. 8
( 13)





Figure 7.2
The evolution of cooperation is strongly affected by rate of mixing between groups. Part
(a) plots the long run average frequency of cooperation (i.e., the sum of the frequencies of
contributors and punishers) as a function of group size when there is no punishment
( p ¼ k ¼ 0) for three mixing rates. Group selection is ineffective unless groups are quite
small. Part (b) shows that when there is punishment ( p ¼ 0:8, k ¼ 0:2), group selection
can maintain cooperation in larger groups for all rates of mixing. However, at higher
rates of mixing, cooperation does not persist in the largest groups.
222 Boyd, Gintis, Bowles, and Richerson

Figure 7.3
The evolution of cooperation is sensitive to the cost of being punished ( p). Here we plot
the long run average frequency of cooperation with the base case cost of being punished
( p ¼ 0:8) and with a lower value of p. Lower values of p result in much lower levels of

The long run average amount of cooperation is also sensitive to the
cost of being punished (¬gure 7.3). When the cost of being punished is
at base case value °p ¼ 4cÞ, even a modest frequency of punishers will
cause defectors to be selected against, and, as a result, there is a sub-
stantial correlation between the frequency of cooperation and punish-
ment across groups. When the cost of being punished is the same as
the cost of cooperation ° p ¼ cÞ punishment does not suf¬ciently reduce
the relative payoff of defectors, and the correlation between the fre-
quency of cooperators and punishers declines. Lower correlations
mean that selection among groups cannot compensate for the decline
of punishers within groups, and eventually both punishers and con-
tributors decline.
It is important to see that punishment leads to increased cooperation
only to the extent that the costs associated with being an altruistic
punisher decline as defectors become rare. Monitoring costs, for exam-
ple, must be paid whether or not there are any defectors in the group.
When such costs are substantial”or when the probability of mistaken
defection is high enough that altruistic punishers bear signi¬cant costs
even when defectors are rare”group selection does not lead to the
evolution of altruistic punishment (¬gure 7.4). However, because peo-
ple live in long-lasting social groups, and language allows the spread
The Evolution of Altruistic Punishment 223

Figure 7.4
Punishment does not aid in the evolution of cooperation when the costs borne by punish-
ers are ¬xed, independent of the number of defectors in the group. Here we plot the long
run average frequency of cooperation when the costs of punishing are proportional to the
frequency of defectors (variable cost), ¬xed at a constant cost equal to the cost of cooper-
ating (c), and when there is no punishment.

of information about who did what, it is plausible that monitoring
costs may often be small compared to enforcement costs. This result
also leads to an empirical prediction: People should be less inclined to
pay ¬xed rather than variable punishment costs if the mechanism out-
lined here is responsible for the psychology of altruistic punishment.
The effectiveness of group selection is especially sensitive to the rate
of mutation when there is punishment. For example, decreasing the
mutation rate from 0.05c to 0.005c leads to very high levels of coopera-
tion even when groups include 256 individuals. Random drift-like
processes have an important effect on trait frequencies in this model.
Standard models of genetic drift suggest that lower mutation rates will
cause groups to stay nearer the boundaries of the state space (Crow
and Kimura 1970), and our simulations con¬rm this prediction (¬gure
7.5). When the mutation rate is low, there are very few groups in
which defectors are common; most of the groups lie very close to the
contributor-punisher boundary. In contrast, when the mutation rate is
higher, groups with a wide range of defector frequencies are present.
Thus, an increasing mutation rate, on average, increases the amount
of punishment that must be administered and therefore increases the
224 Boyd, Gintis, Bowles, and Richerson


Figure 7.5
Decreasing the mutation rate reduces the number of groups in which defectors are com-
mon. Each point represents the frequencies of each of the three strategies in 1 of 128
groups during a single, representative time period (t ¼ 1500) from the interval in time
(t ¼ 1000“2000) over which we calculated the average steady state frequencies. There are
not 128 points because many groups have the same frequencies. In (A) m ¼ 0:01 while (B),
m ¼ 0:001, n ¼ 64, and other parameters are as in the base case. If punishment and group
selection are eliminated ( p ¼ k ¼ e ¼ 0), these mutation rates maintain ˜˜cooperation™™
(now just an individually disadvantageous trait) approximately at frequencies 0.1 and
0.01 respectively. When defectors are less common, punishing is less costly, and therefore
group selection is more effective at maintaining punishment at high frequency. Note,
however, even when there are many groups in which defectors are common as in (A)
group selection can still maintain punishment and therefore sustain cooperation in fairly
sizable groups.
The Evolution of Altruistic Punishment 225

payoff advantage of second order free-riders compared to altruistic
Additional sensitivity analyses suggest that these results are robust.
In addition to the results described in the last several paragraphs, we
have studied the sensitivity of the model to variations in the remaining
parameter values. Increasing e, the error rate, reduces the steady state
amount of cooperation. Reducing N adds random noise to the results.
We also tested the sensitivity of the model to three structural
changes. We modi¬ed the payoffs so that each cooperative act produ-
ces a per capita bene¬t of b=n for each other group member and also
modi¬ed the extinction model so that the probability of group extinc-
tion is proportional to the difference between warring groups in aver-
age payoffs including the costs of punishment (rather than simply the
difference in frequency of cooperators). The dynamics of this model
are more complicated because group selection now acts against altruis-
tic punishers as punishment reduces mean group payoffs. However,
the correlated effect of group selection on cooperation still tends to in-
crease punishment as in the original model. The relative magnitude of
these two effects depends on the magnitude of the per capita bene¬t to
group members of each cooperative act, b=n. For reasonable values of
b, (2c; 4c, and 8c), the results of this model are qualitatively similar to
those shown above.
We also investigated a model in which the amount of cooperation
and punishment vary continuously. An individual with cooperation
value x behaves like a cooperator with probability x and a defector
with probability 1 À x. Similarly, an individual with a punishment
value y behaves like an altruistic punisher with probability y and a
nonpunisher with probability 1 À y. New mutants are uniformly dis-
tributed. The steady state mean levels of cooperation in this model are
similar to the base model.
Finally, we studied a model without extinction analogous to a re-
cent model of selection among stable equilibria due to biased imitation
(Boyd and Richerson 2002). In this model, populations are arranged in
a ring, and individuals imitate only other individuals drawn from the
neighboring two groups. Cooperative acts produce a per capita bene¬t
b=n so that groups with more cooperators have higher average payoff,
and thus cooperation will, all other things being equal, tend to spread
because individuals are prone to imitate successful neighbors. We
could ¬nd no reasonable parameter combination that led to signi¬cant
steady state levels of cooperation in this last model.
226 Boyd, Gintis, Bowles, and Richerson

7.4 Discussion

We have shown that while the logic underlying altruistic cooperation
and altruistic punishment is similar, their evolutionary dynamics are
not. In the absence of punishment, within-group adaptation acts to de-
crease the frequency of altruistic cooperation, and as a consequence,
weak drift-like forces are insuf¬cient to maintain substantial variation
between groups. In groups where altruistic punishers are common,
defectors are excluded, and this maintains variation in the amount
of cooperation between groups. Moreover, in such groups, punishers
bear few costs, and altruistic punishers decrease only very slowly in
competition with contributors. As a result, group selection is more ef-
fective at maintaining altruistic punishment than maintaining altruistic
These results suggest that group selection can play an important role
in human cultural evolution, because rapid cultural adaptation pre-
serves cultural variation among groups. The importance of group se-
lection is always a quantitative issue. There is no doubt that selection
among groups favors individually costly, group-bene¬cial behaviors.
The question is always: Does group selection play an important role
under plausible conditions? Our results suggest that group selection
acting on genetic variation will not be important even when punish-
ment is possible, because natural selection will rarely be strong enough
to overcome homogenizing effects of migration between groups, and,
as a result, there will be insuf¬cient genetic variation among groups.
In contrast, rates of cultural adaptation are often greater than rates of
mixing”as is re¬‚ected by the parameter values used in our simula-
tions. With these parameter values, cooperation is sustained in groups
on the order of 100 individuals. If the ˜˜individuals™™ in the model rep-
resent family groups (on the grounds that they migrate together and
adopt common practices), altruistic punishment could be sustained in
groups of 600 people”a size much larger than typical foraging bands
and approximately the size of many ethno-linguistic units in nonagri-
cultural societies.


Aoki, Kenichi. ˜˜A Condition for Group Selection to Prevail over Counteracting Individ-
ual Selection,™™ Evolution 36 (1982): 832“842.
Axelrod, Robert, and William D. Hamilton. ˜˜The Evolution of Cooperation,™™ Science 211
(1981): 1390“1396.
The Evolution of Altruistic Punishment 227

Boehm, Christopher. ˜˜Egalitarian Behavior and Reverse Dominance Hierarchy,™™ Current
Anthropology 34, 3 ( June 1993): 227“254.
Bowles, Samuel. ˜˜Individual Interactions, Group Con¬‚icts, and the Evolution of Prefer-
ences,™™ in Steven N. Durlauf and H. Peyton Young (eds.), Social Dynamics. Cambridge,
MA: MIT Press, 2001, pp. 155“190.
Boyd, Robert, and Peter J. Richerson. ˜˜The Evolution of Cooperation,™™ Journal of Theoreti-
cal Biology 132 (1988): 337“356.
””” and Peter J. Richerson. ˜˜Group Bene¬cial Norms Can Spread Rapidly in a Struc-
tured Population,™™ Journal of Theoretical Biology 215 (2002): 287“296.
Clutton-Brock, T. H., and G. A. Parker. ˜˜Punishment in Animal Societies,™™ Nature 373
(1995): 58“60.
Crow, James F., and Motoo Kimura. An Introduction to Population Genetic Theory. New
York: Harper & Row, 1970.

Eshel, Ilan. ˜˜On the Neighbor Effect and the Evolution of Altruistic Traits,™™ Theoretical
Population Biology 3 (1972): 258“277.

Gintis, Herbert. ˜˜The Hitchhiker™s Guide to Altruism: Genes, Culture, and the Internal-
ization of Norms,™™ Journal of Theoretical Biology 220, 4 (2003): 407“418.
Hauert, Christoph, Silvia DeMonte, Josef Hofbauer, and Karl Sigmund. ˜˜Volunteering as
Red Queen Mechanism for Cooperation in Public Goods Game,™™ Science 296 (May 2002):
Henrich, Joseph. ˜˜Cultural Transmission and the Diffusion of Innovations,™™ American An-
thropologist 103 (2001): 992“1013.
””” and Robert Boyd. ˜˜Why People Punish Defectors: Weak Conformist Transmission
Can Stabilize Costly Enforcement of Norms in Cooperative Dilemmas,™™ Journal of Theoret-
ical Biology 208 (2001): 79“89.

Rogers, Alan R. ˜˜Group Selection by Selective Emigration: The Effects of Migration and
Kin Structure,™™ American Naturalist 135, 3 (March 1990): 398“413.

Rogers, E. M. Diffusion of Innovations. New York: Free Press, 1983.
Sethi, Rajiv, and E. Somanathan. ˜˜The Evolution of Social Norms in Common Property
Resource Use,™™ American Economic Review 86, 4 (September 1996): 766“788.

Sober, Elliot, and David Sloan Wilson. Unto Others: The Evolution and Psychology of Unself-
ish Behavior. Cambridge, MA: Harvard University Press, 1998.
Soltis, Joseph, Robert Boyd, and Peter Richerson. ˜˜Can Group-functional Behaviors
Evolve by Cultural Group Selection: An Empirical Test,™™ Current Anthropology 36, 3
( June 1995): 473“483.
Trivers, R. L. ˜˜The Evolution of Reciprocal Altruism,™™ Quarterly Review of Biology 46
(1971): 35“57.
8 Norm Compliance and
Strong Reciprocity

Rajiv Sethi and E. Somanathan

8.1 Introduction

A central feature of strong reciprocity is the propensity to punish
others for opportunistic actions and to reward them for acts of uncom-
mon generosity, where such rewards and punishments are not moti-
vated by the prospect of future gain. The social norms that serve as the
benchmark for evaluating behavior may vary from one culture to an-
other, but given some such set of broadly shared norms, strong reci-
procity provides a decentralized mechanism for their enforcement. The
extent and persistence of strong reciprocity poses something of a theo-
retical puzzle because monitoring and sanctioning activities, while po-
tentially bene¬cial to the group, place a net material burden on the
reciprocator. Since opportunistic individuals choose to comply with
or violate norms based on the likelihood and severity of sanctioning
they anticipate, such individuals will always outperform reciprocators
within any group. Even under complete compliance, reciprocators in-
cur costs that opportunists are able to avoid. One would expect this
payoff differential to exert evolutionary pressure on the population
composition until reciprocators are entirely displaced from the group.
This suggests that any population composed of immutable groups
with no intergroup mobility will not sustain strong reciprocity in the
long run.
The situation can be quite different if groups can dissolve and new
groups are formed periodically. Strong reciprocity differs from pure
altruism in one important respect: The presence of reciprocators in a
group can, under very general conditions, alter the behavior of oppor-
tunists in such a manner as to bene¬t all members of the group
(including reciprocators).1 This creates the possibility that in groups
230 Sethi and Somanathan

containing reciprocators, all group members including reciprocators
obtain greater payoffs than are obtained in homogeneous groups of
self-regarding individuals. We argue below that under these circum-
stances, reciprocators can invade a population of opportunists when
groups are dissolving and new groups are forming according to a pro-
cess of purely random (non-assortative) matching. Furthermore, we
show that even when these conditions are not satis¬ed (so that an op-
portunistic population is stable), there may exist additional stable pop-
ulation states in which reciprocators are present.
The conditions under which strong reciprocity can survive and
spread in evolutionary competition with opportunism within the con-
text of a common pool resource environment are explored in this chap-
ter. Such environments consist of economically valuable resources to
which multiple unrelated users have access. Common pool resources
have been the dominant form of property through all of human prehis-
tory and history until the advent of agriculture and remain economi-
cally signi¬cant to this day. Coastal ¬sheries, grazing lands, forests,
groundwater basins, and irrigation systems are all examples of re-
sources that have traditionally been held as common property. A well-
known problem that arises in the management of such resources is that
when all appropriators independently attempt to maximize their own
private gains from resource extraction, the result is a ˜˜tragedy of the
commons™™”with overextraction resulting in excessive resource deple-
tion. The tragedy is that all appropriators may end up with smaller
net gains than would be obtained under a system of resource man-
agement in which restraints on extraction were enforced. In the ab-
sence of a government, such enforcement can only come from the
appropriators themselves through a decentralized system of monitor-
ing and enforcement.
Strong reciprocity can motivate individuals to undertake such moni-
toring and enforcement. Field studies of local commons, of which there
are several thousand, show that in many cases resource extraction is
regulated and restrained by a complex network of social norms held in
place by credible threats of sanction.2 Such systems coerce ordinarily
self-interested individuals to behave in ways that re¬‚ect prosocial con-
cerns. Overextraction is therefore limited, and it is possible for all indi-
viduals (including reciprocators) to obtain higher material rewards
than the tragedy of the commons model would predict. In this chapter,
we argue that this effect helps us understand not only how local com-
mons have been able to survive conditions of extreme scarcity, but
Norm Compliance and Strong Reciprocity 231

also how strong reciprocity itself has been able to survive under evolu-
tionary pressure.
The evolutionary theory of strong reciprocity advanced in sections
8.2 and 8.3 relies on the ability of reciprocators to make a credible com-
mitment to monitor and sanction norm violators even when it is not in
their interest to do so. Alternative evolutionary accounts of strong reci-
procity that differ in signi¬cant ways from this one have been pro-
posed, and these are reviewed in section 8.4. Aside from the power of
commitment, two additional themes”which we identify as assortation
and parochialism”appear repeatedly in this literature. Our survey of
this sometimes technical and specialized literature is neither exhaus-
tive nor mathematical and should be accessible to a broad range of re-
searchers across disciplinary boundaries.

8.2 Common Property

The following simple model of common pool resource extraction pro-
vides an analytical framework within which the question of prefer-
ence evolution can be explored.3 Consider a group of individuals with
shared access to a resource that is valuable but costly to appropriate.
Each appropriator makes an independent choice regarding her level of
resource extraction. The aggregate amount of resource extraction is
simply the sum of all individual extraction levels. The total cost of ex-
traction incurred by the group as a whole rises with aggregate extrac-
tion in accordance with the following hypothesis: The higher the level
of aggregate extraction, the more it costs to extract an additional unit of
the resource. The share of the total cost of extraction that is paid by any
given appropriator is equal to the share of that appropriator™s extrac-
tion in the total extraction by the group. These are standard assump-
tions in the analysis of common pool resource environments and
imply that an increase in extraction by one appropriator raises the cost
of extraction for all appropriators.
Figure 8.1 depicts the manner in which aggregate bene¬ts and costs
vary with the level of aggregate extraction. The straight line corre-
sponds to aggregate extraction and the curve to the aggregate costs of
extraction. The costs rise gradually at ¬rst and then rapidly, so that
there is a unique level of aggregate extraction X at which net bene¬ts
are maximized. If each appropriator were to extract an equal share of
this amount, the resulting outcome would be optimal from the per-
spective of the group. However, if all appropriators were to chose this
232 Sethi and Somanathan

Aggregate Benefits and Costs


Aggregate Resource Extraction

Figure 8.1
Aggregate costs and bene¬ts of extraction.

level of extraction, self-interested individuals would prefer to extract
more, since this would increase their own private payoffs. The fact
that this increase would come at the cost of lowering the combined
payoff to the group as a whole would not deter a self-interested appro-
priator. If all appropriators were self-interested, and made indepen-
dent choices regarding their extraction levels, the resulting level of
aggregate extraction would not be optimal from the perspective of the
group. It is possible to show that in an equilibrium of the game played
by a group of self-interested appropriators, each appropriator would
choose the same extraction level and that the resulting aggregate ex-
traction X e would exceed X (as shown in ¬gure 8.1). The level of
extraction under decentralized, self-interested choice is inef¬cient. Each
member of the group could obtain higher payoffs if all were forced to
limit their extraction. This is the tragedy of the commons, in which the
optimal pursuit of one™s own interest by each appropriator leads to
lower payoffs for all than could be realized under a system of ˜˜mutual
coercion, mutually agreed upon™™ (Hardin 1968).
Coordinated mutual coercion, however, requires a central authority
capable of imposing sanctions on violators. Can groups avoid the trag-
Norm Compliance and Strong Reciprocity 233

edy of the commons even in the absence of centralized enforcement?
Consider the possibility that individuals may monitor each other (at a
cost) and impose decentralized sanctions on those who choose extrac-
tion levels that are above some threshold. Speci¬cally, suppose that
individuals are of two types, whom we call reciprocators and oppor-
tunists. Reciprocators comply with and enforce a norm that prescribes,
for each individual, an equal share of the ef¬cient extraction level X Ã .4
Reciprocators monitor others at a cost and are able to detect and sanc-
tion all violators. Violators incur a cost as a result of each sanction.
Opportunists simply choose extraction levels that maximize their pri-
vate net bene¬ts from extraction. In doing so, they face a choice be-
tween norm compliance, which allows them to escape punishment,
and norm violation, which enables them to choose optimal extraction
levels. Which of these two options is more pro¬table for a given oppor-
tunist depends on the population composition of the community and
the choices made by other opportunists.
Consider a group in which both reciprocators and opportunists are
present. The opportunists are involved in a strategic interaction in
which each must determine her level of extraction. In equilibrium,
opportunists fall into one of two groups: those who violate the norm
and incur the cost of being punished and those who comply with the
norm and escape punishment. It can be shown that this game has a
unique equilibrium in which all opportunists who violate the norm
will choose the same extraction level. For reasons discussed earlier in
this chapter the extraction level of violators will exceed that of those
individuals (some of whom may be opportunists) who are in compli-
ance with the norm.
The equilibrium number of violators will depend, among other
things, on the severity of the sanction that reciprocators impose, and it
can be shown that the equilibrium number of violators is nonincreas-
ing in the severity of the sanction. This is illustrated in ¬gure 8.2 for a
particular speci¬cation of the model (with one reciprocator in a group
of thirty individuals). The relationship between the number of viola-
tors and the severity of the sanction is nonlinear. A relatively small
sanction can achieve some compliance, and the extent of compliance
rises rapidly with the severity of the sanction at ¬rst. However, achiev-
ing complete or almost complete compliance requires very substantial
increases in sanction severity. The reason is because increased compli-
ance by others reduces the incremental cost of extraction and therefore
raises the incentives to violate the norm. To counteract this phenome-
non, the penalty from violation must rise commensurately.
234 Sethi and Somanathan


Number of Violators





0 1 2 3 4 5 6
Severity of Sanction

Figure 8.2
Severity of sanctions and the incidence of compliance.

Not all opportunists need to receive the same payoff, since the pay-
offs from compliance and norm violation are not necessarily equal in
equilibrium. All opportunists earn more than reciprocators, however.
This is the case because compliance is an option that opportunists may
choose to exercise, and since they do not engage in monitoring or
enforcement, compliance always yields opportunists a greater payoff
than reciprocators can ever attain. Hence if opportunists choose to vio-
late the norm, they do so in the expectation that this will be at least as
pro¬table as compliance, and hence strictly more pro¬table than the
behavior of reciprocators. This raises the question of how reciprocators
can survive under evolutionary pressure.

8.3 Evolution

Suppose that groups are formed by randomly sampling individuals
from a large global population, a certain proportion of which are recip-
rocators. Groups formed in this manner will show some variation in
composition as a direct result of randomness in the sampling process.
If the global population share of reciprocators is close to zero, there
Norm Compliance and Strong Reciprocity 235

will be a very high probability that most communities consist entirely
of opportunists, and most reciprocators will ¬nd themselves in com-
munities in which no other reciprocators are present. Similarly, if the
global population share of reciprocators is close to one, most groups
will consist exclusively of reciprocators and most opportunists will
¬nd themselves in groups without other opportunists. For intermedi-
ate values of the global population composition, there will be greater
variety across groups and most groups will consist of a mixture of
reciprocators and opportunists.
The average payoff obtained by opportunists in any given group is
fully determined by the composition of the group. Hence, the average
payoff to opportunists in the population as a whole is obtained by tak-
ing a weighted average of opportunist payoffs, with the weight ap-
plied to each type of group proportional to the probability with which
this type of group will form. The same procedure applied to reciproca-
tor payoffs yields the average payoff to reciprocators in the population
as a whole. When these average payoffs differ, the population compo-
sition itself will change. We assume that the dynamics of the popula-
tion composition are such that the type with the higher payoff grows
relative to the type with the lower payoff (a special case of this is the
replicator dynamics). We are interested in identifying stable rest points
in this dynamic process with a view toward identifying whether or not
reciprocators can be present at such states.
Consider ¬rst a population consisting only of opportunists. Can
reciprocators invade such a population under evolutionary dynamics?
Note that when the global reciprocator share is small, almost all recip-
rocators ¬nd themselves in groups with exactly one reciprocator, while
almost all opportunists ¬nd themselves in groups with no reciproca-
tors. In groups of the former type, reciprocators necessarily obtain
lower payoffs than do opportunists (regardless of the extent of compli-
ance). However, this does not imply that a population of opportunists
must be stable. Such a population will be unstable as long as reciproca-
tors obtain greater payoffs in groups consisting of a single reciprocator
than do opportunists in groups consisting of no reciprocators. This is
clearly possible only if the presence of a single reciprocator induces
at least some opportunists to choose compliance, and this in turn
depends on the severity of the sanction.
It can be shown that if the severity of the sanction falls below some
threshold (which depends on group size and the cost parameter), then
an opportunist population is necessarily stable. On the other hand, if
236 Sethi and Somanathan


Cost of Monitoring and Enforcement







0 1 2 3 4 5 6
Severity of Sanction

Figure 8.3
Conditions for the instability of an opportunist population.

the severity of the sanction exceeds this threshold, then an opportunist
population will be invadable if the cost to reciprocators of imposing
sanctions is suf¬ciently small. In particular, raising the severity of sanc-
tions increases or leaves unchanged the range of costs that are consis-
tent with the instability of the opportunist population. However, there
is a boundary that the enforcement cost cannot exceed if an opportun-
ist population is to be invadable, no matter how great the severity of
sanctions happens to be. Figure 8.3 illustrates this phenomenon for a
particular speci¬cation of the model.
While an opportunist population may or may not be stable, a popu-
lation consisting of reciprocators alone is unstable for all parameter
values. As the global reciprocator population share approaches one,
reciprocators almost certainly ¬nd themselves in homogeneous groups
in which each person complies with the norm and pays the cost of
monitoring, while opportunists almost certainly ¬nd themselves in
groups in which they are the only opportunist. Since they have the op-
tion of complying with the norm and escaping both the monitoring
cost and the sanction, they can guarantee for themselves a payoff
strictly greater than that which reciprocators get in all-reciprocator
Norm Compliance and Strong Reciprocity 237

Expected Payoffs to Opportunists and Reciprocators







0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Global Population Share of Reciprocators

Figure 8.4
Multiple stable steady states.

groups. Since this is feasible, their optimal choice must yield them
at least this amount. Opportunists therefore have a greater expected
payoff than reciprocators when they are suf¬ciently rare in the global
For those parameter values that render an opportunist population
unstable, the only stable states will be polymorphic (that is, they will
consist of a mixture of the two types). Polymorphic states can also arise
when an opportunist population is stable, and it is not dif¬cult to ¬nd
parameter ranges consistent with two or even three stable states. Fig-
ure 8.4 shows how the average payoffs obtained by opportunists and
reciprocators vary with the population share of the latter in the case of
one such example. Aside from the stable state in which only opportun-
ists are present, there is a second stable state in which about 40 percent
of the population is composed of reciprocators. In fact, it is easy to ¬nd
speci¬cations in which three stable states exist”one of which consists
almost exclusively of reciprocators.
238 Sethi and Somanathan

The reason why a mixture of reciprocators and opportunists can be
stable”even when a population consisting only of opportunists is it-
self stable”is subtle. If the severity of sanctions is insuf¬ciently great,
a single reciprocator in a group of opportunists will induce little or no
compliance, and opportunists will outperform reciprocators when the
population share of the latter is small. However, when the population
share of reciprocators is not too small, most groups in which reciproca-
tors ¬nd themselves will also contain other reciprocators, and in such
groups there may be signi¬cant compliance. Opportunists will do
even better than reciprocators in any such group, but even with ran-
dom group formation, the probability with which an opportunist ¬nds
herself in a group with signi¬cant compliance will be somewhat lower
than the probability with which reciprocators ¬nd themselves in such
groups. This effect can outweigh the effect of greater opportunist pay-
offs in each group and permit a mixed population to be stable.
This evolutionary theory of reciprocity is based on the power of com-
mitment. Reciprocators are able to in¬‚uence the behavior of oppor-
tunists in their group because they can credibly commit to punishing
them if they violate the norm of limited resource extraction. Their com-
mitment to do so is credible because they are strong reciprocators who
prefer to punish violators even at some material cost to themselves. As
a result, the disadvantage faced by reciprocators within their group
can be outweighed by the fact that groups in which they are present
can be signi¬cantly more successful than those in which they are ab-
sent. In the next section, we review other approaches to the evolution
of strong reciprocity that do not rely on commitment but rather on
assortative interaction or parochialism.5

8.4 Assortation, Parochialism, and Identi¬ability

The preceding analysis was based on the hypothesis of random (non-
assortative) group formation. If, instead, group formation is suf¬-
ciently assortative, stable norm compliance can occur even in the
absence of a sanctioning mechanism. To take an extreme case, suppose
that there were perfect assortation so that all groups were homoge-
neous. In this case, each opportunist would be in a group in which all
appropriators extract opportunistically, while each reciprocator would
be in a group in which all appropriators extract ef¬ciently. Recipro-
cators would obtain greater net bene¬ts and opportunists would be
displaced under evolutionary selection. It is easily seen that the same
Norm Compliance and Strong Reciprocity 239

outcome arises if there is a suf¬ciently high degree of assortative
How might assortative interaction among unrelated individuals
arise? One possibility is that group formation results from a process of
conscious choice in which reciprocators seek out those of their own
type. Even if individuals of all types prefer to be in groups consisting
largely of reciprocators, this will result in assortative interaction as
long as reciprocators avoid interaction with opportunists. Endoge-
nous group formation along these lines requires some degree of type
identi¬ability”for instance, through a signal by which reciprocators
can be identi¬ed. When the signal is informative but imperfect, some
opportunists will appear to be reciprocators and vice versa. The result-
ing sorting process leads to partial assortation: Reciprocators are more
likely to be matched with other reciprocators than with opportunists.
Opportunists who happen to be matched with reciprocators do ex-
tremely well because they violate the norm while others in their group
are in compliance. However, as long as the degree of assortation is suf-
¬ciently great, this advantage can be swamped by the disadvantage
that opportunists face in being more likely to be matched with other
opportunists. If, in addition, the process of sorting on the basis of sig-
nal observation is costly, then the long-run population will consist of a
mixture of types. The intuition for this is that when most members of
the population are reciprocators, then investment in sorting not worth-
while and individuals forego the opportunity to seek out reciprocators
and avoid opportunists. This allows the share of opportunists to grow
until a point is reached when reciprocators ¬nd investment in sorting
to be worthwhile.7
A less direct route to assortative interaction occurs when individuals
may be ostracized from groups for noncompliance with social norms. In
this case, opportunists must take into account not simply the direct
payoff consequences of norm compliance and violation, but also the
payoff implications of possible detection and expulsion. Since oppor-
tunists violate norms with greater frequency than do reciprocators,
they will be expelled with greater likelihood. The result is assortative
interaction: Reciprocators are more likely than opportunists to be in a
group with a large proportion of reciprocators. This compensates for
the losses incurred by costly sanctioning of noncooperative behavior,
and both types can coexist in the long run.8
Even in the absence of assortative interaction, reciprocity can survive
if individuals condition their behavior on the distribution of types in
240 Sethi and Somanathan

their group. We refer to this dependence of actions on the group com-
position as parochialism. The basic idea can be illustrated by consider-
ing the extreme case in which reciprocators comply with the norm and
engage in monitoring and enforcement only if they are present in suf¬-
ciently large numbers to ensure complete compliance on the part of
opportunists. In this case, the behavior (and hence the payoffs) of
opportunists and reciprocators are identical in groups containing an
insuf¬cient number of reciprocators. The remaining groups achieve
norm compliance and signi¬cantly higher payoffs, although opportun-
ists in such groups escape the cost of monitoring and hence have
a payoff advantage over reciprocators. If the cost of monitoring is
suf¬ciently small, this advantage to opportunists will be outweighed
by the fact that reciprocators are more likely to ¬nd themselves in
groups that achieve norm compliance and ef¬ciency, even under non-
assortative group formation. In this case, reciprocators will survive
and spread in a population consisting largely of opportunists, just as
they would under assortative interaction. Suppose further that the
monitoring costs incurred by reciprocators in groups in which they
predominate decrease their payoffs below those of opportunists in
these groups. If opportunists are rare, most groups containing both
types will be of this kind, and opportunists will therefore invade a
population of reciprocators. In this case, the model predicts the evolu-
tion of a mixed population.9
The preceding discussion has been based on the assumption that
individuals know the composition of their group, as would be the case
if reciprocators and opportunists could be distinguished by some ob-
servable trait. In this situation, an opportunist who carried the trait
identifying reciprocators would outperform identi¬able opportunists
and would gradually displace the latter in the population. As this
happened, however, it would generate selection pressure favoring
reciprocators who could distinguish themselves from the disguised
opportunists. Reciprocators who evolved a signal that achieved this
objective would reap the gains from ef¬cient norm compliance in the
presence of their own type. Hence, rather than assuming that recipro-
cators and opportunists are either perfectly distinguishable or perfectly
indistinguishable, it is more realistic to assume that they are neither.
As in the earlier discussion of assortative interaction, this assump-
tion can be made by supposing that prior to choosing actions, each in-
dividual emits a signal with some ¬xed probability that depends on
the individual™s type. Speci¬cally, suppose that reciprocators are more
Norm Compliance and Strong Reciprocity 241

likely to emit the signal than are opportunists. After the signaling
phase, each member of the group updates her assessment of the proba-
bility distribution describing the composition of the group. When the
global population consists almost exclusively of opportunists, it is ex-
tremely likely that even a person who emits the reciprocator signal is
an opportunist. This follows from the fact that the fraction of the popu-
lation who are opportunists with reciprocator signals will be much
larger than the fraction of the population who are reciprocators with
reciprocator signals. The signal then conveys almost no information,
and opportunists will not be deterred from overextraction by the pros-
pect of punishment even when they are matched with a person with a
signal. Recognizing this, reciprocators will behave exactly like oppor-
tunists when the opportunist population share is large. Thus, both
types ignore the signal, choose the same inef¬cient extraction level,
and get the same payoff.
Now consider the other extreme case of a population consisting
almost exclusively of reciprocators. Again, the reciprocator signal will
convey virtually no information since it is extremely likely, regardless
of whether or not the signal is observed, that each player in one™s
group is a reciprocator. In this situation, if reciprocators were to en-
gage in monitoring, then opportunists would always comply with the
norm and get higher payoffs than reciprocators by escaping the moni-
toring cost. If, on the other hand, reciprocators did not monitor, then
opportunists would extract more than the norm, thus getting higher
payoffs than reciprocators. In either case, we see that opportunists will
always be able to invade a reciprocator population. (Reciprocators will
always comply with the norm since they expect with near-certainty
that the other group members are fellow reciprocators.)
However, there will exist an intermediate range for the global popu-
lation composition such that the reciprocator signal does convey useful
information. This case is complicated and we illustrate it under the
simplifying assumption that the group size is two. Suppose that re-
ciprocators who emit a reciprocator signal comply with the norm and
engage in monitoring, and that reciprocators who do not emit recipro-
cator signals never monitor and comply if and only if they are matched
with someone who emits the reciprocator signal. Given this behavior
of reciprocators, opportunists™ best response, provided the damage
from punishment is high enough, is to comply when their partner
emits a reciprocator signal and to extract more than the norm when
their partner emits none. In this case, players emitting reciprocator
242 Sethi and Somanathan

signals do much better than those emitting none. Within this group,
opportunists obtain greater payoffs than reciprocators since they never
monitor and comply only when they observe a reciprocator signal
from their partner. Nevertheless, this advantage can be outweighed by
the fact that reciprocators are more likely to emit reciprocator signals in
the ¬rst place.10
Can reciprocity be evolutionarily stable even in the absence of com-
mitment, assortation, or parochialism? If the costs of monitoring and
sanctioning are negligible when there is complete norm compliance,
there can be stable groups consisting of a mixture of reciprocators and
pure cooperators (who comply with, but do not enforce, the norm).
This stability is of a rather tenuous nature since it can be disrupted by
the periodic appearance of individuals who violate the norm and are
punished by reciprocators for doing so. If, however, behavior is trans-
mitted across generations through a cultural process that is partly con-
formist (in the sense that widespread behaviors are replicated at greater
rates than less common but equally rewarding behaviors), then such
groups can be stable in a more robust sense. Conformist transmission,
however, can result in the stabilization of virtually any behavioral
norms, including those that are antisocial and inef¬cient. One way to
reduce the multiplicity of potential outcomes is to allow for cultural se-
lection to operate in structured populations. In this model, groups are
located in accordance with a spatial pattern in which each group has
well-de¬ned neighbors. Members of groups that exhibit ef¬cient norms
will enjoy higher material payoffs than members of groups that do not,
and such norms may therefore spread through the population by the
imitation of successful practices found in neighboring groups. The
study of structured populations holds considerable promise in helping
identify additional mechanisms for the survival and spread of strong
One further direction in which work on the evolution of reciproc-
ity can pro¬tably proceed is the following. Several researchers have
recently provided parsimonious representations of preferences that can
be used to account simultaneously for data from a variety of strategic
environments. These speci¬cations are free of any particular experi-
mental context and re¬‚ect concerns for distribution, ef¬ciency, and rec-
iprocity.12 Evolutionary models can build on this literature by shifting
focus from the analysis of behavioral norms in particular environ-
ments to the emergence and stability of general purpose rules that are
equipped to deal with multiple and novel situations.
Norm Compliance and Strong Reciprocity 243

8.5 Conclusions

Social norms that have evolved in a particular economic environment
often continue to govern behavior in other contexts. Even as the rela-
tive economic importance of traditional local commons has diminished
with the expansion of state and private property, norms of restraint
and enforcement that arose as a substitute for governments and mar-
kets in earlier environments continue to make their presence felt in
more modern institutions such as ¬rms, unions, and bureaucracies.
Compliance with such norms often results in greater economic ef¬-
ciency than does opportunistic behavior. Viewed in this light, norms
of reciprocity are an important component of social capital, and an un-
derstanding of their origins and persistence may help to prevent their
erosion. The literature on the evolution of strong reciprocity is a patch-
work of models, each of which emphasizes a different mechanism
under which reciprocators can survive in competition with purely
opportunistic individuals. We have identi¬ed three broad themes”
commitment, parochialism, and assortation”that appear repeatedly
in the literature. These effects, separately or in combination, are largely
responsible for the departures from narrow self-interest that humans
display in the experimental laboratory and in daily life.


The claims made in sections 8.2“8.3 in the text are proved formally be-
low. The common pool resource game involves n players with appro-
priator i choosing extraction x i at a cost °aXÞx i where X ¼ i¼1 x i
denotes aggregate extraction. The payoffs of individual i are:

p i ¼ x i °1 À aXÞ: °1Þ
The ef¬cient level of aggregate extraction maximizes aggregate payoffs
i¼1 p i ¼ X°1 À aXÞ and is given by:

XÃ ¼ °2Þ

Reciprocators comply with and enforce a norm that prescribes, for each
individual, the extraction level:

1 1
xr ¼ XÃ ¼ °3Þ
n 2an
244 Sethi and Somanathan

Reciprocators monitor others at a cost g > 0 and are able to detect and
sanction all violators. Violators incur a cost d as a result of each sanc-
tion. Opportunists simply choose extraction levels that maximize their
payoffs (1). Let r denote the number of reciprocators in the community.
A opportunist i who has chosen to extract optimally (and hence violate
the norm) must choose a level of extraction:

1 À aX
xi ¼ :

Since X is common to all individuals, all opportunists who violate the
norm will choose the same extraction level. Let x v denote this level,
and let v a n À r represent the number of opportunists who choose it.
X ¼ °n À vÞx r þ vx v :

Using the two previous equations we obtain:
ax v ¼ 1 À a°°n À vÞx r þ vx v Þ
which, using (3), simpli¬es to yield:
1 nþv
x¼ °4Þ
2an 1 þ v

Aggregate extraction is
1 nþv 1 n þ v°2n À 1Þ
X ¼ °n À vÞ þv ¼ °5Þ
2an 1 þ v 1þv
2an 2an

Using (1), (4), and (5) and taking into account the sanctions imposed on
violators, we get:

1 °n þ vÞ 2
v v
p ¼ x °1 À aXÞ À dr ¼ À dr °6Þ
4a n 2 °1 þ vÞ 2

where p v is the payoff from violation. The payoff from compliance is:
p c ¼ x r °1 À aXÞ ¼ °7Þ
4a n 2 °1 þ vÞ

In equilibrium, a unilateral deviation should not bene¬t any opportun-
ist. If v A ½1; n À r À 1Š, this implies the following conditions:
1 °n þ v þ 1Þ 2
À dr °8Þ
4a n 2 °2 þ vÞ 2
4a n 2 °1 þ vÞ
Norm Compliance and Strong Reciprocity 245

1 °n þ vÞ 2 1 nþvÀ1
À dr b °9Þ
4a n 2 °1 þ vÞ 2 n2v

The ¬rst states that an opportunist in compliance cannot pro¬t by
switching to noncompliance; the second that an opportunist in viola-
tion cannot pro¬t by switching to compliance. If v ¼ 0 in equilibrium,
only the former condition need be satis¬ed, and if v ¼ 1, only the
The parameters n; a, and d and the number of reciprocators r de¬ne
a game played by the n À r opportunists who choose their extraction
levels strategically, with the number of violators v being determined in
equilibrium. Let this game be denoted G°n; a; d; rÞ. We then have:

Proposition 1 Every game G°n; a; d; rÞ has a unique equilibrium. The
equilibrium number of violators v is nonincreasing in d.

From (8“9), the number of violators v at any asymmetric equilibrium
must satisfy:

F°vÞ a d a G°vÞ
1 °n þ v þ 1Þ 2 nþv
F°vÞ ¼ À
4ar n 2 °2 þ vÞ 2 4ar n 2 °1 þ vÞ
1 °n þ vÞ 2 1 nþvÀ1
G°vÞ ¼ À
4ar n 2 °1 þ vÞ 2 4ar n2v

Note that F°v À 1Þ ¼ G°vÞ. Hence, (10) de¬nes a sequence of intervals
f½F°vÞ; F°v À 1ފgv¼1 such that there is an asymmetric equilibrium
with v violators if and only if d A ½F°vÞ; F°v À 1ފ. If d does not fall within
any of these intervals, then equilibrium is symmetric. If d > F°0Þ, there
is no violation in equilibrium, while if d < F°n À rÞ there is no compli-
ance in equilibrium. Note that raising d lowers or leaves unchanged
the equilibrium value of v.
Proposition 1 allows us to write the number of violators as a func-
tion of the number of reciprocators v ¼ v°rÞ. This in turn de¬nes ag-
gregate extraction and the payoffs from compliance and violation as
functions of r. The payoff obtained by reciprocators is therefore
p r °rÞ ¼ p c °rÞ À g: °11Þ

and the mean payoff received by opportunists is:
246 Sethi and Somanathan

v°rÞp v °rÞ þ °n À r À v°rÞÞp c °rÞ
p m °rÞ ¼ °12Þ

Suppose that the share of reciprocators in the population as a whole is
given by r, and that this population is randomly distributed across
communities. The probability that a community formed in this manner
will contain precisely r reciprocators is given by:

r r °1 À rÞ nÀr :
p°r; rÞ ¼
°n À rÞ!r!

The expected payoffs of reciprocators and opportunists in the popula-
tion as a whole is given by:
Pn r
r¼1 p°r; rÞp °rÞ
p °rÞ ¼
r¼1 p°r; rÞ
P nÀ1 m
r¼0 p°r; rÞp °rÞ
p °rÞ ¼
r¼1 p°r; rÞ

The mean payoff in the population as a whole is simply:

p°rÞ ¼ rp r °rÞ þ °1 À rÞp m °rÞ:
Suppose that the evolution of the population share r is governed by
the replicator dynamics:
r ¼ °p r °rÞ À p°rÞÞr:

Then we have:

Proposition 2 Suppose n and a are given. Then there exists d > 0 such
that an opportunist population is stable if d a d. If d > d, then there
exists a nondecreasing and bounded function g°dÞ such that an oppor-
tunist population is stable if and only if g > g°dÞ.

The stability of r ¼ 0 depends on whether or not p m °0Þ is greater
than p r °1Þ. This is because:

lim p m °rÞ ¼ p m °0Þ

lim p r °rÞ ¼ p r °1Þ

All opportunists violate the norm when r ¼ 0, so in this case v ¼ n
Norm Compliance and Strong Reciprocity 247

1 °n þ nÞ 2 1
m v
p °0Þ ¼ p °0Þ ¼ ¼ :
4a n 2 °1 þ nÞ 2 a°1 þ nÞ 2

From (11) and (7), we have:
p r °1Þ ¼ Àg
4a n 2 °1 þ vÞ

1 1
r m
p °1Þ À p °0Þ ¼ À Àg
a°1 þ nÞ 2
4a n 2 °1 þ vÞ

1 °n À 1Þ°n 2 À n À 3vn À vÞ
¼ Àg
an 2 °1 þ vÞ°1 þ nÞ 2

The ¬rst term is positive if and only if n 2 À n À 3vn À v > 0. This
v< n
3n þ 1

There exists d > 0 such that the above will not be satis¬ed for any
d < d, in which case the opportunist population must be stable. If
d > d, then stability holds if and only if g < g where:

1 °n À 1Þ°n 2 À n À 3vn À vÞ
g¼ :
an 2 °1 þ vÞ°1 þ nÞ 2

The right-hand side of the above expression is decreasing in v. Since v
is nonincreasing in d; g is nondecreasing in d. Finally we have:

Proposition 3 A reciprocator population is unstable for all parameter

The stability of r ¼ 1 requires p r °nÞ to be greater than p m °n À 1Þ.
When r ¼ n À 1, the single opportunist can comply with the norm and
obtain a payoff p r °nÞ þ g. Since this payoff is feasible, under optimal
choice we must have p m °n À 1Þ b p r °nÞ þ g > p r °nÞ. Hence r ¼ 1 is


1. Altruism may also have this effect, but does so in a narrower range of environments
which exclude those considered in this chapter (Bester and Guth 1998).
248 Sethi and Somanathan

2. For an overview of the evidence from ¬eld studies, see Bromley (1992) and Ostrom
(1990). Laboratory experiments designed to replicate common pool resource environ-
ments reveal extensive sanctioning behavior that is broadly consistent with the ¬ndings
from ¬eld studies (Ostrom, Walker, and Gardner 1992); see also Fehr and Fischbacher
(this volume, chapter 6).

3. A mathematical analysis of this model with proofs of all claims made in the text may
be found in the appendix.

4. It is not essential to the argument that the norm prescribe behavior that is optimal
in this sense, only that it result in greater payoffs for the group than would be observed
under opportunistic extraction.
5. The section to follow draws on our considerably more extensive survey (Sethi and
Somanathan 2003). Other evolutionary models of reciprocity that rely on the power of
¨ ¨
commitment include Guth and Yaari (1992), Guth (1995), Sethi (1996), Huck and Oechss-
ler (1999), and Friedman and Singh (1999). Gintis (2000) and Sethi and Somanathan
(2001) analyze models in which both commitment (the power to in¬‚uence the actions of
others) and parochialism (the conditioning of one™s behavior on the composition of one™s
group) play a role.
6. This is, of course, analogous to Hamilton™s argument that an altruistic gene will
spread in a population if individuals share a suf¬ciently high proportion of their genes
on average with those with whom they interact (Hamilton 1964).
7. This model of partial assortation on the basis of signaling is due to Frank (1987, 1988);
see also Guttman (2002). For a model in which prior cooperative acts are themselves used
as signals, see Nowak and Sigmund (1998).
8. See Bowles and Gintis (2004) for a model along these lines.
9. Gintis (2000) models this effect in an empirically motivated model of public goods
10. See Frank (1987), Robson (1990), Guttman (2002) and Smith and Bliege Bird (this vol-
ume, chapter 4) for further discussion and variations on the theme of signaling.
11. Models in which stable mixtures of reciprocators and pure cooperators can arise
include Axelrod (1986) and Sethi and Somanathan (1996); see Gale, Binmore, and
Samuelson (1995) for similar ¬ndings in a different context. A discussion of conformist
transmission and its implications may be found in Boyd and Richerson (1995). The model
of structured populations mentioned here is due to Boyd and Richerson (2000); see also
Boyd et al. (this volume, chapter 7).
12. Important contributions include Rabin (1993), Levine (1998), Fehr and Schmidt
(1999), Bolton and Ockenfels (2000), Falk and Fischbacher (1998), Dufwenberg and Kirch-
steiger (1998), and Charness and Rabin (2002); see also Falk and Fischbacher (this vol-
ume, chapter 6).


Axelrod, R. 1986. ˜˜An Evolutionary Approach to Norms.™™ American Political Science Re-
view 80:1095“1111.
Bester, H., and W. Guth. 1998. ˜˜Is Altruism Evolutionarily Stable?™™ Journal of Economic Be-
havior and Organization 34:193“209.
Norm Compliance and Strong Reciprocity 249

Bolton, G. E., and A. Ockenfels. 2000. ˜˜ERC: A Theory of Equity, Reciprocity and Compe-
tition.™™ American Economic Review 90:166“193.
Bromley, D., ed. 1992. Making the Commons Work. San Francisco: ICS Press.

Bowles, S. and H. Gintis. 2004. ˜˜The Evolution of Strong Reciprocity,™™ Theoretical Popula-
tion Biology 65:17“28.

Boyd, R., H. Gintis, S. Bowles, and P. J. Richerson. 2004. Chapter 7, this volume.
Boyd, R., and P. J. Richerson. 1985. Culture and the Evolutionary Process. Chicago: Univer-
sity of Chicago Press.

Boyd, R., and P. J. Richerson. 2000. ˜˜Group Bene¬cial Norms Can Spread Rapidly in a
Structured Population.™™ Mimeo. University of California at Los Angeles.
Charness, G., and M. Rabin. 2002. ˜˜Understanding Social Preferences with Simple Tests.™™
Quarterly Journal of Economics 117:817“869.
Dufwenberg, M., and G. Kirchsteiger. 1998. ˜˜A Theory of Sequential Reciprocity.™™ Center
Discussion Paper 9837. Tilburg University.
Falk, A., and U. Fischbacher. 1998. ˜˜A Theory of Reciprocity.™™ Mimeo. University of
Falk, A., and U. Fischbacher. 2004. Chapter 5, this volume.
Fehr, E., and U. Fischbacher. 2004. Chapter 6, this volume.

Fehr, E., and K. M. Schmidt. 1999. ˜˜A Theory of Fairness, Competition, and Coopera-
tion.™™ Quarterly Journal of Economics 114:817“868.
Frank, R. H. 1987. ˜˜If Homo economicus Could Choose His Own Utility Function, Would
He Want One with a Conscience?™™ American Economic Review 77:593“604.
Frank, R. H. 1988. Passions within Reason: The Strategic Role of the Emotions. New York:
W. W. Norton.
Friedman, D., and N. Singh. 1999. ˜˜On the Viability of Vengeance.™™ Mimeo. University of
California at Santa Cruz.
Gale, J., K. Binmore, and L. Samuelson. 1995. ˜˜Learning to be Imperfect: The Ultimatum
Game.™™ Games and Economic Behavior 8:56“90.
Gintis, H. 2000. ˜˜Strong Reciprocity and Human Sociality.™™ Journal of Theoretical Biology
Guth, W. 1995. ˜˜An Evolutionary Approach to Explaining Cooperative Behavior by Re-
ciprocal Incentives.™™ International Journal of Game Theory 24:323“344.
Guth, W., and M. Yaari. 1992. ˜˜Explaining Reciprocal Behavior in Simple Strategic
Games: An Evolutionary Approach.™™ In U. Witt (ed.) Explaining Forces and Change:
Approaches to Evolutionary Economics. Ann Arbor: University of Michigan Press.
Guttman, J. M. 2003. ˜˜Repeated Interaction and the Evolution of Preferences for Reciproc-
ity.™™ Economic Journal 113:631“656.
Hamilton, W. D. 1964. ˜˜The Genetical Evolution of Social Behavior.™™ Journal of Theoretical
Biology 7:1“16.
Hardin, G. 1968. ˜˜The Tragedy of the Commons.™™ Science 162:1243“1248.
250 Sethi and Somanathan

Huck, S., and J. Oechssler. 1999. ˜˜The Indirect Evolutionary Approach to Explaining Fair
Allocations.™™ Games and Economic Behavior 28:13“24.
Levine, D. K. 1998. ˜˜Modeling Altruism and Spitefulness in Experiments.™™ Review of Eco-
nomic Dynamics 1:593“622.
Nowak, M. A., and K. Sigmund. 1998. ˜˜Evolution of Indirect Reciprocity by Image Scor-
ing.™™ Nature 393:573“577.
Ostrom, E. 1990. Governing the Commons: The Evolution of Institutions for Collective Action.
Cambridge: Cambridge University Press.
Ostrom, E., J. Walker, and R. Gardner. 1992. ˜˜Covenants with and without a Sword: Self-
Governance Is Possible.™™ American Political Science Review 86:404“417.
Rabin, M. 1993. ˜˜Incorporating Fairness into Game Theory and Economics™™ American Eco-
nomic Review 83:1281“1302.
Robson, A. 1990. ˜˜Ef¬ciency in Evolutionary Games: Darwin, Nash, and the Secret Hand-
shake.™™ Journal of Theoretical Biology 144:379“396.
Sethi, R. 1996. ˜˜Evolutionary Stability and Social Norms.™™ Journal of Economic Behavior and
Organization 29:113“140.
Sethi, R., and E. Somanathan. 1996. ˜˜The Evolution of Social Norms in Common Property
Resource Use.™™ American Economic Review 86:766“788.

Sethi, R., and E. Somanathan. 2001. ˜˜Preference Evolution and Reciprocity.™™ Journal of
Economic Theory 97:273“297.

Sethi, R., and E. Somanathan. 2003. ˜˜Understanding Reciprocity.™™ Journal of Economic Be-
havior and Organization 50:1“27.
Smith, E. A., and R. B. Bird. 2004. Chapter 4, this volume.
IV Reciprocity and Social
9 Policies That Crowd out
Reciprocity and Collective

Elinor Ostrom

9.1 Introduction

The extensive empirical research presented in this volume and else-
where (see reviews by Bowles 1998; Frey and Jegen 2001; E. Ostrom
1998, 2000) challenges the assumption that human behavior is driven
in all settings entirely by external material inducements and sanctions.
Instead of assuming the existence of a single type of ˜˜pro¬t maxi-
mizing™™ or ˜˜utility maximizing™™ individual, a better foundation for
explaining human behavior is the assumption that multiple types of
individuals exist in most settings. Among the types of individuals
likely to be present in any situation are ˜˜rational egoists,™™ who focus
entirely on their own expected material payoffs. Neoclassical econom-
ics and non-cooperative game theory have usually assumed that ratio-
nal egoists are the only type of player that scholars need to assume in
order to generate useful and validated predictions about behavior.
Substantial research in nonmarket experimental settings now provides
strong evidence that in addition to rational egoists, many settings also
involve ˜˜strong reciprocators,™™ who are motivated by both intrinsic
preferences and material payoffs. As discussed in this volume, strong
reciprocators will frequently adopt strategies of conditional coopera-
tion and conditional punishment in settings where individuals can ob-
serve each other™s behavior.
Laboratory experiments of social dilemmas, trust games, dictator
games, and ultimatum games repeatedly ¬nd higher-than-predicted
cooperative behavior that cannot be explained by theories assuming
the existence of only rational egoists. ˜˜It is a well known fact in the ex-
perimental literature that in games like the trust game, there is always
a 30“40 percentage of individuals who act in a purely egoistic way™™
(Frey and Benz 2001, 9). This leaves 60 to 70 percent of the other
254 Ostrom

individuals who tend to follow more complex strategies involving
some levels of trust and reciprocity. Furthermore, the proportion of
different types of individuals is likely to change over time due to the
self-selection of individuals into diverse types of situations and due to
endogenous changes in preferences and expectations over time as a
result of the patterns of interactions and outcomes achieved (see E.
Ostrom and Walker 2003).
A considerable body of contemporary policy analysis is, however,
based on the earlier widely accepted presumption that all individu-
als are strictly rational egoists motivated entirely by external payoffs.
When rational egoists ¬nd themselves in a wide diversity of collective-
action situations, the predicted result is a de¬cient equilibrium of zero
or very low contributions to joint outcomes. Consequently, centrally
designed and externally implemented material incentives”both posi-
tive and negative”are seen as universally needed to overcome these
Pareto-de¬cient equilibria. Leviathan is alive and well in our policy
textbooks. The state is viewed as a substitute for the shortcomings of
individual behavior and the presumed failure of community. Some-
how, the agents of the state are assumed to pay little attention to their
own material self-interest when making of¬cial decisions and to know
and seek ˜˜the public interest.™™
For contemporary policy analysis to have a ¬rm empirical founda-
tion, it is necessary to adopt a broader theory of human behavior that
posits multiple types of individuals”including rational egoists as well
as strong reciprocators”and examines how the contexts of collective
action affect the mix of individuals involved.
In section 9.3, I will brie¬‚y review the evidence regarding intrinsic
motivations. The evidence shows that in some settings (particularly
those where individuals lose a sense of control over their own fate),
providing external inducements to contribute to collective bene¬ts
may actually produce counterintentional consequences. External in-
centives may ˜˜crowd out™™ behaviors that are based on intrinsic prefer-
ences so that lower levels of contributions are achieved with the
incentives than would be achieved without them (Frey 1994, 1997). Ex-
ternal incentives may also ˜˜crowd in™™ behaviors based on intrinsic
preferences and enhance what could have been achieved without these
In section 9.4, I will then discuss the delicate problem of designing
institutions that enhance cooperation rather than crowding it out. In-
stead of relying on the state as the central, top-down substitute for all
Policies That Crowd out Reciprocity and Collective Action 255

public problem solving, it is necessary to design complex, polycentric
orders that involve both public governance mechanisms and private
market and community institutions that complement each other (see
McGinnis 1999a, 1999b, 2000). Reliance primarily on national govern-
ments crowds out public and private problem solving at regional and
local levels (and radical decentralization would crowd out public prob-


. 8
( 13)