
= (ii) 1 if f (kt ) > g(kt )
n ∗ shot Nash situation, as in the ALL C case. In contrast, the very
(iii) kmin,t , if f (kt ) > g(kt ).
small group (SG) yields some minimum critical mass, kSG ∗ , beyond

We will shortly discuss the simple logic of the population
perspective with the expectation ‘to meet a certain agent next inter-
action’.
For cooperation to become ‘evolutionarily stable’ [case (ii)], the
condition of Eq. (7b)>! Eq. (7a) yields the minimum critical mass
∗ ∗
kmin or the critical portions (k/n)crit :
 k ∗ (c − d)
= .
n crit [(a − c)/(1 − ı2 ) − b − d + 2c]
This, then, can easily be depicted in Schelling’s graphics as illus-
trated in Fig. 2 below (see also, e.g., Elster, 1989, pp. 37–39, who,
however, did not discuss a clear-cut analytical result in terms of
population or group sizes). The cases are obvious:14
• (1a) In the very large population where ı2 → 0, when nobody coop-
which cooperation pays for all individual agents (see Fig. 1 below;
note that all graphs go from k = 0 to k = n, for the sake of simplic-
ity).
The emergence, or creation, of the minimum critical mass may
be a ‘motivational’ problem as described, or an ‘organizational prob-
lem’ (Schelling) (if we would refer to hierarchies or authorities), or
might require some policy design (on this see below).
The ‘relevant cooperating group’ will be the whole population
(8)
in this case as said, i.e. all defectors will switch to cooperation in a
logical second.
There is a maximum group size (at some ı2min ) in which kSG ∗ comes
∗ ∗
into existence, i.e. f (ı2 , kSG = n) = g(ı2 , kSG = n). At this maximum
∗ = n, the ‘minimum critical mass’ and the ‘relevant cooperating
kSG
group’ equal the whole population size. With decreasing group size
∗ will start moving, with an increasing ı , from the right end of
kSG 2
the k-axis to the left, stopping at some kSG ∗ , as illustrated. With b = 4,

erates (k = 0), the defectors continuously receive c on average a = 3, c = 2, d = 1, a calculation of ı2 that allows a kSG ∗ to come into

per interaction. However, when all others cooperate (k = n − 1),
the last remaining defector (the nth member of the population)
will receive b, i.e. [(n − 1)/(n − 1)][c/(1 − ı2 ) + b − c], all the time
since he will always meet another cooperative interaction part-
ner whom he can exploit in each of their first encounters (which
12
As said this still represents an ex-ante single-shot calculation.
13
We will consider agency capacity below.
14
Note that the agent who decides to cooperate or defect based on these equations
will not be part of the relevant k’s and n’s.
being at k = n (i.e., (k/n) = 1) of course yields ı2 = 0.5. (See also Fig. 3
below, where we enter the relevant area at (k/n) = 1 on the graph
from above [i.e., (k/n) = 1 at ı2 = 0.5].)
In the following, we will elaborate on the population perspective
and build upon mentioned behavioral diversification, considering
agency, specifically information gathering and selection capabilities.
We start from ‘contingent trust’.
15
Possibly, a case of pathological behavior similar to an ALL C player.
852 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
∗
Fig. 3. Graph of the function of (k/n)crit . (For Axelrod’s (1984, 2006) Incentive struc-
ture a = 3, b = 5, c = 1, d = 0). Note: The graph of the function is drawn black, the
asymptotes are fat grey lines, and the borders of the area of permitted and relevant
values are thin grey lines.
7. ‘Loosening’ connectivity: ‘contingent trust’, monitoring,
memorizing, reputation, and selection
7.1. Contingent trust
In the population perspective, i.e. in a ‘structured’ population
with portions of (representatives of) different strategies, agents can
no longer focus on ‘the one’ interaction partner whom they know
and need to meet again next interaction and next round in the
single-shot logic. Rather, they have to learn about many potential
partners and the distribution of strategies among them, in order to
increase their contingent trust ık in their inter-round transitions.
While we assume agents to face random occurrences of partners
with different behaviors, what counts now is their expectation to
meet a cooperative partner next round, i.e.
k behavior of third agents (being outside of his own funnel but inside
ık =
n their funnels). Assume further, for the sake of simplicity,17 that only
(similarly, e.g., Elster, 1989, p. 34).
Compared to the ‘benchmark knowledge’ of only one interaction
partner, the population perspective, therefore, increases the num-
ber of agents that the individual has to ‘know’. At least, he must get
to know a representative sample of the population.
Let us shortly consider which agency capacities specifically are
required in order to increase this expectation to a sufficient level.
7.2. Memory and monitoring
First, knowledge of others will increase by adding some memory
capability (a memory of one period only has been implied in TFT so
far).
Second, agents should be considered capable of monitoring con-
current interactions between identifiable third parties (see also,
e.g., Elster, 1989, 40f.). In the fully deterministic single-shot he could
only ‘monitor’ the same agent, so to speak.
Assume six memory periods for own interactions and fewer
memory periods for interactions among monitored third parties.
The agent memorizes only his own interactions over his maximum
number of memory periods, while the more ‘distant’ third parties
are16 from the monitoring agent the shorter the agent memorizes
16
We will consider a geographical or social topology, or neighborhood, below.
their monitored actions. Assume the agent can monitor ten other
interactions.
Thus, a monitoring-memory funnel may exist where the most ‘dis-
tant’ monitored third-party interactions are memorized only for
one period, i.e. if they have been monitored in t−1 , the next less
distant only for two past memory periods, t−2 , etc., and the own
interactions from t−6 . For a simple numerical example of informa-
tion gathering on a ‘meso’-sized number of agents, see Fig. 4 for an
illustration.
Consider, for example, agents endowed with the ability to mon-
itor, for example, 10 other concurrent interactions at a time. Agent
A then will ‘know about’ 21 agents excluding himself from current
monitoring. From memory, he would know about 21 more from t−1 ,
17 from t−2 , 13 from t−3 , etc., in sum 87 other agents.
However, there is no perfect mobility in that monitored and
memorized part of the population. Assume that in addition to A all
Xs stand for immobile agents with fixed positions in the neighbor-
hood of A who appear repeatedly while only 50% of the monitored
and memorized part of the population are changing all the time;
assume for the sake of simplicity that these are the Ys. In this way,
A knows only about 58 different other agents.
Also assume 50% observed cooperative different other agents (i.e.
ık = 0.5) (or actions, resp.). The agent’s ‘knowledge’ of (the behaviors
of) cooperative different other agents then is 29.
Compared to the ‘benchmark’ of ‘the 1’ he knows (and only needs
to know) in the fully deterministic single-shot logic, thus the rele-
vant population size which was 3 at p2 = 0.5 in the logic so far, has
increased to 59, and the number of cooperators from a maximum
of 3 up to 30 (including A), i.e., by 10.
7.3. Reputation chain
Third, reputation may further increase the ‘knowledge’ about
other agents, i.e. it further increases the probability to meet a part-
ner next round whose (earlier) behavior the agent knows, in this
case through a third person.
Assume, for example, that each agent has a monitoring-memory
funnel that includes 29 known cooperative different other agents
as above, and each agent can ask all of these cooperators about the
(9)
half of them have own funnels that differ from A’s funnel. Thus A may
get new information only from 14 known cooperators about 28 new
third (different other) cooperative agents.
In this way, the number of other agents ‘knowable’ by an agent
will multiply and sum up over the number x of steps of the chain
according to k = x = 0 → ∞ 14x ·28, which, for instance, would yield
around 77,000 known agents in only three steps of the reputation
chain. This would definitely have to be considered ‘meso’-
sized if these agents would constitute a ‘relevant cooperating
group’.
Of course, the reputation chain will not work with the same
effectiveness across the different steps or links of the chain
(the ‘neighborhood’ geography), but rather may get ‘thinner’ and
‘weaker’ with increasing distance from the interrogating agent.
Information feedback to the original agent will be increasingly
incomplete and subject to leakage. Even if we assumed a consid-
erable leakage of 50% per step, we would yield 14·7·3.5·28 ≈ 9600
17
A ‘half-distant’ neighbor’s funnel, of course, will not completely be identical with
A’s funnel but will overlap so that he might add, say 10 or 20 or more % informa-
tion (depending on his distance) to A’s knowledge from his own funnel. This would
require a more elaborated model of overlaps of funnels in a topology. We will neglect
this and simply assume that only 50% of cooperators in A’s funnel can add informa-
tion, these, however, may ask the complete set of their assumed 28 different other
third cooperative agents.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
Fig. 4. A monitoring-memory funnel—illustration.
agents which also may be considered to reach into ‘meso’
sizes.
Also, the set of agents may ‘bend’ into a ‘spherical’ topology so
that more remote agents may increasingly be already ‘known’ ones.
In this way, it may easily be that k may reach some limit at some n.
In all, this should illustrate that these agency mechanisms, i.e.
active acquisition of information about others, may easily lead us into
a ‘meso’-size scale of populations of ‘known’ agents and ‘known’
cooperators, and with this into the individual experience of the
‘true’ composition of the whole population.
No further devices and mechanisms such as signaling and substi-
tute indicators such as sex, race, age, living area, formal education,
certifications, identity cards, corporate uniforms, appearance, cere-
monial behaviors or other signals would be needed. Note, however,
that this is just a first intuitive understanding of a ‘relevant
meso-sized cooperating group’ while we develop a theoretical
understanding below.
Note also that we have to assume ‘correct’ reputation here, i.e.
an information requested and received by A from B on a potential
interaction partner C is not subject to some ‘strategic’ tampering in
favor of B.
Finally, if such verbal communication comes into considera-
tion, this is something that non-cooperative game theory basically
would render senseless and ignore (the ‘cheap talk’ issue).18 ‘True’
verbal communication itself can of course not be explained through
the same process and logic of institutional emergence that is inves-
tigated here. Qualitative theorizing on institutions and multi-level
systems of reference among norms and institutions, of course, pro-
vides more opportunities to go beyond such limitations (cf., e.g.,
Hodgson, 2006) and cooperative game theory may have a major role
to play here.19
However, why and how should a certain maximum percentage
of cooperators above the critical minimum mass (which comes
into existence through another agency mechanism, as explained)
come into existence in the population? ‘Loosening’ the connection
between expectation, namely contingent trust, and population size
through active partner selection will help to understand group size
adaptation.
7.4. Adapting group size through partner selection and a
‘maximum critical mass’ smaller than the whole population—a
graphical display
Monitoring, memorizing and using a reputation chain can be
considered informational preconditions for selecting agents. Assume
that interaction partners appear in some random sequence sub-
18
This might be a point for a complementary use of cooperative game theory. We
owe the advice on cooperative game theory an anonymous referee and will denote
the logical points of its complementary use in two more places in our argument.
19
One referee has mentioned that there is no rule, institution or game in isolation,
but systems of games, rules and institutions, partially hierarchic, and by all means
mutually interdependent, often mutually stabilizing. In our view another case for
a co-evolutionary analysis. This of course is not confined to informal institutions
but must include the state and formal institutions where, again, cooperative game
theory may be very promising.
853
ject to ‘distance’, i.e. in a neighborhood topology the probability
of appearance of potential interaction partners to the individual
decreases with decreasing proximity. But also knowledge about
agents decreases with decreasing proximity as discussed in the
reputation chain.
Assume further that in the process of search, experimentation
and behavioral diversification motivated as described above, coop-
erators start to select agents (on such agency capacities, see, e.g.,
Davis, 2007, 2008; Dolfsma and Verburg, 2008; Leydesdorff, 2007).
It is supposed that the cooperative agent has some capacity to reject
interaction with a known defector, say, a capability to reject on
average every second defector he encounters.
Cooperators also may prefer to interact recurrently with agents
they cooperated with in the past rather than with known defectors
or new unknown agents. The size of the ‘relevant cooperating group’
thus may be determined by myriads of acceptances and denials of
interactions.
Lab experiments have shown that agents indeed try to reduce
net complexity through active selection and active building of
neighborhoods (see, e.g., Harmsen-van Hout et al., 2008), and that
network effects can informally be attained by groups constituted
through the selective interactions of individuals (e.g., Tucker, 2008).
Spiekermann (2009), for instance, has recently shown that a selec-
tion mechanism (i.e. choosing or excluding/rejecting partners)
allow may already allow for the emergence of cooperation in an n-
person public-good game. However, again, size way not an explicit
issue in his analysis.
In this way, the cooperator will be able to increase the ratio
of cooperators in his individual interactions. Individual i thus will
be able to increase (ki /ni ) of the group of his past and present
interaction partners and in this way increase ıki ↑ = (ki ↑/ni ) while
the surrounding population remains the same size n. He decou-
ples ıki from the general ık existent in the population, namely
ıki > ık . (Note that he is not assumed to be able to reject every
defector.)
Therefore, he will also increase his average outcome in a pop-
ulation of given size and structure, and cooperators together
can make cooperation increasingly more attractive. Particularly,
through selection cooperators can change their payoff curves and
realize higher growth rates than in the linear case of the simple pop-
ulation curve above which is dependent on ık rather than ıki (Eq.
(7b)). Since cooperators’ payoff cannot exceed a we have to assume
a degressively increasing payoff function. With a perfect selection
capacity, the curve would start at d at k = 1 and jump to a with
k ≥ 2, i.e. cooperators would always interact only with cooperators.
Since selection, however, is not perfect the curve will increase more
slowly as will their ıki , in spite of the fact that they may become
more in the whole population through both initial experimenta-
tion and the increasing attractiveness of cooperation (through that
very selection).
Specifically, we may assume that the effectiveness of the part-
ner selection declines because the knowledge about agents does not
grow by the same rate as new (potential) interaction partners occur.
Memory and monitoring provide knowledge about a limited num-
ber of agents and the reputation chain mechanism fades away with
854 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
distance as said. So the more new partners occur and the greater
the distance they come from the less the agent will already know
about them.
Overlapping individual partner selections may then constitute
the ‘relevant cooperating group’ [see also, e.g., Oestreicher-Singer
and Sundararajan, 2008, for minimum relations and networks that
still allow for institutional diffusion (‘small worlds’) and segregation
of a population, see, e.g., Foley, 1998, 18ff., 38ff., 61ff.; Watts, 1999,
204ff.; Batten, 2001, 89ff.].
In all, the function thus may increase according to some factor
˛, 0 < ˛ > 1, connected to ki /ni , i.e. (ki /ni )˛ .
Defectors, on the other hand, will be sorted out to some degree by
the early cooperators and thus will not fully profit from an increas-
ing number of cooperators as in Eq. (7a). Their population payoff
curve thus will remain below the curve given by Eq. (7a) and coop-
erators’ payoff curves may rise above the defectors’ payoff curves
already for a small number of cooperators.
However, while cooperators cannot get above a, defectors will
still increasingly profit from a growing number of cooperators. Their
payoff function may be considered progressively increasing, subject
to (k/n)1/˛ . This may constitute a maximum critical mass, or rele-
vant cooperating group below the size of the whole population (see
below).
In order to theorize and illustrate the simple logic of such seg-
regation, different non-linear cooperation payoff curves have often
been considered in the literature.
For instance, Schelling (1978, 104f., 239ff.) has referred to net-
externalities to explain progressively increasing cooperative payoff
functions. This implies some additional positive mutual externalities
among cooperators, perhaps even above what cooperators gener-
ate on average in the standard PD (i.e. a). Schumpeterian innovation
economists have argued in favor of the cumulative character of
new knowledge, both inter-personally and inter-temporarily, so that
cooperative payoffs may increase through an extra synergetic effect
(through mutual learning, imitation, etc.) (see, e.g., Pyka, 1999,
98ff.).
Specifically, S-shaped curves have been considered in such con-
texts. For instance, in his technology choice model with increasing
returns (i.e. net-effects), Arthur (1989, 123ff.) has made use of
logistics curves. While a technology adoption function maps the
probability of choice of a certain technology by the next choosing
agent against the number of those who have chosen this technology
so far, i.e. equivalent to a payoff function depending on the num-
ber k who have ‘chosen cooperation’, some ‘improvement function’
would mirror additional increasing returns to adoption. Particu-
larly, he considered cumulative learning (by using) effects and
‘coordination externalities’ (p. 126). He also considered a ‘bounded
improvement function’ when effects become ‘exhausted’. The popu-
lation would split up then, in an equilibrium, with the coexistence
of more than one technology, one portion using the ‘dominant tech-
nology’ another one some ‘minority technology’.
Also, Cooper and John (e.g., 1988) elaborated on economies
with ‘strategic complementarities’, or synergies. Going beyond ‘pos-
itive externalities’ of agent A which just increase the payoff of
agent B, ‘synergies’ imply that an ‘increase’ in agent A’s strategy
increases the marginal return of agent B’s own action. A’s strat-
egy thus is an increasing function of agent B’s strategy, and vice
versa. A ‘synergetic’ reaction function displays some ‘multiplier
effect’ and is be S-shaped (pp. 445ff.). It can be matched against
a benchmark of a linear curve representing a series of Nash equi-
libria. This involves multiple equilibria at the intersections of both
curves. The higher intersection basically is equivalent to a coop-
erative equilibrium. The economic examples that Cooper and John
discuss include net-externalities through coordination in supplier
networks and demand coordination among multiple industries in the
business cycle.
The sigmoid function reflects the idea of initially exponentially
growing payoffs for cooperative behavior (cumulative learning,
‘synergies’, net-externalities, and the like) which later ends in
maturity, when the ‘resource of cooperation’ somehow becomes
‘exhausted’ (see also Elster, 1989, 28f., pp. 32–34, for a logistics
curve, while discussing the ‘technology of collective action’).
While the composition and ‘contingent trust’ for the whole
population may remain relatively low cooperators increase the
composition of their own interaction partners (ki /ni ) and thus may
increase their average payoffs above the level of the average payoff of
defectors through selection. This in turn will increase the portion of
cooperators in the population and contingent trust. With degres-
sively increasing cooperators’ payoffs and progressively increasing
defectors’ payoffs this will occur at a relatively small ‘minimum criti-
cal mass’ compared to the linear case of payoff functions above (Eqs.
(7a) and (7b)).
Also, sorting out (to some degree) one’s interaction partners
is equivalent with the assumption that cooperators ‘self-commit’
to interact with ‘their kind’. Or: self-commitment to interact a cer-
tain number of rounds among cooperators is equivalent to some
capability to select partners.
Here, Axelrod’s use of ı applies. He focused on the number of
interaction rounds among agents with a somehow committed ongo-
ing relationship. If two agents had a given (expected) round length
of one interaction, the probability that the interrelation will end is
1.0; accordingly the (expected) probability to ‘meet the same again’
is ı = 0 (i.e. p1 and ı1 according to our definitions would be zero,
one-interaction rounds; if this applied also to the inter-round prob-
abilities p2 , also ı2 became zero; this is equivalent with the very
large group). With two interactions in total, the probability that the
interrelation (the round) will end is 0.5, and also ı = 0.5. Axelrod had
set up his tournaments in this way, with an average round length of
200 interactions, i.e. a probabilistically determined chance of about
.003 of ending an interrelation after each interaction, corresponding
to ı ∼ 0.997 (1984/2006, 63ff., 212f.; see also footnote 10). While this
is no advantage for defectors a high ı generates a relative advan-
tage for TFTs since a > c. In this way TFTs can be considered to be
self-committed for some 200 interactions per round see footnote
10.
In addition, Axelrod had assumed that an invading group of TFTs
is relatively small so that it only marginally affects the ALL Ds’ pay-
offs. Defectors’ payoffs are considered unchanged (!) since the great
bulk of the ALL Ds’ interactions would still take place among them-
selves. If, for instance, ı = 0.9 for both cooperators and defectors and
b = 5, a = 3, c = 1, d = 0 the defectors gain on average only 10 with each
other (a simplified version of Eq. (7a). The TFTs, in contrast, have to
take into account considerable portions of interactions with ALL
Ds, according to Eq. (7b). The example yields k/n > ! 1/21, i.e. only
around 4.8% of invaders can succeed and survive in a ‘hostile’ social
environment.
In this way, Axelrod was able, through ‘committed’ large num-
bers of interactions, both to keep expectations (ı) high and increase
cooperators’ group size (k or k/n, resp.), while the minimum critical
mass would be very small (below 5% in the example). This implies
abandoning total connectivity by introducing selectivity through self-
commitment.
We illustrate the effect of partner selection in Fig. 5. Coop-
erative payoffs now may quickly exceed the average defector’s
payoff.
The payoff functions ‘with selection’, gsel. for cooperators and
fsel. for defectors, can be represented as explained by:
k  k 1/˛  c
  (n − k)
 c

fsel.ALLD = +
n n (1 − ı2 ) + b − c n (1 − ı2 )
(10a)
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858 855

Also, we must not necessarily think of individuals being clear-

cut cooperators or defectors at any given point in time, but may
think of certain portions of cooperative and defective actions in the
sets of actions of every single individual.
Finally, note again that we must not and do not assume that the
average payoff from cooperation ever exceeds the original payoff
value a from common cooperation at any group size.

8. A real-world example: trust polls, country size, and

macro-performance

Before we come to conclude, a short example should suffice to

indicate the empirical, and topical, relevance of ‘meso’ size.
Trust polls, nowadays carried out regularly in most countries
on behalf of the World Bank and others major organizations (see,
e.g., Knack and Keefer, 1997; Keele and Stimson, 2002), namely the
Fig. 5. Illustration of the effect of partner selection (and ‘exhaustion’ thereof) on the World Value Survey, include general trust questions, such as: ‘Do you
payoffs from cooperation (gsel. ) and defection (fsel. ), indicating the ‘meso’-size area think you can trust the next person you will encounter?’ Obviously,
∗
of the ‘relevant cooperating group’, kmax . this is fully equivalent to the expectation to meet a cooperative
agent in the next round, i.e. contingent trust.
Such polls have brought about surprising differences and even
and considerable divergence over time in trust levels among presumably
k  k ˛  a
  (n − k)
 c
 similar and converging countries (e.g., leading OECD countries).
gsel.TFT = + , However, this is particularly the case between large and small coun-
n n (1 − ı2 ) n (1 − ı2 ) + d − c tries. Similarly surprising was the fact that such trust levels have
0 ≤ ˛ ≤ 1. (10b) turned out to be highly correlated with economic and social perfor-
mance in broad areas (see also, e.g., O’Hara, 2008).
It has indeed been argued in the more recent ‘varieties of capi-
Note that the cooperators’ curve is set on top of the ‘worst-case’
talism’ literature that it may particularly be smaller countries which
cooperative function, the ALL C vs. ALL D case, equivalent to the
display such high trust and performance (see, e.g., Kesting and
very large group, where ı2 may be very small. This may illustrate
Nielsen, 2008; see also already Fukuyama, 1995, Chapters 3, 28,
that the constitution of a meso-sized relevant cooperating group
pp. 23ff., 335ff.). Conventional economics has mostly stressed the
may occur even in a large population, depending on the exponents
disadvantages and volatility of small countries (see, e.g., Barro and
˛ and 1/˛. This would of course be the case all the more on top of a
Sala-i-Martin, 2003; Alesina and Spolaore, 2003). The empirical
‘small group’, i.e. with flatter curves (as given in Fig. 2).
record of the relation between smallness and economic perfor-
As said, a reduction of the ‘minimum critical mass’ compared
mance is not obvious and clear-cut. Many have argued in favor of
to the linear case (Fig. 5) may occur, so that random diversification
advantages of smaller countries (see, e.g., Kuznets, 1960; Easterly
of behavior combined with early selection might bring about the
and Kraay, 2000), particularly their superior adaptability, learning
‘minimum critical mass’.
and cooperation conditions as dependent on proximity and inter-
A general solution to determine the lower and upper intersections
action density (see, e.g., Cantner and Meder, 2008).
of fsel. and gsel. , from the condition
Considering critical factors in the ‘deep structure’ of coun-
Eq. (10a) = ! Eq. (10b), is given by
tries, such as futurity (expectations) and size as investigated in this
 k 1/˛  c
 k  k ˛  a
 paper, specifically small countries that are internally structured
+ (d − c) − in overlapping functional, spatial, professional, organizational, or
n (1 − ı2 ) + b − c n n (1 − ı2 )
jurisdictional communities, groups, networks, associations, etc.,
= d − c. (11) have turned out to display such favorable trust (and high social
capital, indicated, for instance through membership and participa-
This yields a complex function of k which has solutions with at tion), and performance properties. It appears that the principle of
least two values of k. ‘smallness’ has been internally generalized by them (through cumu-
As a numerical example assume n = 1000, ˛ = 0.5, ı2 = 0.1, and lative historical process rather than deliberate political design, of
∗
again b = 5, a = 3, c = 1 and d = 0. This will yield kmin ∗
≈ 82 and kmax ≈ course), so that they can make use of ‘meso’-sized ‘arenas’, ‘plat-
696. This illustrates that the equilibrium ‘relevant cooperating group’ forms’, and groups. In this way they may generate trust and facilitate
size is below the size of the whole population and a ‘meso’-sized the emergence of institutionalized cooperation, related innovative
institutional carrier group. action capabilities, and macro performance.
This also reflects the fact that the established informal institu- This typically applies to Scandinavian countries, given their
tion may bear, and even provoke, some degree of defection, by making overall sizes, residential structures, enterprise size structures, dom-
it, with increasing k, ever more profitable to deviate. Any institu- inant interactive workplace organization, general organizational
tion, in fact, exists, and may survive, in the face of a certain number participation, informal networks structures and policy frame-setting
of defectors. However, these defectors do no longer endanger the (particularly safeguarding some level of social integration, stability
institution as such.20 and security) (see for instance the rich material on the Danish case
in Jorgensen, 2002; Lundvall, 2002; Edquist and Hommen, 2008;
Christensen et al., 2008; Holm et al., 2008). In this way, even a 5.4
20
A referee has suggested here that wherever full rule adherence can be observed,
non-cooperative game theory would explain only the spontaneously voluntary part.
The ‘other segment involves cooperative game theory’, he argued. We basically game–theoretic argument, we will suggest a slightly different place for cooperative
agree. Regarding a role for policy with respect to the results of our non-cooperative game theory, though.
856 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858

Table 1 dimensions of contingent trust (expectations) and ‘meso’-sized

Indications of ‘meso’ structures, institutionalized cooperation, general trust and
group formation, might play a more important role in this field in
macro performance.
the future.
1.1 General trust survey data Other than the usual political, welfare-state, social integration
Selected countries 1990 1999/2000 2004–2008 and social-security argument explaining the ‘Scandinavia plus’ vari-
Denmark 57.7 66.5 71.3*
ety of capitalism, we would suggest to empirically investigate the
Finland 62.7 57.4 58.8 size dimension working in the ‘deep structure’ of the interaction sys-
Norway 65.1 67.2* 69.2* tems of countries as one critical factor among others.
Sweden 66.1 66.3 68.0 The methodological problem such an empirical study immedi-
Netherlands 53.5 60.1 44.5
ately faces is that many ‘meso’ structures are already historically
To compare: Germany 32.9 37.5 34.1
given in those countries, e.g., settlement systems (little urbaniza-
1.2 General trust correlations (13 smaller and larger OECD countries, if not tion, rural structures), membership and participation rates (i.e. social
indicated else) capital) network cooperations of firms and the like. Can we iso-
Indicators Correlations with general trust, survey wave late these factors and their impacts on institutionalization and
2004–2008 ‘meso platform’ emergence? On the other hand, can empirically
Population size −0.34 identify the crucial factors and mechanisms of our model, such
Membership 0.88 (9 countries) as incentives, expectations, reputation chains and partner selec-
Share of cooperating innovators 0.80 tion to trigger further ‘meso’ generation? We can expect only to
GDPpc 0.51
explain the continuing reproduction, sometimes the growth and the
GDP growth 0.34
sometimes surprising resilience under adverse conditions rather
1.3 Other correlations between empirical indicators of ‘meso’ and cooperation than the original emergence of meso structures, supporting expec-
tations (namely general trust), institutionalization of cooperation
Population size/share of cooperating innovators: −0.89 (6 countries)
Membership/share of cooperating innovators: 0.84 (5 countries) and, finally, macro performance. Specifically, can we explain further
trust growth in spite of fast change, partial de-regulation, global-
Source: Own calculations based on World Value Survey data;
http://www.worldvaluessurvey.org.
ization?
*
Estimated. The overall causal chain might overlap between the model and
empirical data in the following way:

mio population country may be sufficiently interconnected through

‘meso’-arenas to mobilize reputation chains and to generate high
levels of trust, institutionalized cooperation, and socio economic
performance.
Note that the Scandinavian countries (incl. NL, A, and some oth-
ers) are not only leading in social areas like little poverty, even
distribution, high employment, education, social security or subjec-
tive well-being but are also in the leading groups with innovation
rates, GDP pc, speed of structural change, labor market mobility,
globalization rates, future expectations, etc. Also, they are largely
not too much affected by the current financial meltdown. Find some
indicators of the relative positions of Scandinavian countries and
small countries in general in Table 1.
The World Bank, the OECD, EU, and others may have to pro-
ceed to a more complex explanation of trust, and the theoretical
framework of institutional emergence, with its co-evolutionary
(1) given formal and informal inner ‘meso’ structures: empiri-
cal (e.g., population size; settlement structures; membership;
share of cooperative innovators)
(2) the ‘mechanisms’: model
(3) reproduction of informal inner ‘meso’ structures: model, empir-
ical (e.g.,  membership)
(4) high and increasing trust/institutionalization of coopera-
tion/social capital regeneration: model, empirical (e.g., 
general trust;  share of cooperative innovators)
(5) high macro-performance: model, empirical (e.g., GDPpc;
GDP).
Fig. 6 gives an illustration of the sequence of an empirical
research based on our model and focusing on trust and performance
in ‘small (and well-structured)’ vs large countries.

Fig. 6. An empirical research setting: socio-economic conditions and results of ‘meso’-structures—empirical indicators (in circles) and theoretical mechanisms and predictions
(in rectangles) combined. A tentative scheme.
 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
9. Conclusion
Against the background of an ‘evolutionarily embedded’ sim-
ple theory of ‘meso’ as investigated in this paper, the co-evolution
of ‘meso’ size, futurity (expectations), and institutional emergence,
given the incentive structure, will have to be further theorized, for-
malized, specified and finally quantified in simulations. The critical
factors explored here are:
(1) the incentive structure;
(2) a distribution of a population in a topology, i.e. proximity and
‘neighborhood structure’, with the corresponding rules of inter-
action, (e.g., mobility rules);
(3) an initial (stochastic, although ‘motivated’) distribution of strate-
gies, and a ‘minimum critical mass’ of cooperators;
(4) ‘agency’ mechanisms such as monitoring, memorizing, reputa-
tion building and chain using, as well as partner selection based
on related information;
(5) potential additional mutual externalities of cooperation, cumu-
lative learning, or ‘synergies’, typically justifying a degressively
growing payoff curve; and, finally;
(6) the ‘relevant cooperating group’, typically smaller than the whole
population involved and thus ‘meso’-sized. This may be repre-
sented also by segregation patterns.
The process, logic and critical factors discussed here may
allow for a class of complex models and simulations focusing on
‘meso’. This may easily merge with lines of complex economic
research based on increasing returns, positive feedback or cumula-
tive causation. Technically speaking, such models will escape total
connectivity among agents in patterned networks with a more
loose connectivity. While the bounds of the simple logic are nar-
row, the space of formal modeling and quantitative simulation is
virtually infinite. Complex modeling and simulation, though, has
been beyond the scope of this paper.
A more general ‘meso’-economics might be envisaged, i.e. the
economics of the emergent ‘mid-size’ level for the coordination
and capability increasing of a population in an evolutionary pro-
cess. Its applications, such as cultural emergence, production and
innovation standards, paradigms, and cultures, information gover-
nance and open source in a fragmented and interconnected ‘new’
economy, spatial industrial organization and agglomeration, general
trust and macro performance, etc., have high practical relevance.
Of course, our approach confirms that we cannot expect
a non-trivial, automatic emergence of a stable, self-sustaining,
Pareto-efficient institutionalized equilibrium, i.e. no ‘deliberation-
free’, ‘hierarchy-free’, or ‘state-free’ stable private self-organization
or ‘spontaneous order’. For instance, the initiation and generation of
a ‘minimum critical mass’ may require considering the role of pub-
lic policy (see, e.g., Schelling, 1978; Cooper and John, 1988; Elster,
1989, 31ff.; Elsner, 2001,2008; on the public provision of selective
incentives and high futurity). This may lead to a new interactive
public policy design, focused on specific ‘frame-setting’ to trigger the
causal factors, which in turn may allow for institutional emergence,
co-evolving with meso structure, and collective action capability
among private agents.21 This may particularly give space for the
emergence of the capability and inclination to learn and innovate,
where otherwise too much volatility and turbulence would under-
mine efforts to learn and innovate. Thus, innovative behavior may,
paradoxically, emerge through the stabilization of behaviors and
21
This also would be the logical place for a major role of cooperative game theory
that one reviewer has suggested. We have put an interactive or institutional, i.e.
frame-oriented public policy in the focus of a ‘negotiated economy’ or ‘new meritorics’
(see Elsner, 2001). This would have to be modeled using cooperative game theory.
857
expectations attained by institutionalized coordination (see, e.g.,
Boudreau et al., 2008).
The co-evolutionary process of ‘meso’-economics investigated
here is far from being fully understood and sufficiently elaborated.
Although network formation and size has already been a dimension
of many socio-economic approaches and complex models, fur-
ther strengthening relevant, applied, empirical, and policy-oriented
economic research, such as the topical research on ‘general trust’
and macro performance, may not least require further elaboration
and simulation of the logic and process of ‘meso’.
Acknowledgements
The authors are grateful to K. Dopfer and the discussants of the
2006 Schloss Wartensee Workshop on ‘Evolutionary Economics’,
St. Gallen, CH, to the participants of a 2008 AFEE session in New
Orleans and G. Hodgson, the discussant in that session, further to
D. Foley (and the students of his complex-modeling workshop at the
New School, NYC), J.B. Hall (and the discussants at a PSU department
seminar, Portland, OR), J. Sturgeon (and the discussants at a UMKC
seminar, Kansas City, MO), and U. Witt (and discussants at a Seminar
at MPI Jena) for comments on an earlier version. K. Dopfer has also
commented on the paper at the EAEPE conference in Rome, Novem-
ber 2008. Also many thanks to M. Greiff. An earlier version has
also been presented at the ASSA meetings in San Francisco, January
2009. The authors are particularly grateful to Henning Schwardt
for numerous discussions of all details of the paper. Finally, we are
grateful to an anonymous referee. However, the authors insist on
the property rights for all remaining deficiencies.
References
Alesina, A., Spolaore, E., 2003. The Size of Nations. MIT Press, Cambridge, MA.
Arthur, W.B., 1994. Inductive reasoning and bounded rationality (the El Farol prob-
lem). American Economic Review, 84–406.
Arthur, W.B., 1989. Competing technologies, increasing returns, and lock-in by his-
torical events. Economic Journal 99, 116–131.
Axelrod, R., 1984/2006. The Evolution of Cooperation. Basic Books, New York (rev.
ed. 2006).
Aydinonat, N.E., 2007. Models, conjectures and exploration: an analysis of Schelling’s
checkerboard model of residential segregation. Journal of Economic Methodol-
ogy 14 (4), 429–454.
Ayres, R.U., Martinás, K., 2005. In: Elgar, E. (Ed.), On the Reappraisal of Microeco-
nomics. Cheltenham, UK, Northampton, MA, USA.
Barro, R.J., Sala-i-Martin, X., 2003. Economic Growth. MIT Press, Cambridge, MA.
Batten, D.F., 2001. Complex landscapes of spatial interaction. The Annals of Regional
Science 35 (1), 81–111.
Binmore, K., 1998. Review of Axelrod, R., The Complexity of Cooperation. JASSS 1.1.
Boudreau, K.J., Lacetera, N., Lakhani, K.R., 2008. Parallel Search, Incentives and Prob-
lem Type: Revisiting the Competition and Innovation Link. Harvard Business
School Technology and Operations Mgt. Unit Research Paper No. 1264038.
Bush, P.D., 1987. The theory of institutional change. Journal of Economic Issues XXI,
1075–1116.
Cantner, U., Meder, A., 2008. Regional Effects of Cooperative Behavior and the Related
Variety of Regional Knowledge Bases. University of Jena, Germany, Mimeo.
Chen, P., 2008. Equilibrium illusion economic complexity and evolutionary founda-
tion in economic analysis. Evolutionary and Institutional Economics Review 5
(1), 81–127.
Christensen, J.L., Gregersen, B., Johnson, B., Lundvall, B.A., Tomlinson, M., 2008. An NSI
in transition? Denmark. In: Edquist, C., Hommen, L. (Eds.), Small Country Inno-
vation Systems. Globalization, Change and Policy in Asia and Europe. Edward
Elgar, Cheltenham, UK, Northampton, MA, USA, pp. 403–441.
Cooper, R., John, A., 1988. Coordinating coordination failures in Keynesian models.
Quarterly Journal of Economics CIII (3), 441–463.
David, P.A., 1985. Clio and the economics of QWERTY. American Economic Review
75 (2), 332–337.
Davis, J.B., 2007. Complexity theory’s network conception of the individual. In: Gia-
comin, A., Marcuzzo, M.C. (Eds.), Money and Markets. A Doctrinal Approach.
Routledge, Abingdon, UK, New York, USA, pp. 30–47.
Davis, J.B., 2008. Complex individuals: the individual in non-euclidian space. In:
Hanappi, G., Elsner, W. (Eds.), Advances in Evolutionary Institutional Economics:
Evolutionary Mechanisms, Non-Knowledge, and Strategy. Edward Elgar, Chel-
tenham, UK, Northampton, MA, USA.
Dejean, S., Penard, T., Suire, R., 2008. Olson’s Paradox Revisited: An Empirical Analysis
of Filesharing Behaviour in P2P Communities. CREM, Univ. of Rennes, F, Mimeo.
858 W. Elsner, T. Heinrich / The Journal of Socio-Economics 38 (2009) 843–858
Demange, G., Wooders, M. (Eds.), 2005. Group Formation in Economics. Networks,
Clubs, and Coalitions. Cambridge University Press, Cambridge, UK, New York.
Devezas, T.C., Corredine, J.T., 2002. The nonlinear dynamics of technoeconomic
systems: an informational interpretation. Technological Forecasting & Social
Change 69, 317–357.
Dolfsma, W., Verburg, R., 2008. Structure, agency and the role of values in processes
of institutional change. Journal of Economic Issues XLII (4), 1031–1054.
Dolfsma, W., Leydesdorff, L., 2009. Lock-in & break-out from technological trajec-
tories: modeling and policy implications, Technological Forecasting & Social
Change.
Dopfer, K. (Ed.), 2001. Evolutionary Economics: Program and Scope. Kluwer Acad.
Publ., Boston, Dordrecht, London.
Dopfer, K., 2005. Evolutionary economics: a theoretical framework, in: Id. (Ed.), The
Evolutionary Foundations of Economics, Cambridge Univ. Pr., Cambridge, pp.
3–26.
Dopfer, K., 2007. Schumpeter’s economics. The origins of a meso revolution. In:
Hanusch, H., Pyka, A. (Eds.), The Elgar Companion to Neo-Schumpeterian Eco-
nomics. Edward Elgar, Cheltenham, UK, Northampton, MA, USA.
Dopfer, K., Potts, J., 2008. The General Theory of Economic Evolution. Routledge,
London, New York (Chapter 4).
Dopfer, K., Foster, J., Potts, J., 2004. Micro–meso–macro. Journal of Evolutionary
Economics 14, 263–279.
Dosi, G., Winter, S.G., 2000. Interpreting economic change: evolution, structures and
games. LEM Working Paper Series 2000/08, Pisa, Italy.
Easterly, W., Kraay, A., 2000. Small states, small problems? Income, growth, and
volatility in small states. World Development 28 (11), 2013–2027.
Eckert, D., Koch, St., Mitloehner, J., 2005. Using the iterated prisoner’s dilemma for
explaining the evolution of cooperation in open source communities. In: Scotto,
M., Succi, G. (Eds.), Proceedings of the First International Conference on Open
Source Systems. Genova.
Edquist, C., Hommen, L. (Eds.), 2008. Small Country Innovation Systems. Global-
ization, Change and Policy in Asia and Europe. Edward Elgar, Cheltenham, UK,
Northampton, MA, USA (Chapters 1 and 12).
Elsner, W., 2000. An industrial policy agenda 2000 and beyond—experience theory
and policy. In: Elsner, W., Groenewegen, J. (Eds.), Industrial Policies After 2000.
Kluwer Acad. Publ., Boston, Dordrecht, London, pp. 411–486.
Elsner, W., 2001. Interactive economic policy: toward a cooperative policy approach
for a negotiated economy. Journal of Economic Issues XXXV (1), 61–83.
Elsner, W., 2005. Real-world economics today: the new complexity co-ordination
and policy. Review of Social Economy LXIII (1), 19–53.
Elsner, W., 2008. Market and state. In: O’Hara, P.A. (Ed.), International Encyclopedia
of Public Policy <http://pohara.homestead.com/Encyclopedia/Volume-3.pdf>.
Elster, J., 1989. The Cement of Society. A Study of Social Order. Cambridge University
Press, Cambridge.
Foley, D.K., 1998. Introduction to ‘barriers and bounds to rationality’. In: Foley, D.K.
(Ed.), Barriers and Bounds to Rationality: Essays on Economic Complexity and
Dynamics in Interactive Systems, by Albin, P.S., with an Introduction by Foley,
D.K., Princeton Univ. Pr., Princeton, N.J.
Foster, J., 2005. From simplistic to complex systems in economics. Cambridge Journal
of Economics 29, 873–892.
Fukuyama, F., 1995. Trust. The Social Virtues and the Creation of Prosperity. The Free
Press and Penguin Books, New York.
Gibbons, R., 2006. What the Folk Theorem doesn’t tell us. Industrial and Corporate
Change 15 (2), 381–386.
Goyal, S., 2005. Learning in networks. In: Demange, G., Wooders, M. (Eds.), Group For-
mation in Economics. Networks, Clubs, and Coalitions’, as given under Demange,
Wooders, pp. 122–167.
Gruene-Yanoff, T., Schweinzer, P., 2008. The roles of stories in applying game theory.
Journal of Economic Methodology 15 (2), 131–146.
Harmsen-van Hout, M.J.W., Dellaert, B.G.C., Herings, P.J.J., 2008. Behavioral Effects
in Individual Decisions of Network Formation. Maastricht Research School of
Economics of Technology and Organizations RM/08/019.
Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., 2004. Foundations
of Human Sociality. Economic Experiments and Ethnographic Evidence from
Fifteen Small-Scale Societies. Oxford University Press, Oxford.
Hodgson, G.M., 2000. From micro to macro: the concept of emergence and
the role of institutions. In: Burlamaqui, L., et al. (Eds.), Institutions and the
Role of the State. Edward Elgar, Cheltenham, UK, Northampton, MA, USA,
pp. 103–126.
Hodgson, G.M., 2006. What are institutions? Journal of Economic Issues XL (1), 1–25.
Holm, J.R., Lorenz, E., Lundvall, B.A., Valeyre, A., 2008. Work organisation and systems
of labour market regulation in Europe. In: Paper Presented at the 20th EAEPE
Annual Conference, Rome (Mimeo).
Jennings, F.B., 2005. How efficiency/equity tradeoffs resolve through horizon effects.
Journal of Economic Issues 39 (2), 365–373.
Jorgensen, H., 2002. Consensus Cooperation and Conflict. The Policy Making Process
in Denmark. Edward Elgar, Cheltenham, UK, Northampton, MA, USA.
Keele, L.J., Stimson, J.A., 2002. Measurement issues in the analysis of social capital.
World Bank Working Paper.
Kesting, St., Nielsen, K., 2008. Varieties of capitalism—theoretical critique and empir-
ical observations. In: Elsner, W., Hanappi, G. (Eds.), Varieties of Capitalism and
New Institutional Deals: Regulation, Welfare, and the New Economy. Edward
Elgar, Cheltenham, UK, Northampton, MA, USA.
Knack, S., Keefer, P., 1997. Does social capital have an economic payoff? A cross-
country investigation. Quarterly Journal of Economics 112 (4), 1251–1288.
Knudsen, T., 2002. The evolution of cooperation in structured populations. Consti-
tutional Political Economy 13, 129–148.
Kuznets, S., 1960. Economic growth of small nations. In: Robinson, E.A.G. (Ed.), Eco-
nomic Consequences of the Size of Nations. Macmillan, London, pp. 14–32 (repr.
1963).
Leydesdorff, L., 2007. The communication of meaning in anticipatory systems: a
simulation study of the dynamics of intentionality in social interactions. In:
Vice-Presidential Address at the 8th International Conference of Computing
Anticipatory Systems, Liège, Belgium. Mimeo.
Lindgren, K., Nordahl, M.G., 1994. Evolutionary dynamics of spatial games. Physica
D 75, 292–309.
Lundvall, B.A., 2002. Innovation Growth and Social Cohesion. The Danish Model.
Edward Elgar, Cheltenham, UK, Northampton, MA, USA.
Marwell, G., Oliver, P., 1993. The Critical Mass in Collective Action. A Micro-Social
Theory. Cambridge University Press, Cambridge.
Oestreicher-Singer, G., Sundararajan, A., 2008. The Visible Hand of Social Networks
in Electronic Markets. New York University, Mimeo, New York.
O’Hara, P.A., 2008. Uneven development, global inequality and ecological sustain-
ability: recent trends and patterns. GPERU Working Paper, Curtin Univ., Perth,
Australia.
Olson, M., 1965. The Logic of Collective Action. Harvard University Press, Cambridge,
MA.
Ostrom, E., 2007. Challenges and growth: the development of the interdisciplinary
field of institutional analysis. Journal of Institutional Economics 3 (3),239–264.
Page, F., Wooders, M., 2007. Strategic basins of attraction, the path dominance
core, and network formation games. Center for Applied Economics and Policy
Research, Working Paper 2007-020.
Pyka, A., 1999. Der kollektive Innovationsprozess. Eine theoretische Analyse
informeller Netzwerke und absorptiver Fähigkeiten. Duncker&Humblot, Berlin.
Schelling, T.C., 1969. Models of segregation. American Economic Review 59 (2),
488–493.
Schelling, T.C., 1973. Hockey helmets, concealed weapons, and daylight saving: a
study of binary choices with externalities. The Journal of Conflict Resolution
XVII.3 (repr. in: Schelling, 1978, pp. 211–243).
Schelling, T.C., 1978. Micromotives and Macrobehavior. W.W. Norton, New York,
London.
Schotter, A., 1981. The Economic Theory of Social Institutions. Cambridge University
Press, Cambridge, UK.
Spiekermann, K.P., 2009. Sort out your neighbourhood. Public good games on
dynamic networks. Synthese 168, 273–294 (Knowledge, Rationality & Action,
583-604).
Traulsen, A., Nowak, M.A., 2006. Evolution of cooperation by multilevel selection.
PNAS 103 (29), 10952–10955.
Tucker, C., 2008. Social interactions, network fluidity and network effects. NET Insti-
tute Working Paper No. 08-30.
van Staveren, I., 2001. The Values of Economics. An Aristotelian Perspective. Rout-
ledge, London, New York.
Vinkovic, D., Kirman, A., 2006. A physical analogue of the Schelling model. PNAS 103
(51), 19261–19265.
Watts, D.J., 1999. Small Worlds. The Dynamics of Networks between Order and
Randomness. Princeton University Press, Princeton, NJ, Oxford, UK (Chapter 8).
Yamagishi, T., 1992. Group size and the provision of a sanctioning system in a social
dilemma. In: Liebrand, W.B.G., Messick, D.M., Wilke, H.A.M. (Eds.), Social Dilem-
mas. Theoretical Issues and Research Findings. Pergamon Press, Oxford, pp.
267–287.