. 6
( 13)


cuniary valuations of the good are public information, all involved
parties know the surplus. Since there is a known ¬xed surplus, there
is no uncertainty regarding the quality of the good provided by the
seller. The situation can thus be viewed as a market in which the con-
tract (the quality of the good) is enforced exogenously.
If all parties are sel¬sh, competition among the responders does not
matter because the proposer is already predicted to receive the whole
surplus in the bilateral case. Adding competition to the bilateral ulti-
matum game therefore has no effect on the power of the proposer. It
is also irrelevant whether there are two, three, or more competing
responders. The self-interest hypothesis thus implies a very counterin-
tuitive result”namely, that increasing the competition among the res-
The Economics of Strong Reciprocity 159

ponders does not affect the share of the surplus that the responders re-
ceive. Fischbacher, Fong, and Fehr (2003) tested this prediction by con-
ducting ultimatum games with one, two, and ¬ve responders. To allow
for convergence and learning effects, in each experimental session a
large group of subjects played the same game for twenty periods. For
example, in the case with two responders, one-third of the subjects
were always in the role of the proposer and two-thirds of the subjects
were in the role of the responder. In every period the proposers and
the responders were randomly rematched to ensure the one-shot na-
ture of the interactions. All subjects knew that the experiment would
end after period twenty. The results of these experiments are presented
in ¬gure 5.1.
Figure 5.1 shows the responder™s average share of the surplus in
each period across conditions. In the bilateral case, the average share
is”except for period 1”very close to 40 percent. Moreover, the share
does not change much over time. In the ¬nal period, the responder still
appropriates slightly more than 40 percent of the surplus. In the case of
two responders, however, the situation changes dramatically. Already
in period 1 the responder™s share is reduced by 16.5 percentage points
relative to the bilateral case. Moreover, from period 1 until period 12


50 1 responder
Average responder share


2 responders


5 responders

1 2 34 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Figure 5.1
Responder share in the ultimatum game with one, two, and ¬ve competing responders.
Source: Fischbacher, Fong, and Fehr 2003.
160 Fehr and Fischbacher

responder competition induces a further reduction of the responder™s
share by 14 percentage points (from 35 percent to 21 percent) and in
the ¬nal period the share is even below 20 percent. Thus, the addition
of just one more responder has a dramatic impact on the responder™s
share. If we add three additional responders, the share goes down
even further. From period 3 onwards, it is below 20 percent and comes
close to 10 percent in the second half of the session.7
The reason why the responder™s share decreases when competition
increases is that the rejection probability of the responders declines
when there are more competing responders. These facts can be parsi-
moniously explained if one takes the presence of strongly reciprocal or
inequity averse responders into account (Falk and Fischbacher this vol-
ume, chapter 6; Fehr and Schmidt 1999). Recall that strongly reciprocal
responders reject low offers in the bilateral ultimatum game because
by rejecting they are able to punish the unfair proposers. They can al-
ways ensure this punishment in the bilateral case, while in the compet-
itive case this is no longer possible. In particular, if one of the other
responders accepts a low offer, it is impossible for a reciprocal res-
ponder to punish the proposer. Since there is a substantial fraction of
sel¬sh responders, the probability that one of the other responders is
sel¬sh increases with a larger number of competing responders. This
means, in turn, that the expected non-pecuniary return from the rejec-
tion of a low offer is smaller when the number of competing respond-
ers increases. Therefore, strongly reciprocal responders will reject less
frequently as more competing responders are added.

5.3.2 The Effects of Competition under Endogenous Contract
The previous example illustrates that the self-interest model underesti-
mates the power of competition. This example should not, however,
make us believe that suf¬cient competition will in general weaken or
remove the impact of fairness on market outcomes. Quite the contrary.
In this section, we will show that the presence of strongly reciprocal
individuals may completely nullify the impact of competition on mar-
ket outcomes. Whether competition does have the effects illustrated in
¬gure 5.1 depends critically on the enforceability of the contracts.
The double auction experiments conducted by Fehr and Falk (1999)
help to illustrate this argument. Fehr and Falk deliberately chose the
double auction as the trading institution because a large body of re-
search has shown the striking competitive properties of experimental
The Economics of Strong Reciprocity 161

double auctions. In hundreds of such experiments, prices and quan-
tities quickly converged to the competitive equilibrium predicted by
standard self-interest theory (see Davis and Holt 1993, for a survey of
important results). Therefore, showing that strong reciprocity renders
competition completely powerless in an experimental double auction
provides a strong piece of evidence in favor of the importance of
strong reciprocity in markets.
Fehr and Falk use two treatment conditions: a competitive condition
and a bilateral condition in which competition is completely removed.
In the competitive condition, they embed the gift exchange framework
into the context of an experimental double auction that is framed in la-
bor market terms.8 The crucial difference between the competitive con-
dition and the gift exchange game described earlier in this chapter is
that in the competitive condition, both experimental ¬rms and experi-
mental workers can make wage bids in the interval [20, 120] because
the workers™ reservation wage is 20 and the maximum revenue from a
trade is 120. If a bid is accepted, a labor contract is concluded and the
worker has to choose the effort level. As in the gift exchange game, the
workers (˜˜responders™™) can freely choose any feasible effort level. They
have to bear effort costs while the ¬rm (˜˜proposer™™) bene¬ts from the
effort. Thus, the experiment captures a market in which the quality of
the good traded (˜˜effort™™) is not exogenously enforced but is chosen
by the workers. Workers may or may not provide the effort level that
is expected by the ¬rms. In the competitive double auction, there are
8 ¬rms and 12 workers. Each ¬rm can employ at most 1 worker. A
worker who enters into a contract has costs of 20. Therefore, due to the
excess supply of labor, the competitive wage level is 20. A double auc-
tion lasts for ten periods and each period lasts for three minutes.9
In contrast to the double auction, ¬rms in the bilateral condition are
exogenously matched with a worker. If a worker rejects the ¬rm™s of-
fer, both parties earn nothing. The bilateral condition consists of a se-
ries of ten one-shot gift exchange games that are also framed in labor
market terms. There are ten ¬rms (proposers) and ten workers (res-
ponders). In each of the ten periods, each ¬rm is matched with a differ-
ent worker. Firms have to make an offer to the matched worker in each
period. If the worker accepts, he has to choose the effort level. As in the
competitive condition, a worker who accepts a wage offer has costs of
20, and the maximum revenue from a trade is 120. The self-interest
model predicts that in both conditions the workers will only provide
the minimum effort so that the ¬rms will pay a wage of 20 or 21
162 Fehr and Fischbacher

in equilibrium. However, we know already from bilateral ultimatum
games that ¬rms (proposers) cannot reap the whole surplus”that is,
wages in the bilateral gift exchange game can also be expected to be
much higher than predicted by the self-interest model. Moreover, since
in the gift exchange game the effort is in general increasing in the wage
level, ¬rms have even an additional reason to offer workers a substan-
tial share of the surplus. The task, therefore, is to determine what ex-
tent competition in the double auction pushes wages below the level
in the bilateral condition.
Figure 5.2 shows the evolution of wages in both conditions. This ¬g-
ure indicates the startling result that competition has no long-term im-
pact on wage formation in this setting. Wages in the double auction are
slightly lower than the wages in the bilateral condition only in the
beginning periods but since workers responded to lower wages with
lower effort levels, ¬rms raised their wages from period four onwards.
In the last ¬ve periods, ¬rms paid even slightly higher wages in the
double auction than in the bilateral condition; this difference is not sig-
ni¬cant, however. It is also noteworthy that competition among the


70 Competitive double auction

Average wages

Bilateral condition



1 2 3 4 5 6 7 8 9 10

Figure 5.2
Wage levels in the competitive double auction and the bilateral condition. Source: Fehr
and Falk 1999.
The Economics of Strong Reciprocity 163

workers was extremely intense in the double auction. In each period,
many workers offered to work for wages below 30 but ¬rms preferred
instead to pay workers on average wages around 60. It was impossible
for the workers to get a job by underbidding the going wages because
the positive effort-wage relation made it pro¬table for the ¬rms to pay
high, noncompetitive, wages.10
The previous evidence in this section indicates that strong reciproc-
ity severely limits the impact of competition in markets in which effort
or quality is not enforced exogenously. It restricts the impact of compe-
tition on wages by generating an ef¬ciency wage effect that renders
it pro¬table for the ¬rms to pay noncompetitive wages. As is well-
known, such noncompetitive wages may in turn cause involuntary
unemployment (Akerlof 1982). In addition, the existence of strongly
reciprocal types may endogenously generate a distinction between
insiders and outsiders. Firms are of course interested in workers who
do not exploit every opportunity to shirk”workers who are loyal and
who also perform when they are unobserved. Since workers are heter-
ogeneous in this regard (and since a worker™s type may be dif¬cult
to ascertain), ¬rms are generally reluctant to replace existing workers
with new workers even if the new workers are willing to work for less
than the going wage. This protects ¬rms™ existing workforce from out-
side competition.
Finally, strong reciprocity may also contribute to the existence of
noncompetitive wage differentials. In the 1980s and the early 1990s,
there has been a heated debate about whether inter-industry wage
differentials should be interpreted as noncompetitive job rents. The
debate did not result in a consensus because the results could also be
interpreted as re¬‚ecting unobservable heterogeneity in working con-
ditions and unobservable heterogeneity in skill levels.11 Laboratory
experiments can help resolve some of the open issues because it is pos-
sible to rule out heterogeneity in working conditions and skill levels
in the laboratory. This situation was done by Fehr, Gachter, and Kirch-
steiger (1996), who embedded the gift exchange framework into a
competitive market environment in which experimental ¬rms differed
according to their pro¬t opportunities. Under their experiment, once a
worker has accepted a ¬rm™s wage offer and before she makes her ef-
fort choice, she is informed about the ¬rm™s pro¬t opportunity. This
procedure ensured that only the effort decision”but not the contract
acceptance decision”of the worker is affected by the ¬rm™s pro¬t
164 Fehr and Fischbacher

opportunity. Both ¬rms and workers know this information revelation
procedure in advance. The experiment shows that ¬rms with better
pro¬t opportunities pay systematically higher wages and higher job
rents. This wage policy is quite rational because a given effort increase
leads to a larger pro¬t increase for a more pro¬table ¬rm. Hence, high-
pro¬t ¬rms have a stronger incentive to appeal to the workers™ reci-
procity by paying high wages.

5.4 Cooperation

Free-riding incentives are a pervasive phenomenon in social life. Par-
ticipation in collective actions against a dictatorship or in industrial
disputes, collusion among ¬rms in oligopolistic markets, the preven-
tion of negative environmental externalities, workers™ effort choices
under team-based compensation schemes, or the exploitation of a com-
mon resource are typical examples. In these cases, the free rider cannot
be excluded from the bene¬ts of collective actions or the public good
even though he does not contribute. In view of the ubiquity of cooper-
ation problems in modern societies, it is crucial to understand the
forces shaping people™s cooperation. In this section, we will show that
the neglect of strong reciprocity may induce economists to completely
misunderstand the nature of many cooperation problems. As we will
see, a key to the understanding of cooperation problems is again the
interaction between sel¬sh and strongly reciprocal types and how this
interaction is shaped by the institutional environment.

5.4.1 Conditional Cooperation
Strong reciprocity changes the typical cooperation problem for two
reasons. First, strongly reciprocal subjects are willing to cooperate
if they are sure that the other people who are involved in the coopera-
tion problem will also cooperate. If the others cooperate”despite pe-
cuniary incentives to the contrary”they provide a gift that induces
strongly reciprocal subjects to repay the gift, that is, reciprocators are
conditionally cooperative. Second, strongly reciprocal subjects are will-
ing to punish free-riders because free-riders exploit the cooperators.
Thus, if potential free-riders face reciprocators, they have an incentive
to cooperate in order to prevent being punished.
The impact of strong reciprocity on cooperation can be demon-
strated in the context of the Prisoner™s Dilemma (PD) in the following
situation. Subject A and Subject B both posses £10. Each person can
The Economics of Strong Reciprocity 165

either keep her 10 pounds or transfer it to the other person. If either
subject transfers the money, the experimenter triples the transferred
amount, that is, the recipient receives £30 from the transfer. A and B
each have to decide simultaneously whether to keep or whether to
transfer her £10. If both subjects transfer their money, both earn £30. If
they both keep their own money, both earn only £10. Moreover, irre-
spective of whether the other subject transfers the money, it is always
in the subject™s self-interest to keep the £10.12 The self-interest hypothe-
sis predicts, therefore, that both subjects will keep their money. In fact,
however, many subjects cooperate in situations like this one (see Led-
yard 1995; Dawes 1980). For example, in one-shot PDs cooperation
rates are frequently between 40 and 60 percent.
In the presence of suf¬ciently reciprocal subjects, cooperative out-
comes in the PD can be easily explained because the above game”
although a PD in terms of material payoffs”is not a PD in utility pay-
offs. It is, instead, a coordination game with two equilibria. If both sub-
jects are reciprocators and if A believes that B will cooperate (i.e.,
transfer the money), A prefers to cooperate. The same holds true for B if
B believes that A will cooperate. Thus, the strategy combination (coop-
erate, cooperate) constitutes an equilibrium. Similarly, if both believe
that the other person will defect (i.e., keep the money), A and B both
prefer to defect as well. Therefore, the combination (defect, defect) is
also an equilibrium.13
The fact that the PD is transformed into a coordination game in the
presence of strongly reciprocal players can explain two further facts. It
has been shown dozens of times that communication leads to much
higher cooperation rates in PDs and other social dilemma games (Sally
1995).14 If all subjects were completely sel¬sh, the positive impact of
communication would be dif¬cult to explain. If, however, the PD in
economic terms is in fact a coordination game, communication allows
the subjects to coordinate on the superior equilibrium. It has also been
shown that cooperation is affected by how the PD is framed. If the PD
is framed in ˜˜cooperative™™ terms, subjects are more likely to cooperate
than if it is framed in ˜˜competitive™™ terms. Since it is likely that the ˜˜co-
operative™™ framework induces more optimistic beliefs about the be-
havior of the other player than the ˜˜competitive™™ framework, subjects
are more likely to coordinate on the good equilibrium while working
within the ˜˜cooperative™™ framework.
If the actual preferences of the subjects do transform social dilemma
situations like the PD game into a coordination game, the self-interest
166 Fehr and Fischbacher

hypothesis induces economists to fundamentally misperceive social di-
lemma problems. In view of the importance of this claim, it is desirable
to have more direct evidence on this. Fischbacher, Gachter, and Fehr
(2001) and Croson (1999) elicited subjects™ willingness to cooperate
conditional on the average cooperation of others™ in the context of
four-person public good games in which the dominant strategy for
each subject was to free-ride completely. It was in the sel¬sh interest
of each subject to free-ride, although the socially optimal decision
required the contribution of each individual™s whole endowment to
the public good.
Both studies found considerable evidence for the presence of condi-
tional cooperators.15 The results of the Fischbacher, Gachter, and Fehr
study are presented in ¬gure 5.3. This ¬gure shows that 50 percent
of the subjects are willing to increase their contributions to the public
good if the other group members™ average contribution increases, even
though the pecuniary incentives always implied full free-riding. The
behavior of these subjects is consistent with models of strong reciproc-
ity (or inequity aversion). The ¬gure also reminds us that a substantial
fraction of the subjects (30 percent) are complete free-riders while 14



cooperation: 50 %
Own contribution



Total average

“Hump-shaped”: 14 %

0 Free riding: 30 %
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Average contribution level of other group members

Figure 5.3
Contributions of individual subjects as a function of the contribution of other members™
average contribution. Source: Fischbacher, Gachter, and Fehr 2001.
The Economics of Strong Reciprocity 167

percent exhibit a hump-shaped response. Yet taken together, there are
suf¬ciently many conditional cooperators such that an increase in the
other group members™ contribution level causes an increase in the con-
tribution of the ˜˜average™™ individual (see the bolded line in ¬gure 5.3).
The coexistence of conditional cooperators and sel¬sh subjects has
important implications. It implies, for example, that subtle institutional
details may cause large behavioral effects. To illustrate this, assume
that a sel¬sh and a strongly reciprocal subject are matched in the simul-
taneous PD and that the subjects™ type is common knowledge. Since
the reciprocal type knows that the other player in question is sel¬sh,
he knows that this other player will always defect. Therefore, the recip-
rocal type will also defect”that is, (defect, defect) is the unique equi-
librium. Now consider the sequential PD in which the sel¬sh player
decides ¬rst whether to cooperate or to defect. Then the strongly recip-
rocal player observes what the ¬rst-mover did and chooses his action.
In the sequential case, the unique equilibrium outcome is that both
players cooperate because the reciprocal second-mover will match the
choice of the ¬rst-mover. This means that the sel¬sh ¬rst-mover essen-
tially has the choice between the (cooperate, cooperate)-outcome and
the (defect, defect)-outcome. Since mutual cooperation is better than
mutual defection, the sel¬sh player will cooperate. Thus, while in the
simultaneous PD the sel¬sh player induces the reciprocal player to
defect, in the sequential PD the reciprocal player induces the sel¬sh
player to cooperate. This example neatly illustrates how institutional
details interact in important ways with the heterogeneity of the
Since there are many conditional cooperators, the problem of estab-
lishing and maintaining cooperation involves the management of
people™s beliefs. If people believe that the others cooperate to a large
extent, cooperation will be higher compared to a situation where they
believe that others rarely cooperate. Belief-dependent cooperation can
be viewed as a social interaction effect that is relevant in many impor-
tant domains. For example, if people believe that cheating on taxes,
corruption, or abuses of the welfare state are widespread, they them-
selves are more likely to cheat on taxes, take bribes, or abuse welfare
state institutions. It is therefore important that public policy prevents
the initial unraveling of civic duties, because once people start to
believe that most others engage in unlawful behavior, the belief-
dependency of individuals™ behavior may render it very dif¬cult to
reestablish lawful behavior.
168 Fehr and Fischbacher

In an organizational context, the problem of establishing cooperation
among the members of the organization also involves the selection of
the ˜˜right™™ members. A few shirkers in a group of employees may
quickly spoil the whole group. Bewley (1999), for example, reports
that personnel managers use the possibility of ¬ring workers mainly
as a means to remove ˜˜bad characters and incompetents™™ from the
group and not as a threat to discipline the workers. The reason is that
explicit threats create a hostile atmosphere and may even reduce the
workers™ general willingness to cooperate with the ¬rm. Managers
report that the employees themselves do not want to work with
lazy colleagues because these colleagues do not bear their share of
the burden, which is viewed as unfair. Therefore, the ¬ring of lazy
workers is mainly used to protect the group from ˜˜bad characters and
incompetents,™™”to establish internal equity and to prevent the unrav-
eling of cooperation. This supports the view that conditional coopera-
tion is also important inside ¬rms.
Strong reciprocity and conditional cooperation are also likely to
shape the structure of social policies regarding the poor (Fong, Bowles,
and Gintis this volume, chapter 10; Bowles and Gintis 1998; Wax 2000).
The reason is that the political support for policies regarding the poor
depends to a large extent on whether the poor are perceived as ˜˜de-
serving™™ or as ˜˜undeserving.™™ If people believe that the poor are poor
because they do not want to work hard, support for policies that help
the poor is weakened because the poor are perceived as undeserving.
If, on the other hand, people believe that the poor try hard to escape
poverty but that for reasons beyond their control have not been able to
make it, the poor are perceived as deserving. This indicates that the
extent to which people perceive the poor as deserving is shaped by
strong reciprocity. If the poor exhibit good intentions, try to contribute
to society™s output, or if they are poor for reasons that have nothing to
do with their intentions, they are perceived as deserving. In contrast, if
the poor are perceived as lacking the will to contribute to society™s out-
put, they are perceived as undeserving. This means that social policies
that enable the poor to demonstrate their willingness to reciprocate
the generosity of society will mobilize greater political support than
social policies that do not allow the poor to exhibit their good inten-
tions. Wax (2000) convincingly argues that an important reason for
the popularity of former president Bill Clinton™s 1996 welfare reform
initiative was that the initiative appealed to people™s sense of strong
The Economics of Strong Reciprocity 169

5.4.2 Cooperation and Punishment
We argued above that the presence of a sel¬sh subject will induce
the strongly reciprocal subject in the simultaneous PD to defect. This
proposition also holds more generally in the case of n-person social di-
lemma situations. It can be shown theoretically that even a small mi-
nority of sel¬sh subjects induces a majority of reciprocal (or inequity
averse) subjects to free-ride in simultaneous social dilemmas (Fehr and
Schmidt 1999, proposition 4). In an experiment with anonymous inter-
actions, subjects of course do not know whether the other group mem-
bers are sel¬sh or strongly reciprocal. If they interact repeatedly over
time, however, they may learn the others™ types. Therefore, one would
expect that cooperation will eventually unravel in (¬nitely repeated)
simultaneous social dilemma experiments. This unraveling of coop-
eration has indeed been observed in dozens of experiments (Ledyard
1995; Fehr and Schmidt 1999).
This raises the question of whether there are social mechanisms that
can prevent the decay of cooperation. A potentially important mecha-
nism is social ostracism and peer pressure stemming from reciprocal
subjects. Recall that strongly reciprocal subjects exhibit a willingness
to punish unfair behavior and it is very likely that cooperating recip-
rocators view free-riding as very unfair. Yamagichi (1986), Ostrom,
Gardner, and Walker (1994), and Fehr and Gachter (2000a) studied the
impact of punishment opportunities in public goods and social di-
lemma games where the same players could stay together in the same
group for several periods. In the experiments conducted by Fehr and
Gachter, there were two stages. Stage one was the same public goods
game as described in Fischbacher, Gachter, and Fehr (2001). In particu-
lar, the dominant strategy of each player is to free-ride completely in
the stage game, although the socially optimal decision requires each
player to contribute her whole endowment to the public good. In stage
two, after the group has been informed of the contributions of each
group member, each player can assign up to ten punishment points to
any of the other group members. The assignment of one punishment
point reduces the ¬rst-stage income of the punished subject by three
points on average, but it also reduces the income of the punisher.17
This kind of punishment mimics an angry group member scolding a
free-rider or spreading the word so that the free-rider is ostracized”
there is some cost to the punisher, but a larger cost to the free-rider.
Note that since punishment is costly for the punisher, the self-interest
hypothesis predicts zero punishment. Moreover, since rational players
170 Fehr and Fischbacher

will anticipate this, the self-interest hypothesis predicts no difference in
the contribution behavior between a public goods game without pun-
ishment and the game with a punishment opportunity. In both condi-
tions, zero contributions are predicted.
The experimental evidence completely contradicts this prediction
(see ¬gure 5.4).18 In contrast to the game without a punishment oppor-
tunity, where cooperation declines over time and is close to zero in the
¬nal period, the punishment opportunity causes a sharp jump in coop-
eration (compare period 10 with period 11 in ¬gure 5.4). Moreover, in
the punishment condition, there is a steady increase in contributions
until almost all subjects contribute their whole endowment. This sharp
increase occurs because free-riders often get punished, and the less
they give, the more likely punishment is. Cooperators feel that free-
riders take unfair advantage of them and, as a result are willing to
punish them. This induces the punished free-riders to increase cooper-
ation in the following periods. A nice feature of this design is that the
actual rate of punishment is very low in the last few periods”the
mere threat of punishment, and the memory of the pain from previous
punishments, is enough to induce potential free-riders to cooperate.

Average contributions in percent of endowment

80 With punishment



Without punishment

12 34 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Figure 5.4
Average contributions to the public good. Source: Fehr and Gachter 2000a.
The Economics of Strong Reciprocity 171

5.4.3 Strategic Versus Nonstrategic Punishment
Peer pressure, social ostracism, and the cooperation-enhancing punish-
ment of free-riders in general play a key role in the enforcement of
social norms. They are also important in industrial disputes between
workers and ¬rms, in team production settings, in the management of
common property resources, or as an enforcement device for collusion
in oligopolistic industries. For example, striking workers often ostra-
cize strike breakers (Francis 1985) or, under a piece rate system, viola-
tors of production quotas are punished by those who try to maintain
effort-withholding norms (Roethlisberger and Dickson 1947; Whyte
1955).19 During World War I, British men who did not volunteer for
the army faced strong public contempt and were called ˜˜whimps.™™
Ostrom, Gardner, and Walker (1994) also report that punishment is
frequently imposed on those who excessively use common property
resources. They convincingly argue that the successful management of
such resources requires institutions that render the excess extraction
of common resources visible or easy to detect. This enables the users
of the resource to impose sanctions on the wrongdoers.
A further interesting example is provided by Slade (1990), who ana-
lysed the behavior of ¬rms during price wars in oligopolistic indus-
tries. She shows that during price wars, ¬rms sell their products far
below their marginal costs. While this behavior may be rationalized as
part of a complicated punishment strategy in a repeated game involv-
ing only self-interested players, it seems more likely that players get
angry and that their punishment behavior is driven by non-sel¬sh
forces. Anecdotal evidence from oil company marketers supports this
view. According to Slade (personal communication), the marketers
stated that they would follow a rival™s price cut right down to zero if
that rival started a price war. Yet, anecdotal evidence alone, as sugges-
tive as it might be, is not fully convincing.
All of the examples in this section raise similar questions. To what
extent is the punishment observed strategically motivated”that is,
caused by the expectation of future economic bene¬t”and to what
extent is it due to the mere (nonstrategic) desire to punish? Moreover,
what are the implications of the existence of nonstrategic sanctions
over and above what repeated game theory already tells us?
Our answer to these questions is as follows. First, repeated game
theory tells us in fact very little about the actual behavior in in¬nitely
repeated interactions because for suf¬ciently high discount factors
there is typically a plethora of equilibria, including equilibria with no
172 Fehr and Fischbacher

punishment and no cooperation. Thus, at a minimum, the results on
punishment-based cooperation show that people do punish and they
typically coordinate on cooperative outcomes. Second, there are in
fact many situations in which interactions are only one-shot, ¬nitely
repeated, or where people™s discount factors are so low that self-
interested agents cannot sustain cooperation in equilibrium. The re-
sults of Fehr and Gachter (2000a) show that in these situations nonstra-
tegic punishment is a powerful cooperation-enforcement device. Third,
if fairness considerations are an important driving force of nonstrategic
sanctions, it is quite likely that strategic sanctioning is also shaped in
important ways by fairness concerns. In particular, we believe that
many people will forgo the possibility of sanctioning others for purely
pecuniary reasons if the sanction is viewed as unfair. They may refrain
from sanctioning for intrinsic reasons or because they fear that the
sanctioned player will retaliate.20 Thus, unfair punishments are quite
unlikely even if they yield economic bene¬ts, while fair punishments
will occur even if they cause a net decrease in the punisher™s payoff.
Finally, although it is true that due to the ambiguity of most ¬eld situa-
tions it is not possible to unambiguously attribute the sanctions to non-
pecuniary motives, this does not mean that sanctions are automatically
driven by strategic reasons. In fact, we do not know of any rigorous
evidence that free-riders are punished for strategic reasons.
The lack of evidence in favor of strategic sanctions led Falk, Fehr,
and Fischbacher (2001) to examine this question. They conducted a
public goods experiment with a punishment opportunity in two condi-
tions. In the partner condition, three group members stay together for
six periods. In the perfect stranger condition, the game also lasts for
six periods, but it is ensured that nobody meets any of the other partic-
ipants more than once. Thus, in the partner condition, subjects can ben-
e¬t in economic terms from their punishments because the punished
group members typically raise their contributions in the following
periods; in the perfect stranger condition, no such bene¬ts can accrue.
If there are more sanctions in the partner condition, we have evidence
in favor of strategic sanctions. The results of this experiment are dis-
played in ¬gure 5.5.
Figure 5.5 shows the sanctioning behavior as a function of the devia-
tion of the contribution of the sanctioned subject from the contribution
of the sanctioning subject. It indicates that in the ¬rst ¬ve periods, the
sanctioning pattern and the strength of the sanctions are very similar
in both conditions. The sanctions in the partner condition are only
The Economics of Strong Reciprocity 173

Average number of punishment points


10% Perfect stranger, period 1-5
Partner, period 1-5

19% 15% 11% 8%
20% 41%
10% 6%
[-20,-14) [-14,-8) [-8,-2) [-2,2] (2,8] (8,14] (14,20]
Punished player's contribution - punisher's contribution

Figure 5.5
Punishment pattern in the public goods game”partners versus perfect strangers. Source:
Falk, Fehr, and Fischbacher 2001.

slightly stronger and the difference is not signi¬cant. Thus, the bulk of
the sanctions already exists when there are no pecuniary bene¬ts from
sanctioning”so there is little or no evidence in favor of strategic sanc-
tioning. Moreover, it turns out that in the ¬nal (sixth) period of the
partner treatment, the sanctions are even slightly higher than in the
previous ¬ve periods of this treatment.21 Since subjects know in ad-
vance that the experiment ends after period six, this result also indi-
cates a lack of evidence in favor of strategic sanctions. Although we do
not regard our experiment as the last word on this question, this evi-
dence should remind us that the mere fact that strategic punishments
can be part of an equilibrium does not yet mean that strategic punish-
ments will actually occur in the real world or in the laboratory.22
In view of the ubiquity of opportunities for free-riding, the existence
of a substantial amount of nonstrategic punishment of free-riders is
quite important. It suggests that even in one-shot situations or when
the discount factor is low, collusive practices in output and labor mar-
kets are much more likely than predicted by the self-interest hypothe-
sis. It also lends support to the insider-outsider theory of involuntary
unemployment developed by Lindbeck and Snower (1988). This theory
174 Fehr and Fischbacher

is based on the idea that the ¬rm™s existing workforce will harass out-
siders and will not cooperate with them if the outsiders are employed
below the going wage. Our evidence indicates that insiders will harass
outsiders even if this is costly for the insiders and brings them no eco-
nomic bene¬ts.

5.5 Economic Incentives and Property Rights

In this section, we show that the neglect of strong reciprocity prevents
the understanding of crucial determinants and effects of economic
incentives. We will show, in particular, that economic incentives may
reduce ef¬ciency in situations in which they are predicted to be ef¬-
ciency-enhancing by the self-interest model. In addition, we show that
strong reciprocity may have strong consequences for the optimal pro-
vision of incentives in a moral hazard context. Incentive contracts that
are optimal when there are only sel¬sh actors become inferior when
some agents prefer strong reciprocity. Conversely, contracts that are
doomed to fail when there are only sel¬sh actors provide powerful
incentives and become superior when there are also strongly reciprocal

5.5.1 Economic Incentives May Be Harmful
In the gift exchange game described earlier in the chapter, there are no
economic incentives to provide nonminimal effort levels. Despite this,
many responders (workers) put forward nonminimal effort levels in
case of fair wage offers. In reality, economic incentives are, of course,
also used to induce workers to provide high effort. How do explicit
performance incentives interact with motivations of fairness and strong
reciprocity? One possibility is that strong reciprocity gives rise to extra
effort on top of what is enforced by economic incentives alone. How-
ever, it is also possible that explicit incentives may cause a hostile at-
mosphere of threat and distrust, which reduces any reciprocity-based
extra effort. Bewley (1999, 431), for example, reports that many ˜˜man-
agers stress that punishment should seldom be used to obtain co-
operation™™ because of the negative effects on work atmosphere.
In a series of experiments, Fehr and Gachter (2000b) examined this
possibility. They implemented a baseline gift exchange game with a
slight modi¬cation. In addition to the wage, experimental employers
also stipulate a desired effort level. However, the desired effort repre-
sents merely ˜˜cheap talk™™”it is not binding for the workers. This
The Economics of Strong Reciprocity 175

means that workers still face no economic incentives in this treatment.
Fehr and Gachter (2000b) also implemented a treatment with explicit
performance incentives. This treatment keeps everything constant rela-
tive to the baseline treatment, except employers now have the possibil-
ity to ¬ne an employee in case of veri¬ed shirking. The probability of
veri¬cation is given by 0.33, and the ¬ne is restricted to an interval be-
tween zero and a maximum ¬ne. The maximum ¬ne is ¬xed at a level
such that a sel¬sh risk-neutral worker will choose an effort level of 4
when faced with this ¬ne.23
Figure 5.6 presents the results of this experiment. The line with the
black dots in ¬gure 5.6 shows workers™ effort behavior in the baseline
treatment. It depicts the average effort on the vertical axis as a function
of the rent offered to the workers. The offered rent is implied by the
original contract offer. It is de¬ned as the wage minus the cost of pro-
viding the desired effort level. Due to the presence of many reciprocal
workers, the average effort level is strongly increasing in the offered
rent and rises far above the sel¬sh level of e ¼ 1.
The line with the white dots in ¬gure 5.6 shows the relationship of
rent to effort in the presence of the explicit performance incentive. Ex-
cept at the low-rent levels, the average effort is lower in the presence of



Effort levels in contracts
with no explicit incentives
Average actual effort



Effort levels in contracts
with explicit incentives


0-5 6-10 11-15 16-20 21-25 26-30 31-35 36-40 41-45 46-50
Firms™ offered rents

Figure 5.6
Average effort levels and explicit incentives. Source: Fehr and Gachter 2000b.
176 Fehr and Fischbacher

the explicit incentives! This result suggests that reciprocity-based effort
elicitation and explicit performance incentives may indeed be in con-
¬‚ict with each other. Performance incentives that are perceived as hos-
tile can provoke hostile responses from workers. In the context of our
incentive treatment, this meant that strongly reciprocal workers were
no longer willing to provide nonminimal effort levels.24
In the experiments of Fehr and Gachter (2000b), the average effort
taken over all the different trades (and hence the aggregate monetary
surplus) is lower in the incentive treatment than in the baseline treat-
ment. However, employers™ pro¬ts are higher because in the incentive
treatment they infrequently rely on the carrot of generous wage offers.
Instead, they threaten the workers with the maximal ¬ne in most cases.
For the employers, the savings in wage costs more than offset the
reductions in revenues that are caused by the lower effort in the incen-
tive treatment. However, while the wage savings merely represent a
transfer from the workers to the ¬rms, the reduction in effort levels
reduces the aggregate surplus. This shows that in the presence of recip-
rocal types, ef¬ciency questions and questions of distribution are in-
separable. Since the perceived fairness of the distribution of the gains
from trade affects the effort behavior of the reciprocal types, different
distributions are associated with different levels of the aggregate gains.
Thus, lump-sum transfers between trading parties have ef¬ciency

5.5.2 Reciprocity-based Incentives Versus Explicit Incentives
Standard principal-agent models predict that contracts should be made
contingent on all veri¬able measures that are informative with regard
to the agent™s effort. But in reality, we often observe highly incomplete
contracts. For example, as noted earlier in the chapter, wages are often
paid without explicit performance incentives. On this point, the discus-
sion has focused on demonstrating that strong reciprocity has power-
ful economic effects in situations where explicit incentives are absent.
This section seeks to explore underlying causes for the absence of ex-
plicit incentives. Strong reciprocity plays a twofold role in this context.
First, as the previous experiment has shown, certain kinds of explicit
incentives have negative side effects because they reduce reciprocity-
based voluntary cooperation. Second, it renders contracts that do not
rely on explicit incentives more ef¬cient relative to the prediction of
the self-interest model because strong reciprocity itself constitutes a
powerful contract enforcement device. Each of these two reasons may
induce the principals to prefer contracts without explicit incentives.
The Economics of Strong Reciprocity 177

To study the impact of strong reciprocity on contractual choices,
Fehr, Klein, and Schmidt (2001) conducted an experiment in which
principals had the choice between an explicit incentive contract and an
implicit contract without explicit incentives. In a typical session of this
experiment, there are 12 principals and 12 agents who play for ten
periods. In each of the 10 periods, an agent faces a different principal,
which ensures that all matches are one-shot. A period consists of three
stages. At stage one, the principal has to decide whether to offer the
agent an implicit or an explicit contract. The implicit contract speci¬es
a ¬xed wage and a desired effort level (where effort choices can range
from 1 to 10). In addition, the principal can promise a bonus that may
be paid after the actual effort has been observed. In the implicit con-
tract, there is no contractual obligation to pay the announced bonus,
nor is the agent obliged to choose the desired effort level. The principal
is, however, committed to pay the wage. An explicit contract also
speci¬es a binding ¬xed wage and a desired effort level between 1 and
10. Here, however, the principal can impose a ¬ne on the agent that
has to be paid to the principal in case of veri¬ed shirking. Except for
one detail, the explicit contract is identical to the performance contract
discussed in the previous section. The difference concerns the fact that
the choice of the explicit contract involves a ¬xed veri¬cation cost of 10
units. This re¬‚ects the fact that the veri¬cation of effort is, in general,
costly. Note that the implicit contract does not require third-party veri-
¬cation of effort. It is only necessary that effort is observable by the
At stage two, the agent observes which contract has been offered
and decides whether to accept or reject the offer. If the agent rejects the
offer, the game ends and both parties get a payoff of zero. If the agent
accepts, the next step is for the agent to choose the actual level of effort.
At stage three, the principal observes the actual effort level. If the
principal has offered an implicit contract, the next decision is whether
the agent should be awarded the bonus payment. If the principal
offered an explicit contract and if the agent™s effort falls short of
the agreed effort level, a random draw decides with probability 0.33
whether shirking is veri¬able, in which case the agent has to pay the
If all players have purely sel¬sh preferences, the analysis of this
game is straightforward. A sel¬sh principal would never pay a bonus.
Anticipating this, there is no incentive for the agent to spend more than
the minimum effort. If the principal chooses the explicit contract, the
principal should go for the maximum punishment because this is the
178 Fehr and Fischbacher

best deterrence for potential shirkers. The parameters of the experi-
ment are chosen such that a risk-neutral and sel¬sh agent maximizes
expected utility by choosing an effort level of 4 if faced with the maxi-
mum ¬ne. Since the enforceable effort level is 4 under the explicit con-
tract, while it is only 1 under an implicit contract, the self-interest
model predicts that principals prefer the explicit contract.
The experimental evidence is completely at odds with these predic-
tions. In total, the implicit contract was chosen in 88 percent of the
cases. In view of the relative pro¬tability of the different contracts, the
popularity of the implicit contract is not surprising. Those principals
choosing the explicit contract made an average loss of 9 tokens per
contract, while those preferring the implicit contract made an average
pro¬t of 26 tokens per contract. Since the ¬xed veri¬cation cost in the
explicit contract was 10 tokens, the explicit contract would have
been much less pro¬table even in the absence of these costs. For both
contracts, the average income of the agents was roughly 18 tokens.
Implicit contracts were more pro¬table because”contrary to the stan-
dard prediction”they induced much higher effort levels. The effort
level in the implicit contract was 5.2 on the average (on a scale of 1 to
10), while the effort level in the explicit contract was 2.1 on the average.
How did implicit contracts induce much higher effort levels than
predicted? A major reason is that in the presence of strongly reciprocal
principals, the promised bonus does not merely represent cheap talk,
because reciprocal principals can”and actually do”condition the bo-
nus payment on the effort level. The average data clearly re¬‚ect this
impact of the reciprocal types because the actual average bonus rises
steeply with the actual effort level. The principal™s capability to commit
himself to paying a conditional bonus is based on his reciprocal incli-
nations. Conditional bonus payments, in turn, provide a strong pecuni-
ary incentive for the agents to perform as desired by the principals.
Why did explicit contracts induce lower effort levels than predicted? A
likely reason is that these contracts are perceived as hostile and even
induce negative reciprocity, as shown in the previous section.
One might also conjecture that the preference for implicit contracts
in this particular experiment is solely caused by the fact that the
explicit contract involves a punishment while the implicit contract
involves a reward. Further experiments by Fehr, Klein, and Schmidt
(2001), however, cast doubt on this explanation. If the implicit contract
described in this section competes with a piece-rate contract, the vast
majority of principals still prefer the implicit contract.
The Economics of Strong Reciprocity 179

5.5.3 Individual Versus Joint Ownership
The impact of strong reciprocity on contractual choices suggests that
it may not only cause substantial changes in the functioning of given
economic institutions, but that it may also have a powerful impact on
the selection and formation of institutions. To provide a further exam-
ple: The present theory of property rights (Hart 1995) predicts that
joint ownership will in general severely inhibit relations-speci¬c invest-
ments so that it emerges only under very restrictive conditions. This
may no longer be true in the presence of strongly reciprocal actors
who are willing to cooperate if they expect the trading partner to
cooperate as well and who are willing to punish even at a cost to
To illustrate this point, consider two parties, A and B, who are
engaged in a joint project (a ˜˜¬rm™™) to which they have to make some
relationship speci¬c investments today in order to generate a joint sur-
plus in the future. An important question that has received consider-
able attention in recent years is who should own the ¬rm. In a seminal
paper, Grossman and Hart (1986) argued that ownership rights allo-
cate residual rights of control to the physical assets that are required to
generate the surplus. For example, if A owns the ¬rm, then he will
have a stronger bargaining position than B in the renegotiation game
in which the surplus between the two parties is shared ex post facto, be-
cause he can exclude B from using the assets which make B™s relation-
ship speci¬c investment less productive. Grossman and Hart showed
that there is no ownership structure that implements ¬rst best invest-
ments, but some ownership structures do better than others, and there
is a unique second-best optimal allocation of ownership rights. They
also show that joint ownership is, in general, not optimal. This result is
at odds with the fact that there are many jointly-owned companies,
partnerships, or joint ventures. Furthermore, the argument neglects
that strong reciprocity may be an important enforcement mechanism
to induce the involved parties to invest more under joint ownership
than otherwise predicted.
In order to test this hypothesis, Fehr, Kremhelmer, and Schmidt
(2001) conducted a series of experiments on the optimal allocation of
ownership rights. The experimental game is a simpli¬ed version of
Grossman and Hart (1986): There are two parties, A and B, who have
to make investments, a; b A f1; . . . ; 10g, respectively, in order to gener-
ate a joint surplus v°a; bÞ. Investments are sequential: B has to invest
¬rst and her investment level b is observed by A, who has to invest
180 Fehr and Fischbacher

thereafter. We consider two possible ownership structures: Under A-
ownership, A hires B as an employee and pays her a ¬xed wage w. In
this case, monetary payoffs are v°a; bÞ À w À a for A and w À b for B.
Under joint ownership, each party gets half of the gross surplus minus
her investment cost” 1 v°a; bÞ À a for A and 1 v°a; bÞ À b for B. The gross
2 2
pro¬t function has been chosen such that maximal investments are
ef¬cient”that is, a FB ¼ b FB ¼ 10, but if each party gets only 50 percent
of the marginal return of his or her investment, then it is a dominant
strategy for a purely self-interested player to choose the minimum
level of investment, a ¼ b ¼ 1. Finally, in the ¬rst stage of the game, A
can decide whether to be the sole owner of the ¬rm and make a wage
offer to B, or whether to have joint ownership.
The prediction of the self-interest model is straightforward. Under
A-ownership, B has no incentive to invest and will choose b ¼ 1. On
the other hand, A is the residual claimant, so she will invest ef¬ciently.
Under joint ownership each party gets only 50 percent of the marginal
return, which is not suf¬cient to induce any investments. In this case,
B™s optimal investment level is unchanged, but A™s investment level is
reduced to a ¼ 1. Thus, A-ownership outperforms joint ownership and
A should hire B as an employee.
In the experiments just the opposite happens. Party A chooses joint
ownership in more than 80 percent (187 out of 230) of all observations
and gives away 50 percent of the gross return to B. Moreover, the frac-
tion of joint ownership contracts increases from 74 percent in the ¬rst
two periods to 89 percent in the last two periods. With joint ownership,
B players choose on the average an investment level of 8.9, and A
responds with an investment of 6.5 (on the average). On the other
hand, if A-ownership is chosen and A hires B as an employee, B™s aver-
age investment is only 1.3, while all A players choose an investment
level of 10. Furthermore, A players earn much more on the average if
they choose joint ownership rather than A-ownership.
These results are inconsistent with the self-interest model, but it is
straightforward to explain them with concerns for strong reciprocity.
Under joint ownership, the investments are associated with positive
externalities and, hence, joint ownership favors positively reciprocal
behavior. If, under joint ownership, B expects A to behave reciprocally,
even a sel¬sh B player has a strong incentive to make high investments
because this induces the reciprocal players A to invest, too. Under A-
ownership, the incentives for B are quite different because B does not
bene¬t from the investments of A. Hence, the sel¬sh Bs choose the
The Economics of Strong Reciprocity 181

minimal investment under A-ownership. If there is suf¬cient comple-
mentarity between the investments of A and B, the joint surplus is
therefore much higher under joint ownership. This makes it pro¬table
for A to choose joint ownership.

5.6 Proximate Models of Strong Reciprocity

The evidence from this chapter indicates that strong reciprocity has a
deep impact on fundamental economic issues. It is an important be-
havioral force that shapes the functioning of competition, governs the
laws of cooperation, and has a decisive impact on how incentives
work. Strong reciprocity creates implicit incentives and renders some
explicit incentives quite inef¬cient. By changing the incentives for the
sel¬sh types, it also affects the prevailing interaction patterns and con-
straints on individual behavior”the prevailing contracts and institu-
tions relative to a world that is exclusively populated by self-interested
We believe that”in view of the importance of strong reciprocity”
mainstream economics as well as the social sciences in general have
much to gain by routinely incorporating concerns for strong reciprocity
into the analysis. This means that when analyzing an economic or so-
cial problem, one should routinely try to derive the implications of
the assumption that, in addition to the purely self-interested types,
roughly 40 to 50 percent of the people exhibit strongly reciprocal pref-
erences. It is obvious that to achieve this a precise mathematical model
of strongly reciprocal preferences is desirable. In the past few years,
several authors have worked on models of strong reciprocity (Rabin
1993; Levine 1998; Dufwenberg and Kirchsteiger 2004; Falk and Fisch-
bacher, this volume, chapter 6; Segal and Sobel 1999; Charness and
Rabin 2002). These papers are very useful because they sharpen the no-
tion of strong reciprocity. However, they also show that it is extremely
dif¬cult to build simple and tractable models of strong reciprocity. The
problem is that the explicit modeling of intention-based or type-based
strong reciprocity quickly renders these models very complex and dif-
¬cult to handle.
The ¬rst best solution to the modeling problem would surely be a
simple and tractable model of strong reciprocity. However, since this
solution is not presently available, there is also a need for simpler
models that mimic the outcomes of strong reciprocity models in a
wide variety of circumstances but that do not explicitly model strong
182 Fehr and Fischbacher

reciprocity. Such models have been developed by Fehr and Schmidt
(1999), and Bolton and Ockenfels (2000). They are based on the assump-
tion that ˜˜fair™™ types dislike an inequitable distribution of economic
resources. The impressive feature of these models is that although they
are much simpler than strong reciprocity models, they correctly predict
the outcome of experiments in a wide variety of games. The model of
Fehr and Schmidt (1999), for example, is consistent with the stylized
facts in scenarios presented earlier in the chapter”the bilateral ulti-
matum and gift exchange game, market games under exogenous and
endogenous contract enforcement, cooperation games with and with-
out a punishment opportunity, and contract choice and property rights
experiments. This suggests that it is possible in many instances to
capture the behavioral predictions of strong reciprocity with simpler
models of fairness.
However, the black-boxing of strong reciprocity via simple fairness
models must be done with a background knowledge about the limits
of these models. Mindless application of these models may lead to
wrong predictions, as is demonstrated in the experiments of Brandts
and Sola (2001), and Falk, Fehr, and Fischbacher (2003). In the Falk,
Fehr, and Fischbacher (2003), paper, the rejection rates of the (8/2)-
offer (8 for the proposer and 2 for the responder) in four different
mini-ultimatum games are compared. The games differ only with re-
gard to the available alternative to the (8/2)-offer. In one game the al-
ternative was (5/5), in the second game it was (2/8), in the third game
it was (8/2), and in the last game it was (10/0). Note that if the res-
ponder cares only about the distribution of payoffs, the rejection rate
of the (8/2)-offer should be the same in all four games.
In fact, however, the rejection rate is monotonically declining across
the four games. It is highest in the (5/5)-game, where (5/5) was the al-
ternative, and lowest in the (10/0)-game. A plausible interpretation of
this result is that in the (5/5)-game an offer of (8/2) indicates an unfair
intention or an unfair type, while in the (10/0)-game this is not the
case. Thus, if responders punish unfair intentions or unfair types, they
should exhibit a higher rejection rate in the (5/5)-game. This example
indicates that if the set of feasible alternatives changes across situa-
tions such that the possibilities for expressing good or bad intentions
changes, simple fairness models do not capture important aspects of
It is, however, interesting that even in these situations, a simple
model can be useful because the prediction of the model provides hints
The Economics of Strong Reciprocity 183

when intention- or type-based strong reciprocity is likely to matter.
The prediction thus alerts the researcher about the limits of the model.
For instance, if (5/5) instead of (10/0) is the alternative to (8/2), the
Fehr-Schmidt model (1999) predicts that for reasonable rejection rates,
the population of proposers who make the (8/2)-offer is less fair. Thus
responders will make different inferences about the type or the inten-
tion of the population who made the (8/2)-offer when (5/5) is the al-
ternative compared to when (10/0) is the alternative. This inference
induces strongly reciprocal responders to reject the (8/2)-offer more
frequently when (5/5) is the alternative.26
It is also important to keep in mind that models that have been
developed to explain a diverse set of facts in laboratory experiments
must be used with care and need perhaps some adaptations when
applied to ¬eld situations. For example, it is often not possible to deter-
mine the relevant reference agents in the ¬eld without further em-
pirical analysis, while in an experiment, the set of the other players in
the group is often a good ¬rst approximation. Likewise, it does not
seem likely that the effort-relevant fairness judgements of a worker are
based on a comparison between the worker™s income and the in-
come of the top managers of the ¬rm. Instead, the behaviorally rele-
vant comparisons tend to be more local”that is, comparisons with her
coworkers or comparisons with the average value of the output she

5.7 Conclusion

The self-interest hypothesis assumes that all people are exclusively
motivated by their economic self-interest. This hypothesis is sometimes
a convenient simpli¬cation and there are, no doubt, situations in which
almost all people behave as if they were strictly self-interested. In
particular, for comparative static predictions of aggregate behavior,
self-interest models may make empirically correct predictions because
models with more complex motivational assumptions predict the same
outcome. However, the evidence presented in this paper also shows
that fundamental issues in economics and the social sciences in general
cannot be understood solely on the basis of the self-interest model. The
evidence indicates that concerns for fairness and strong reciprocity are
important for bilateral negotiations, for the functioning of markets and
incentives, for the structure of property rights and contracts, and for
the laws governing collective action and cooperation.
184 Fehr and Fischbacher


1. Economists and biologists de¬ned the term ˜˜reciprocity™™ in the past in different ways.
Biologists think of reciprocity, or ˜˜reciprocal altruism,™™ as tit-for-tat strategies in repeated
interactions (Trivers 1971; Axelrod and Hamilton 1981). Some economists (Binmore 1998)
use the term in a similar way. During the last ten years, however, an increasing number
of contributions show that reciprocal behavior also exists in sequentially structured one-
shot interactions. Reciprocity in one-shot interactions cannot be explained on the basis of
sel¬sh motives. Therefore, we use the term ˜˜strong reciprocity™™ to describe these non“
self-interested behaviors to distinguish them from reciprocal behaviour of self-interested
agents in a repeated game.
2. Strictly speaking, Levine™s model of reciprocity is not based on intentions, but on the
reciprocation to the other players™ preferences. A subject with Levine-type preferences
is more altruistic (or less spiteful) towards an altruistic player and more spiteful (or less
altruistic) towards a spiteful player. The model thus captures a kind of type-based
3. For evidence suggesting that fairness and reciprocity is important in the ¬eld see, for
example, Agell and Lundborg (1995), Bewley (1999), Frey and Weck-Hanneman (1984),
Frey and Pommerehne (1993), Greenberg (1990), Kahneman, Knetsch, and Thaler (1986),
Lind and Tyler (1988), Ostrom (1990, 2000), Seidl and Traub (1999), and Zajac (1995).
4. In the experiments, human subjects make decisions with real monetary consequences
in carefully controlled laboratory settings. In particular, the experimenter can implement
one-shot interactions between the subjects so that long-term self-interest can be ruled out
as an explanation for what we observe. As we will see, in some experiments the mone-
tary stakes involved are quite high”amounting up to the income of three months™ work.
In the experiments reviewed in this chapter, subjects do not know each other™s identities,
interact anonymously, and sometimes even the experimenter cannot observe their indi-
vidual choices. Due to the anonymity conditions, the laboratory environment is quite
unfavorable to the emergence of reciprocal behavior. Yet, if we observe reciprocal
behavior under such unfavorable conditions, it is even more likely to prevail in non-
anonymous interactions between people who know each other.
5. For surveys on ultimatum games, see Roth (1995) or Camerer (2003). Gift exchange
games have been conducted by scholars such as Brandts and Charness (forthcoming),
Charness (2000, forthcoming), Fehr and Falk (1999), Gachter and Falk (2002), Falk,
Gachter, and Kovacs (1999), and Hannan, Kagel, and Moser (2002).
6. The remaining subjects, except one, exhibit no signi¬cant change in the acceptance
threshold. Only 1 out of 70 subjects exhibits a signi¬cant decrease in the threshold rela-
tive to the baseline. Note that if a subject places a very high value on fairness, the accep-
tance threshold may already be very high in the baseline condition so that there is little
reason to change the threshold in the reputation condition. Identical thresholds across
conditions are, therefore, also compatible with a social preference approach. Only a de-
crease in the acceptance threshold is incompatible with theories of social preferences.
7. In the study of Roth et al. (1991), competition led to an even more extreme outcome.
However, in their market experiments, 9 competing proposers faced only 1 responder,
and the responder was forced to accept the highest offer.
8. In the meantime, the gift exchange game has been framed in goods market terms, la-
bor market terms, and in a completely neutral language. The results indicate that there
are no framing effects.
The Economics of Strong Reciprocity 185

9. In each period the same stationary situation is implemented”there are 12 workers,
8 ¬rms, and each worker™s reservation wage is 20. In a given period, employers and
workers can make as many wage bids as they like, as long as they have not yet been
signed on. Trading is anonymous. Every worker can accept an offer made by a ¬rm and
every ¬rm can accept an offer made by a worker.

10. A variety of studies have found that one major reason why managers are reluctant to
cut wages in a recession is the fear that wage cuts may hamper work performance.
Among others, Bewley (1999, this volume, chapter 11) reports that managers are afraid
that pay cuts ˜˜express hostility to the work force™™ and will be ˜˜interpreted as an insult.™™
For similar results see Agell and Lundborg (1995), and Campbell and Kamlani (1997).
11. For the severe dif¬culties created by unobservable heterogeneity in this context, see
Murphy and Topel (1990), and Gibbons and Katz (1992).

12. This situation mimics a classic exchange problem in the absence of exogenous con-
tract enforcement. A would like to have the good that B possesses, because she values
that good more than B does and vice versa. Since A and B cannot write a contract that is
enforced by a third party and since both have to send their goods simultaneously to the
other person, they have a strong incentive to cheat.
13. For rigorous proofs that reciprocity (or inequity aversion) transform the PD into a co-
ordination game, see section IV in Fehr and Schmidt (1999).

14. Social dilemma games are generalised PD-games in the following sense: There is a
Pareto-superior cooperative outcome that renders everybody strictly better off relative to
the Nash equilibrium.
15. Social psychologists have also found evidence that people who believe that others
cooperate more will themselves cooperate more (Dawes 1980; Messick and Brewer 1983;
Hayashi et al. 1999). Some of them have interpreted this in terms of a false consensus
effect. According to the false consensus effect, the causality runs from a subject™s cooper-
ativeness to the subject™s belief that others cooperate. However, the evidence in Fisch-
bacher, Gachter and Fehr (2001) and in Hayashi et al. (1999) shows that the causality
goes the other way round: If the other players contribute more strongly, reciprocal sub-
jects contribute more on average.

16. The of¬cial title of former president Clinton™s reform initiative”˜˜The Personal Re-
sponsibility and Work Opportunity Reconciliation Act™™”is telling in this regard.
17. The written instructions for the subjects do not use value-laden terms such as ˜˜pun-
ishment points.™™ Instead, the instructions are framed in neutral terms. For example, sub-
jects do not assign ˜˜punishment points™™ but just ˜˜points™™ to the other players.
18. In the experiments, subjects ¬rst participate in the game without a punishment op-
portunity for ten periods. After this, they are told that a new experiment takes place. In
the new experiment, which lasts again for ten periods, the punishment opportunity is
implemented. In both conditions, subjects remain in the same group for ten periods and
they know that after ten periods the experiment will be over.
19. Francis™ (1985, 269) description of social ostracism in communities of the British
miners provides a particularly vivid example. During the 1984 miners™ strike, which
lasted for several months, he observed the following: ˜˜To isolate those who supported
the ˜scab union,™ cinemas and shops were boycotted, there were expulsions from football
teams, bands, and choirs, and ˜scabs™ were compelled to sing on their own in their chapel
services. ˜Scabs™ witnessed their own ˜death™ in communities which no longer accepted
186 Fehr and Fischbacher

20. Suppose we offer you £100 for hitting a stranger in the face. Even if the stranger had
no possibility to hit back, most people would probably reject this offer.
21. A plausible reason for this is that if subjects cooperate successfully for ¬ve periods
and then some group members try to cheat (free-ride) in the ¬nal period, the cooperators
may be more angry than when they face free-riding in earlier periods. Being cheated by a
˜˜friend™™ might make people angrier than being cheated by a ˜˜stranger.™™
22. There is an interesting difference between the ultimatum game experiments with rep-
utation formation discussed in this chapter and the punishment of free-riders in the part-
ner treatment. Recall that responder™s acceptance thresholds were signi¬cantly higher in
the reputation treatment of the ultimatum game relative to the baseline treatment. In the
reputation treatment, a responder could aqcuire an individual reputation for being a
tough bargainer and he could reap the full bene¬ts of his reputation. In the partner treat-
ment of the public goods game, the punishment of free-riders constitutes a second-order
public good because all group members bene¬t from the cooperation-enhancing effect of
the punishment. This may be one reason why we observe so little strategic punishment in
the partner treatment.

23. To prevent hostility from being introduced merely by the use of value-laden terms,
we avoided terms like ˜˜¬ne,™™ ˜˜performance,™™ and so forth. Instead we used a rather neu-
tral language”for example, ˜˜price deduction.™™
24. Note that according to this interpretation, there is no crowding out of an intrinsic
(reciprocal) motivation here. Instead, the preference for reciprocity implies that workers
respond in a hostile manner to incentives that are perceived as hostile.
25. Employers are, in general, not free to cut a worker™s wage for shirking, while they
have few legal problems when they refuse to pay a promised bonus.
26. One way to explain this evidence is to modify the Fehr-Schmidt (1999) model of in-
equity aversion such that the disutility from disadvantageous inequality is lower if a
person faces a subject with a high preference against advantageous inequality. This
basically boils down to a type-based model of reciprocity. The model by Falk and
Fischbacher (this volume, chapter 6) can also explain this evidence.


Agell, Jonas, and Per Lundborg. 1995. ˜˜Theories of Pay and Unemployment: Survey Evi-
dence from Swedish Manufacturing Firms.™™ Scandinavian Journal of Economics 97:295“308.
Akerlof, George. 1982. ˜˜Labor Contracts as Partial Gift Exchange.™™ Quarterly Journal of
Economics 97:543“569.
Andreoni, James. 1989. ˜˜Giving with Impure Altruism: Applications to Charity and
Ricardian Equivalence.™™ Journal of Political Economy 97:1447“1458.
Andreoni, James, and John Miller. 2002. ˜˜Giving According to GARP: An Experimental
Test of the Rationality of Altruism.™™ Econometrica 70:737“753.
Axelrod, Robert, and William D. Hamilton. 1981. ˜˜The Evolution of Cooperation.™™ Science
Bewley, Truman. 1999. Why Wages Don™t Fall During a Recession. Cambridge, MA: Har-
vard University Press.
The Economics of Strong Reciprocity 187

Bewley, Truman. 2003. ˜˜Fairness, Reciprocity, and Wage Rigidity,™™ chapter 11, this
Binmore, Ken. 1998. Game Theory and the Social Contract: Just Playing. Cambridge, MA:
MIT Press.
Bolton, Gary E., and Axel Ockenfels. 2000. ˜˜A Theory of Equity, Reciprocity and Compe-
tition.™™ American Economic Review 100:166“193.
Bowles, Samuel, and Herbert Gintis. 1998. ˜˜Is Equality Passe.™™ Boston Review 23(6).
Brandts, Jordi, and Gary Charness. Forthcoming. ˜˜Do Labor Market Condition Affect
Gift-Exchange?™™ Economic Journal.
Brandts, Jordi, and Carles Sola. 2001. ˜˜Reference Points and Negative Reciprocity in Sim-
ple Sequential Games.™™ Games and Economic Behavior 36:138“157.
Camerer, Colin F. 2003. Behavioral Game Theory. Princeton, NJ: Princeton University Press.
Cameron, Lisa. 1999. ˜˜Raising the Stakes in the Ultimatum Game: Experimental Evidence
from Indonesia.™™ Economic Inquiry 37:47“59.
Campbell, Carl M., and Kunal S. Kamlani. 1997. ˜˜The Reasons for Wage Rigidity: Evi-
dence from a Survey of Firms.™™ Quarterly Journal of Economics 112:759“789.
Charness, Gary. 2000. ˜˜Responsibility and Effort in an Experimental Labor Market.™™ Jour-
nal of Economic Behavior and Organization 42:375“384.
Charness, Gary. Forthcoming. ˜˜Attribution and Reciprocity in an Experimental Labor
Market.™™ Journal of Labor Economics.
Charness, Gary, and Matthew Rabin. 2002. ˜˜Understanding Social Preferences with Sim-
ple Tests.™™ Quarterly Journal of Economics 117:817“869.
Cox, Jim. 2000. ˜˜Trust and Reciprocity: Implications of Game Triads and Social Con-
texts.™™ Mimeo. Tucson: University of Arizona.
Cox, Jim, Klarita Sadiraj, and Vjollca Sadiraj. 2001. ˜˜A Theory of Fairness and Competi-
tion without Inequality Aversion.™™ Mimeo.
Croson, Rachel T. A. 1999. ˜˜Contributions to Public Goods: Altruism or Reciprocity.™™ Dis-
cussion Paper. Wharton School of Business, University of Pennsylvania.

Davis, Douglas D., and Charles A. Holt. 1993. Experimental Economics. Princeton, NJ:
Princeton University Press.
Dawes, Robin. 1980. ˜˜Social Dilemmas.™™ Annual Review of Psychology 31:169“193.

Dufwenberg, Martin, and Georg Kirchsteiger. 2004. ˜˜A Theory of Sequential Reciprocity.™™
Games and Economic Behavior 47:268“298.

Falk, Armin, and Urs Fischbacher. 2003. ˜˜Modelling Strong Reciprocity,™™ chapter 6, this
Falk, Armin, Ernst Fehr, and Urs Fischbacher. 2000. ˜˜Testing Theories of Fairness”
Intentions Matter.™™ Working Paper No. 63. Institute for Empirical Research in Economics,
University of Zurich.
Falk, Armin, Ernst Fehr, and Urs Fischbacher. 2001. ˜˜Driving Forces of Informal Sanc-
tions.™™ Institute for Empirical Research in Economics, University of Zurich, Working
Paper No. 59.
188 Fehr and Fischbacher


. 6
( 13)