Eyes are on us, but nobody cares: are eye
cues relevant for strong reciprocity?
Ernst Fehr and Fre´de´ric Schneider*
Institute for Empirical Research in Economics, University of Zu¨rich, Blu¨mlisalpstrasse 10,
8057 Zu¨rich, Switzerland
Strong reciprocity is characterized by the willingness to altruistically reward cooperative acts and to
altruistically punish norm-violating, defecting behaviours. Recent evidence suggests that subtle reputation
cues, such as eyes staring at subjects during their choices, may enhance prosocial behaviour. Thus, in
principle, strong reciprocity could also be affected by eye cues. We investigate the impact of eye cues
on trustees’ altruistic behaviour in a trust game and find zero effect. Neither the subjects who are classified
as prosocial nor the subjects who are classified as selfish respond to these cues. In sharp contrast to
the irrelevance of subtle reputation cues for strong reciprocity, we find a large effect of explicit, pecuniary
reputation incentives on the trustees’ prosociality. Trustees who can acquire a good reputation that
benefits them in future interactions honour trust much more than trustees who cannot build a good reputation.
These results cast doubt on hypotheses suggesting that strong reciprocity is easily malleable by
implicit reputation cues not backed by explicit reputation incentives.
Keywords: altruism; strong reciprocity; trust game; cues; reputation
1. INTRODUCTION
Human altruism represents a huge outlier in the animal
world (Boyd & Richerson 2005). Humans often behave
altruistically towards genetically unrelated strangers,
even if the chance of meeting these strangers again is
extremely small and reputational concerns are unlikely
to play a role (e.g. tipping an unknown taxi driver in a
large foreign city). Altruistic behaviours in the absence
of any opportunity of repeated interaction and reputation
formation have been repeatedly shown in tightly controlled
economic experiments (Camerer 2003; Fehr &
Fischbacher 2003; Gintis et al. 2003). Experimental evidence
(Fehr et al. 2002), social preference theories
(Rabin 1993; Fehr & Schmidt 1999; Dufwenberg &
Kirchsteiger 2004; Falk & Fischbacher 2006) and evolutionary
theories (Gintis 2000; Henrich & Boyd 2001;
Boyd et al. 2003; Bowles & Gintis 2004) also indicate
that a special type of altruistic behaviour—strong reciprocity—plays
a particularly important role in establishing
and sustaining cooperation among strangers. Strong reciprocity
is characterized by the willingness to altruistically
reward cooperative acts and to altruistically punish normviolating,
defecting behaviours. As a consequence, strong
reciprocity generates important incentives for cooperation
among strangers.
The fact that important altruistic behaviours exist even
in the absence of reputation incentives does not mean that
such incentives are irrelevant. A large literature shows that
human cooperation and other forms of prosocial behaviour
are positively affected by the possibility of acquiring
a ‘good’ reputation that may pay off in future interactions
(Ga¨chter & Fehr 1999; Wedekind & Milinski 2000;
Milinski et al. 2001, 2002; Brown et al. 2004; Rege &
Telle 2004; Rockenbach & Milinski 2006; Kurzban
et al. 2007; Engelmann & Fischbacher 2009; Fehr et al.
2009). However, recent articles seem to suggest that
much of human altruistic behaviour may merely be a
response to subtle reputation cues that are not in fact
related to the possibility of benefiting in future interactions
from current altruistic acts (Haley & Fessler
2005; Bateson et al. 2006; Hagen & Hammerstein
2006; Burnham & Hare 2007; Rigdon et al. 2009).
Haley and Fessler argue that reputation incentives in the
ancestral evolutionary environment thoroughly moulded
human social interactions because ‘natural selection can
be expected to have shaped human psychology to be
exquisitely sensitive to cues that are (or were, under
ancestral conditions) informative with respect to the
likely profitability of co-operation in a given situation’
(Haley & Fessler 2005, p. 249). These authors thus
implemented a visual cue in an anonymous experimental
game—eyes staring at the subjects during decisionmaking—a
cue ‘that, over the course of human evolution,
would have reliably indicated the potential observability
of one’s behaviour’ (p. 249). Haley & Fessler (2005),
Bateson et al. (2006), Burnham & Hare (2007) and
Rigdon et al. (2009) indeed found that eyes staring at
the subjects cause an increase in prosocial behaviours in
anonymous games such as the dictator game.
In this paper we examine whether a reputation cue like
that implemented in Haley & Fessler (2005) also affects
strong reciprocity, by implementing an anonymous, oneshot
trust game in which a trustor can send money to a
trustee; the experimenter then quadruples this amount,
so that the trustee receives four times the amount sent.
The trustee observes how much the trustor has sent and
can then send back as little or as much money as he
wants. Thus, the trustee can altruistically reward trustors
who have sent money, which constitutes an instance of
* Author for correspondence (frederic@iew.uzh.ch).
Electronic supplementary material is available at http://dx.doi.org/10.
1098/rspb.2009.1900 or via http://rspb.royalsocietypublishing.org.
Proc. R. Soc. B (2010) 277, 1315–1323
doi:10.1098/rspb.2009.1900
Published online 23 December 2009
Received 3 November 2009
Accepted 4 December 2009 1315 This journal is q 2009 The Royal Society
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
strong reciprocity. By comparing the eye cue condition
with a baseline condition without such cues we can
assess the impact of eye cues on strong reciprocity.
In addition to the eye cue condition we implement
another reputation condition in which subjects face a
real pecuniary incentive for acquiring a good reputation.
Previous work has argued that eye cues activate reputational
concerns, but has not explicitly compared the
effect of eye cues with the effect of explicit pecuniary
reputation incentives. If humans are indeed ‘exquisitely
sensitive’ to reputation cues even if they carry no real
pecuniary incentive power, subjects should generate patterns
in the eye cue condition that resemble the effects
of explicit pecuniary reputation incentives. Our design
enables us to conduct this comparison and investigate
the relative importance of reputation cues for altruistic
behaviour.
We also go beyond previous work by examining
which—if any—subjects respond to the implicit reputation
cue, because we measure subjects’ degree of
selfishness and opportunism with a Machiavellianism
questionnaire (Christie & Geis 1970). Assessing individual
differences in subjects’ responses to reputation cues
is important because on average one might find a null
effect that hides important inter-individual differences.
Implicit reputation cues could increase the altruistic
behaviour of prosocial subjects (i.e. those who score low
on the Machiavellianism (Mach) scale). This has important
consequences on the interpretation of altruistic
behaviours in anonymous one-shot experiments. If prosocial
subjects primarily respond to the implicit reputation
cue, it is possible to argue that they are mostly prone to
all kinds of other subtle reputation cues that are often
not controlled for by the experimenter in the typical
laboratory experiment (e.g. the mere presence of other
subjects and the experimenter in the room, or simply
hearing human voices). It would then be more plausible
to attribute the baseline altruism observed in anonymous
one-shot experiments to such uncontrolled reputation
cues. However, if prosocial subjects do not respond to
the eye cues, it is hard to argue along these lines. It is
then implausible to attribute the observed altruistic
behaviours in anonymous one-shot experiments to uncontrolled
subtle reputation cues. Thus, by measuring
subjects’ Mach scores we can put important constraints
on the interpretation of altruistic behaviours in anonymous
one-shot experiments.
2. THE EXPERIMENT
(a) Experimental procedure
We measured strong reciprocity as second-mover behaviour
in a series of one-shot trust games. A trustor and a
trustee interact with each other in a trust game; the trustor
can send money to the trustee, which is then multiplied by
the experimenter so that the overall money available to the
two parties increases. The trustee can then send back
none or some of the money to the trustor. Fairness
norms typically demand that the trustee sends back
some of the money he received, but the trustee is completely
free to send back nothing if he likes. Details of the
trust game are described in §2b below.
Our experimental design includes three treatments: a
baseline treatment where the trustee faces a neutral
background screen (figure 1a); an ‘implicit reputation
cue’ treatment where the background screen features eyespot-like
shapes (figure 1b) similar to those in Haley &
Fessler (2005); and an ‘explicit reputation’ treatment
where the current trustor is informed of the trustee’s
decisions in the previous periods (same background
screen as in the baseline treatment).
Subjects were seated in isolated compartments and
were assigned either to the role of a trustor or a trustee.
They maintained their roles throughout the experiment.
They then played 10 periods of the trust game, with
randomly re-matched partners each period.
We conducted eight sessions (three sessions each in
baseline and implicit, two in explicit treatment), each session
involving 36 participants, hence 288 participants in
total (144 trustors and 144 trustees). The decisions in a
group of subjects who interact with each other over the
10 periods are statistically not independent. In order to
establish statistically independent observations, we created
three matching groups per session, each consisting
of 12 subjects. Only the subjects within a matching
group were matched with each other during the experiment,
generating three independent observations per
session. With three matching groups per session, we
have nine independent observations both in the baseline
treatment and the implicit cues treatment, and six independent
observations in the explicit reputation treatment.
Immediately after the end of the last period, the participants
had to fill out a questionnaire (emotional state,
fairness attitudes, Machiavellianism, trust, socioeconomic
period
1 of 1 remaining time (s):
period
1 of 1 remaining time (s):
your initial endowment:
transfer from participant B:
you have 38 points at your disposal.
your transfer to participant A: transfer nothing
transfer 1 quarter
equal amount of points
0
0
7
OK
OK
10
your initial endowment:
transfer from participant B:
you have 38 points at your disposal.
your transfer to participant A: transfer nothing
transfer 1 quarter
equal amount of points
7
10
(a)
(b)
Figure 1. Trustee decision screen. (a) Baseline background.
(b) Eyespots background.
1316 E. Fehr & F. Schneider Eye cues and strong reciprocity
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
data). After completion, participants were paid a show-up
fee of CHF 10 plus their earnings from the experiment, at
the rate of 1 point ¼ CHF 0.2. In total, a session lasted
approximately 2 h and subjects earned on an average
CHF 48.88 ($41.77).
(b) Game design
Each period of the experiment was a one-shot trust game.
At the beginning of each period trustors and trustees were
endowed with 10 points. The game itself consisted of two
stages: an investment stage, where trustors had to decide
how many points they would transfer to their current trustee,
and a back-transfer stage, where trustees had to decide
how much they wanted to back-transfer to the trustor.
The amount trustors invested was quadrupled and transferred
to the trustee. Trustors could choose between
four possible transfers: 1 point, 4 points, 7 points, or
10 points. Trustees had three options: they could backtransfer
either nothing, or the amount the trustor sent
(henceforth ‘compensate’), or they could back-transfer
an amount that equalized the period payoff between trustor
and trustee (henceforth ‘equalize’). When the trustee
determined the back-transfer he was perfectly informed
about the trustor’s choice and thus did not have to form
beliefs about the size of the investment.
In table 1 we show the payoff matrix that corresponds
to our trust game. The first number in each cell of the
matrix represents the trustor’s payoff, the second
number denotes the trustee’s payoff. The matrix shows
that for any given investment level, the trustee is always
best off in terms of monetary payoff by back-transferring
nothing. This means that positive back-transfers (i.e. the
choices ‘compensate’ and ‘equalize’) can be regarded as
altruistic acts because the trustee gives up some of his
own payoff to increase the trustor’s payoff.
(c) Personality questionnaires and
fairness norms
As the response to the different treatments may be heterogeneous
depending on the subject’s degree of selfishness,
we also measured each subject’s Mach score. For this
purpose we used the MACH-IV Machiavellianism
Questionnaire (Christie & Geis 1970), which provides a
measure of selfishness and opportunism. Recent results
from a neuroeconomic study (Spitzer et al. 2007) indicate
that Machiavellian subjects are much less willing to share
money in a dictator game, and respond much more
strongly to pecuniary punishment threats for norm violations.
Moreover, subjects’ Machiavellianism also
correlated strongly with activation in the lateral orbitofrontal
cortex that is known to be reliably activated
when subjects face punishing stimuli. Thus, behavioural
and neurophysiological evidence suggests that subjects’
Machiavellianism may affect their responses to our
treatment conditions.
We also measured subjects’ fairness standards by
asking them the following question: ‘suppose that participant
A (i.e. the trustor) transferred 10 points to
participant B (i.e. the trustee). B then chose “compensate”.
How fairly do you judge this behaviour?’
(Additions in brackets did not appear in the questionnaire.)
Subjects indicated their answer to this question
on a Likert scale, coded from 1 (‘very unfair’) to 7
(‘very fair’). Note that subjects with high fairness standards
perceive the choice as unfair and therefore assign
a low score to this question, while subjects with low fairness
standards perceive the choice as fair and assign a
high score.
3. HYPOTHESES
The implicit cues treatment measures the impact of
implicit reputation cues on trustees’ altruistic behaviour
in the trust game. The explicit reputation treatment
enables us to assess the effect of explicit pecuniary reputation
incentives on trustees’ behaviour. Thus, we can
gain insight into the relative importance of the two kinds
of reputation effects by comparing the effect of implicit
cues with the effect of explicit reputation incentives.
Consider the baseline and the implicit reputation treatment
condition first. The game played in these two
conditions constitutes a true one-shot game because the
players remain fully anonymous and they meet a different
anonymous partner in each period. Therefore, if both
players are completely selfish and want to maximize
their money earnings, and the trustor knows this, the following
outcome is predicted. The selfish trustee will
always choose ‘nothing’ (i.e. his back-transfer is zero),
and the trustor will invest the lowest possible amount,
because he knows that the trustee will back-transfer
nothing in any case.
However, there is a large literature indicating that a
substantial share of experimental subjects is not completely
selfish (see Fehr & Fischbacher 2003 for a review).
This literature indicates that subjects may also have
social motives such as inequity aversion (Fehr & Schmidt
1999; Dawes et al. 2007) or intention-based reciprocity
(Rabin 1993; Dufwenberg & Kirchsteiger 2004; Falk &
Fischbacher 2006). Inequity-averse trustees will choose
the ‘equalize’ option, while trustees who interpret high
investments as particularly kind acts will make more
generous back-transfers in response to high investments.
We summarize both these behaviours under the term
‘altruistic rewarding’ because they imply a benefit for
the trustor at the expense of the trustee and they reward
the trustor’s cooperative investment.
A key question then is whether subjects’ social preferences
are affected by implicit reputation cues such as
eyespots. Recent evidence (Haley & Fessler 2005;
Bateson et al. 2006; Burnham & Hare 2007) suggests
that eye cues affect prosocial behaviour in dictator
Table 1. This table shows the monetary payoffs that are
associated with each action combination. The first number
in each cell denotes the trustor’s payoff while the second
number denotes the trustee’s payoff. For example, if the
trustor invests 7 points and the trustee ‘compensates’, the
trustor earns 10 points and the trustee earns 31 points.
trustee
trustor nothing compensate equalize
1 point 9 14 10 13 11.5 11.5
4 points 6 26 10 22 16 16
7 points 3 38 10 31 20.5 20.5
10 points 0 50 10 40 25 25
Eye cues and strong reciprocity E. Fehr & F. Schneider 1317
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
games and public good games. In view of this literature,
one would expect the trustees to respond to the eyespots
in the implicit reputation treatment by making significantly
higher back-transfers compared with the baseline
treatment.
As we are interested in the impact of the implicit reputation
cue on the trustees’ social preferences, the fact that
the trustee chooses his back-transfer with the exact
knowledge of how much the trustor invested is important.
This feature of our design ensures that unknown beliefs
about the trustors’ investments do not affect the trustees’
choices. In this respect, our design differs substantially
from the public goods experiments of Bateson et al.
(2006) and Burnham & Hare (2007), because it is not
clear why subjects change their contributions in response
to a cue in a public goods experiment. In principle, the
cue could cause a more optimistic belief about the other
players’ public good contributions, which will then lead
to an increase in the subject’s own contribution; it is
known that many subjects are conditional cooperators
(Fischbacher et al. 2001; Kurzban & Houser 2005;
Croson 2007; Kocher et al. 2008)—that is, they are willing
to contribute more to the public good if they believe
that other group members contribute more. Alternatively,
the reputation cue could have a direct impact on subjects’
social preferences, implying that subjects are willing to
contribute more for any given belief level. If the first
hypothesis holds, the reputation cue does not affect subjects’
social preferences; it ‘only’ renders their beliefs
about others more optimistic, which then causes the
change in behaviour. If the second hypothesis holds,
the reputation cue has a direct effect on subjects’ social
preferences. In our experimental design a change in the
trustees’ behaviour cannot be attributed to changes in
their beliefs because the trustees know the exact investment
when they make their back-transfer. Thus, we can
measure the impact of the implicit reputation cue for any
given transfer level, which provides a clean behavioural
measure of a change in social preferences.
Because we measure subjects’ degree of Machiavellianism
and their fairness standards, we are able to examine
whether subjects who score differently on these measures
respond differently in the different treatments. We expect,
in particular, that highly Machiavellian subjects tend to
back-transfer less in the baseline condition. It is also
important to examine the impact of the implicit cue condition
for subjects who score high and low on the Mach
score. In particular, if the non-selfish subjects (i.e. those
scoring low on the Mach score) are particularly responsive
to the implicit reputation cue, one may be more inclined
to attribute the observed prosociality in anonymous oneshot
experiments to uncontrolled implicit reputational
features. In contrast, if subjects’ Mach scores do not
affect the response to the implicit cue, one may have
more confidence in the hypothesis that the prosocial
behaviour in anonymous experiments is a true expression
of subjects’ social preferences and not just an artefact of
uncontrolled implicit reputation cues.
In the explicit reputation treatment, the players’ personal
identities are still kept anonymous but we render
the history of the trustees’ back-transfers observable for
their current trustors. Thus, each trustor can assess the
past willingness of the current trustee to back-transfer
resources. Because the trustees know this, even selfish
trustees now have an incentive to choose ‘compensate’
or ‘equalize’, because in this way they can increase the
likelihood that the trustors they face in the future (and
know their past choices) will make large investments.
This explicit reputation incentive ceases in the final
period (when there will be no future encounters with trustors):
the selfish trustees will defect in the last period, and
only the trustees with social preferences will make positive
back-transfers.
The effectiveness of the explicit reputation incentive
requires that the trustees understand that their current
back-transfers will affect average investments of future
trustors. Thus, the explicit reputation incentive will only
raise the trustees’ back-transfers if the trustees exhibit
this kind of rationality. Reputation incentives can also
increase back-transfers of subjects with social preferences.
They may, for example, choose ‘equalize’ instead of only
‘compensate’ when the pecuniary incentive coincides with
their social motive. The hypothesis that explicit reputation
incentives increase trustees’ transfers is also
backed by previous findings (Ga¨chter & Falk 2002;
Cochard et al. 2004).
Our measure of Machiavellianism enables us to examine
whether there is a meaningful heterogeneity in
trustees’ responses to the explicit reputation incentive.
In view of the behavioural and neurophysiological
evidence documented in Spitzer et al. (2007), it seems
plausible to conjecture that highly Machiavellian subjects
respond more strongly to the explicit reputation incentive.
Future trustors are likely to punish low back-transfers by
lowering their investments. By definition, highly Machiavellian
subjects are particularly susceptible to such
threats. Therefore, they should respond more strongly
to the pecuniary reputation incentives.
4. DOES THE IMPLICIT REPUTATION CUE MATTER?
In this section, we examine the impact of implicit reputation
cues on trustees’ back-transfers. If the implicit
cue raises reputational concerns, the trustees in the
implicit cue condition should make higher back-transfers
than those in the baseline condition. Moreover, if the
implicit cue has a sufficiently strong effect, the backtransfer
pattern in the implicit cue condition should
resemble the pattern in the explicit reputation condition.
Finally, if the implicit reputation cue raises the trustees’
back-transfers, this may also increase the trustors’ investments
because higher investments increase the trustors’
payoffs if a sufficiently high share of trustees choose to
equalize payoffs (see the final column in table 1).
Table 2 provides a first indication of the impact of the
implicit cue condition. In the baseline condition, the average
back-transfer is 6.28 points and the trustees’ modal
choice is ‘nothing’. The average back-transfer in the
implicit cue condition is even somewhat lower and the
modal choice is also ‘nothing’. The small difference in
the means across conditions is not significant (Mann–
Whitney test: p ¼ 0.402, n ¼ 18). The trustors’
investment choices are also very similar across the two
conditions. The median investment level in both conditions
is 7; trustors in the baseline condition invest an
average of 5.88, while the average investment in the
implicit cue condition is 5.74 (Mann–Whitney test:
p ¼ 0.825, n ¼ 18).
1318 E. Fehr & F. Schneider Eye cues and strong reciprocity
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
Figure 2a shows the time path of average back-transfers.
The figure indicates that the average back-transfer varies
between 5 and 8 units in both the baseline condition
and the implicit cue condition, with little difference
between the conditions. The figure also displays the
s.e.’s (clustered on matching groups) of the mean,
which indicate that the differences between baseline and
implicit cue conditions are not significant (Mann–Whitney
test: p ! 0.272, Holm-Bonferroni correction for
multiple comparisons).
Thus, table 2 and figure 2a provide little indication
that the implicit cue condition increased average backtransfers.
An examination of the impact of implicit cues
in more detail requires further control for the investments
that the trustees face. Figure 2b shows the trustees’ average
back-transfer conditional on the received investments.
On an average, trustees in the implicit cues condition sent
back the same or a slightly smaller amount than in the
baseline condition for any given investment level. In
table 3 we report the results of ordinary least squares
regressions with the average relative back-transfer as the
dependent variable. The relative back-transfer is defined
as the share of points returned over the points received
(i.e. the quadrupled investment). Thus, a choice of ‘nothing’
implies a relative back-transfer of 0 per cent, the
choice of ‘compensate’ translates into a relative backtransfer
of 25 per cent and the choice of ‘equalize’
means that 62.5 per cent of the received points are sent
back by the trustee. For example, with an investment
level of 4, the trustee receives 4 Â 4 ¼ 16 points and
sends back 10 if he chooses ‘equalize’, giving a payoff of
16 to each of the two players; the relative back-transfer
equals 10/16 ¼ 62.5 per cent. Likewise, if the trustor
sends 10, the trustee receives 40 and sends back 25 in
the case of ‘equalize’, which yields a relative back-transfer
of 25/40 ¼ 62.5 per cent. The advantage of using relative
back-transfers is that a given choice, such as ‘equalize’,
implies the same percentage number regardless of the
investment level. Thus, our regressions implicitly estimate
the conditional frequency of the three choices—‘nothing’,
‘compensate’ and ‘equalized’.
Model (1) in table 3 reports the result of a regression
that takes the average relative back-transfer per matching
group as the dependent variable. The independent variables
in this regression are the average investment per
matching group, dummy variables for the implicit cue
and the explicit reputation treatment, the average Mach
score of the trustees in the matching group, and the average
response to the fairness question (high answer
indicates low fairness norm). In all regressions, the
omitted category is the baseline dummy, implying that
the constant measures the average relative back-transfer
in the baseline condition, while the dummy for the
implicit cue (explicit reputation) condition measures the
difference between the baseline condition and the implicit
cue (explicit reputation) condition.
Regression (1) is the most conservative because the
unit of observation is average behaviour in a matching
group, giving us ‘only’ 24 observations in total. We find
a highly significant positive effect of the investment
level, i.e. higher investments generate higher relative
back-transfers. For our purposes, the most important
result of regression (1) is the small and insignificant
effect of the dummy for the implicit treatment. The coefficient
for this dummy is close to zero, highly insignificant
(p ¼ 0.533) and even has the ‘wrong’ sign, indicating
that eyespots certainly have no positive effect on trustees’
back-transfers. In addition, we find a significant (p ¼ 0.021)
effect of the fairness standard in the baseline condition—
subjects with a lower fairness standard tend to make lower
back-transfers.
In regression (2), we examine the mean relative backtransfer
on the individual level. This yields 144
observations, as we have 36 trustees in the explicit reputation
treatment (‘explicit’) and 54 in each of the other
two conditions (s.e.s are clustered at the matching
Table 2. Descriptive statistics of average, median and modal
behaviour across treatments.
statistic baseline implicit explicit
average back-transfer 6.28 5.36 13.86
median back-transfer compensate nothing equalize
mode trustee decision nothing nothing equalize
average investment 5.88 5.74 7.73
median investment 7 7 10
mode investment 10 10 10
number subjects 108 108 72
number matching groups 9 9 6
20
15
10
meanback-transfer(points)
5
0
20
15
10
meanback-transfer(points)
5
0
0
base impl.expl. base impl.expl.
treatment
base impl.expl. base impl.expl.
1 4 7 10
18.65
10.7510.92
11.91
8.94
6.11
5.33
2.35
3.36
1.1
0.410.43
2 4 6 8 10
period
(a)
(b)
Figure 2. Trustees’ mean back-transfers. (a) Over time across
treatment conditions; error bars represent s.e. on matching
group level (n ¼ 24). Solid line, baseline; dashed line, eyespots;
dotted line, explicit. (b) Per investment level across
treatment conditions.
Eye cues and strong reciprocity E. Fehr & F. Schneider 1319
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
group level). In models (2) and (3), Mach score and
Fairness represent dummies for subjects with an abovemedian
Mach score and a below-median fairness
standard; recall that subjects with a below-median fairness
standard are those who give a high (above-median)
rating on the fairness question. Interestingly, this has
little effect on the impact of the investment level and
implicit cue condition: we get virtually the same result
as in model (1), with respect to both the size and the
significance of these coefficients. In particular, the coefficient
of the implicit cue treatment is still very small,
insignificant (p ¼ 0.242) and has the wrong sign. However,
because of the larger number of observations, the
fairness standard and individuals’ Mach score is now
almost significant at the 5 per cent level (both p ¼
0.056); subjects with lower fairness standards and a
higher Mach score back-transfer less in the baseline condition.
We are also able to examine the interaction
between the fairness standard, the Mach score and the
implicit cue condition in regression (2). The interaction
between the fairness standard and the implicit cue condition
is clearly insignificant (p ¼ 0.998); the same
holds for the Mach score (p ¼ 0.279). This indicates
that the implicit cue condition also does not cause behavioural
changes in trustees with different fairness standards
and different Mach scores.
Finally, we take the decisions in each period as units of
observation and cluster again on matching groups in
model (3). The dependent variable is now the individual
relative back-transfer in a period, which limits the observations
to 0, 25 or 62.5 per cent of the received amount.
In model (3), we also include variables that capture time
effects.
The results of model (3) are interesting in several
respects. First, and most importantly, the coefficient for
the implicit cue treatment remains small in magnitude
and insignificant (p ¼ 0.304), and again has the wrong
sign. Second, subjects who have a low fairness standard
contribute less in the baseline condition (p ¼ 0.054).
Third, subjects with a high Mach score also contribute
less in the baseline condition (p ¼ 0.076). Both the
second and the third effect are substantial, reducing the
mean relative back-transfer by between 8 and 11
per cent. Fourth, the interaction between the implicit
cue condition and the below-median fairness standard/
above-median Mach dummy is not significant (p ¼
0.921/p ¼ 0.319), indicating that individuals with a low
fairness standard/high Mach score do not respond differently
to the implicit reputation cue compared with
individuals with a high fairness standard/low Mach
score. Thus, there is no evidence that individuals who
score low on selfishness and opportunism are more
prone to implicit reputation cues. Both high and low
Mach individuals show little response to the implicit
reputation cue.
5. THE IMPACT OF EXPLICIT REPUTATION
In this section, we examine the effect of pecuniary reputation
incentives on the trustees’ back-transfers and the
trustors’ investments. Table 2 shows that—in contrast to
the implicit reputation condition—the explicit reputation
condition causes an enormous increase in average backtransfers—from
6.28 to 13.86. While the modal response
in the baseline condition is ‘nothing’, the modal response
in the explicit reputation condition is ‘equalize’. This big
Table 3. OLS Regression analysis of trustee decisions. p-values are given in parentheses; in models (2) and (3), we use
Eicker–Huber–White sandwich estimators for the s.e., clustering on matching groups. Models (1) and (2) use aggregated
data on the matching group and trustee level, respectively (i.e. the variables’ relative back-transfer and ‘investment level’ are
averages on the matching group and trustee level, respectively. ‘Mach’ and ‘fairness’ represent matching group averages in
model (1). Regressors: ‘investment level’ is the number of points the trustor transfers; ‘implicit’ and ‘explicit’ are dummies
for the respective treatments (omitted category: baseline treatment). In models (2) and (3), ‘Mach’ is a dummy variable that
equals 1 if the participant scored above the median in the Mach-IV inventory. In models (2) and (3), ‘fairness’ is a dummy
variable that equals 1 if the participant’s answer to the fairness question was above the median response (i.e. if the subject’s
fairness standard is below the median). ‘explicit  last 3’ is a dummy that equals 1 if the observation comes from periods 8, 9
or 10 in the explicit treatment. Its purpose is to capture the end-game effect that occurs when the future benefits from
reputation vanish towards the end of the experiment. ‘Period’ denotes the experimental period and ranges from 1 to 10.
dependent variable
(1) mean relative backtransfer
at the level of
the matching group
(2) mean relative backtransfer
at the level of
the trustee
(3) relative backtransfer
in individual
decisions
(mean of) investment level 0.031 (0.001) 0.033 (0.000) 0.018 (0.000)
implicit 20.013 (0.533) 20.040 (0.242) 20.038 (0.304)
explicit 0.170 (0.000) 0.107 (0.012) 0.203 (0.000)
(mean of) Mach score 0.062 (0.131) 20.086 (0.056) 20.081 (0.076)
Mach  implicit 0.061 (0.279) 0.057 (0.319)
Mach  explicit 0.110 (0.033) 0.105 (0.044)
(mean of) fairness 20.108 (0.021) 20.112 (0.056) 20.109 (0.054)
fairness  implicit 20.000 (0.998) 20.006 (0.921)
fairness  explicit 0.007 (0.919) 20.018 (0.774)
explicit  last 3 20.176 (0.000)
period 20.006 (0.001)
constant 0.049 (0.347) 0.113 (0.019) 0.230 (0.000)
n/no. of variance clusters 24/— 144/24 1440/24
1320 E. Fehr & F. Schneider Eye cues and strong reciprocity
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
change in the trustees’ back-transfers is highly significant
(Mann–Whitney test: p ¼ 0.002) and led to a significant
increase in the trustors’ investments—from 5.88 to 7.74
(Mann–Whitney test: p ¼ 0.006). In the explicit condition,
the maximum investment also represents the
median investment choice. This strong impact of pecuniary
reputation incentives can also be seen in figure 2a.
The average back-transfer is much higher in the explicit
reputation condition in all but the last few periods. The
time path of the average back-transfer in figure 2a also
indicates the relatively high degree of rationality that
seems to be present in our experiment. During the early
periods, a high back-transfer generates a good reputation
for many remaining periods, implying that the pecuniary
return of a good reputation is high. During the final few
periods, a high back-transfer generates a good reputation
only for a few remaining periods, implying that the
pecuniary return of a good reputation is lower. Thus,
individuals who understand this should choose lower
back-transfers during the final few periods because the
selfish returns of behaving in this way are lower. The
time pattern of back-transfers in figure 2a is consistent
with this rational choice argument.
Interestingly, in period 10 of the explicit reputation
condition—in which there are no pecuniary reputation
incentives at all—the average level of back-transfers is
very similar to the level in the other two conditions (in
which explicit reputation incentives are absent by
design). Thus, the trustees seem to understand the logic
of pecuniary reputation incentives quite well: while they
do not respond to merely implicit reputation cues that
carry no explicit incentive power, they respond strongly
to explicit reputation incentives, and they seem to understand
when they can gain from a good reputation and
when not.
The powerful effect of pecuniary reputation incentives
can also be seen in figure 2b, which controls for trustors’
investments: trustees’ back-transfers are higher at every
investment level than in the other two conditions.
Finally, the regressions in table 3 provide further statistical
support for the large effect of the explicit reputation
condition. In models (1) and (2), the explicit reputation
incentive increases the average relative back-transfer
by 17.6 and 16.2 per cent, respectively (p , 0.001 and
p ¼ 0.012). Note that in model (3) the inclusion of the
‘explicit  last 3’ interaction implies that the ‘explicit’
variable captures the effect of the explicit reputation
incentive for the first seven periods while the variable
‘explicit  last 3’ measures the decrease of back-transfers
during the final three periods. The coefficient of
0.203 (p , 0.001) for the variable ‘explicit’ thus indicates
that in the first seven periods subjects increase the
relative back-transfer relative to the baseline condition
by 20.3 per cent if they face an explicit reputation
incentive. Moreover, highly opportunistic trustees
(above-median Mach score) show a 10.5 per cent larger
increase in relative back-transfers when they face an explicit
incentive (coefficient of 0.105; p ¼ 0.044). Taken
together, these results indicate a large effect of the explicit
reputation incentive—an effect that contrasts sharply
with the null effect of the implicit reputation cue.
In fact, an F-test indicates that the difference between
the coefficients of the implicit and the explicit condition
is highly significant (p , 0.001).
The above results confirm the hypothesis that
Machiavellian subjects respond particularly strongly to
social punishment threats such as loss of reputation.
This finding is consistent with the results of another
study (Simpson & Willer 2008), which also observes
that egoistic subjects show a stronger response to
pecuniary reputation incentives.
6. DISCUSSION
There is little disagreement among researchers that
explicit reputation incentives strongly affect human prosocial
behaviour. These explicit incentives can take the
form of higher future material benefits in a dynamic
experimental game—such as in our explicit reputation
treatment—or they can arise when real people (e.g. an
audience) saliently observe other people’s cooperative
or non-cooperative behaviour (Ga¨chter & Fehr 1999;
Rege & Telle 2004; Kurzban et al. 2007; Smith et al.
2009). The strength of merely implicit reputation cues,
in which subjects cannot really acquire a good or bad
reputation, is, however, much less investigated.
Therefore, we examined the impact of such cues on the
strongly reciprocal behaviour of trustees in a trust game.
Previous work has argued that eye cues activate reputational
concerns, but did not explicitly compare the
effect of eye cues with the effect of explicit pecuniary
reputation incentives. If reputational concerns shaped
humans’ altruistic inclinations in ancestral environments
to the extent suggested in some of the recent literature
(Haley & Fessler 2005; Burnham & Hare 2007)—that
is, if humans are indeed so sensitive to reputation cues
that they respond to them even if they carry, in fact, no
real pecuniary incentive power—subjects should generate
patterns in the eye cue condition that resemble the effects
of explicit pecuniary reputation incentives.
However, our results indicate that eye cues, which have
been hypothesized to represent reliable indicators of
potential observability of one’s behaviour over the
course of human evolution, have no effect at all on the
trustees’ altruistic behaviour. The effect of the implicit
cues treatment is close to zero, highly insignificant and
even has the wrong sign. Moreover, this null effect
holds regardless of whether we examine the response of
subjects who score high or low on the Mach scale. Our
results therefore suggest an extremely cautious view of
claims that most of the observed prosocial behaviour in
anonymous one-shot games should be attributed to
uncontrolled implicit reputation cues. At the current
state of our knowledge, this claim represents no more
than a speculation, lacking empirical support. If it
were indeed the case that uncontrolled reputation cues
are so important, behaviour should also respond to
experimentally controlled implicit reputation cues.
The null effect of the implicit reputation cue contrasts
sharply with the large impact of explicit pecuniary reputation
incentives on trustees’ behaviour. The large
contrast between the implicit and the explicit reputation
condition reinforces our conclusions above. The effect
of the implicit cue does not even resemble the effect generated
by the pecuniary reputation incentive, suggesting
that implicit cues are a relatively weak force.
We also found important individual differences in
subjects’ responses to the pecuniary reputation incentive.
Eye cues and strong reciprocity E. Fehr & F. Schneider 1321
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
Subjects who score high on the Mach scale behave less
altruistically in the baseline treatment, but they respond
more strongly to the pecuniary reputation incentive.
Why do other studies find an effect of eye cues on prosocial
behaviour while we find none? With regard to the
studies of Bateson et al. (2006) and Burnham & Hare
(2007), the following feature of their experiments might
have caused the difference. Both experiments investigate
contributions to a public good. As shown above, many
people are conditional cooperators, and their contributions
therefore depend on their beliefs about other
people’s contributions. Eye cues could generate more
optimistic beliefs about other subjects’ cooperation behaviour,
which then induce higher cooperation rates
among subjects with preferences for conditional
cooperation. This contrasts with our study in which we
have full control over subjects’ beliefs because the trustees
know exactly the investment level if they make their backtransfer.
Therefore, in our study, eye cues cannot affect
beliefs about other subjects’ behaviour.
With regard to the study of Bateson et al. (2006)—a
field experiment about voluntary contributions to an honesty
box in a university coffee room—another feature is
also potentially important. Subjects often consume
coffee jointly and observe whether their colleagues pay
for the coffee. In this case the subject’s real reputation—and
not just its imagined reputation—is at stake.
If eye cues draw attention to the moral appropriateness
of paying for one’s coffee, then this real reputation incentive
may be greatly strengthened. Thus, it is possible that
the eye cues in the Bateson et al. experiment enhanced the
already prevailing incentive to maintain one’s reputation
as an honest coffee consumer. This feature of the Bateson
et al. experiment also contrasts with our experiment
because we rule out any interaction between the eye cue
and the pecuniary (explicit) reputation incentive.
Why did the eye cue affect the behaviour in the dictator
game experiment of Haley & Fessler (2005) while lacking
effect in our trust game? A possible reason for this may be
that the dictator game constitutes a less robust situation.
Experimental economists now generally acknowledge
that the dictator game is likely to involve more experimenter
demand effects (Bardsley 2008) and is less
robust than other games in which subjects interact with
each other (Cooper & Kagel 2010). Therefore, relatively
weak forces can affect behaviour in the dictator game.
Perhaps the implicit reputation cue is one of these weak
forces.
The authors would like to thank Charles Efferson, Ryan
McKay, Sonja Vogt and two excellent referees for their
many useful comments on the manuscript.
REFERENCES
Bardsley, N. 2008 Dictator game giving: altruism or
artefact? Exp. Econ. 11, 122–133. (doi:10.1007/s10683-
007-9172-2)
Bateson, M., Nettle, D. & Roberts, G. 2006 Cues of being
watched enhance co-operation in a real-world setting.
Biol. Lett. 2, 412–414. (doi:10.1098/rsbl.2006.0509)
Bowles, S. & Gintis, H. 2004 The evolution of strong reciprocity:
co-operation in heterogeneous populations.
Theoret. Popul. Biol. 65, 17–28. (doi:10.1016/j.tpb.2003.
07.001)
Boyd, R. T. & Richerson, P. 2005 The origin and evolution of
cultures. Oxford, UK: Oxford University Press.
Boyd, R., Gintis, H., Bowles, S. & Richerson, P. J. 2003 The
evolution of altruistic punishment. Proc. Natl Acad. Sci.
USA 100, 3531–3535. (doi:10.1073/pnas.0630443100)
Brown, M., Falk, A. & Fehr, E. 2004 Relational contracts
and the nature of market interactions. Econometrica 72,
747–780. (doi:10.1111/j.1468-0262.2004.00511.x)
Burnham, T. C. & Hare, B. 2007 Engineering human cooperation:
does involuntary neural activation increase
public goods contributions? Human Nat. 18, 88–108.
(doi:10.1007/s12110-007-9012-2)
Camerer, C. F. 2003 Behavioural game theory—experiments in
strategic interaction. Princeton, NJ: Princeton University
Press.
Christie, R. & Geis, F. 1970 Studies in Machiavellianism.
New York, NY: Academic Press.
Cochard, F., Van, P. N. & Willinger, M. 2004 Trusting
behavior in a repeated investment game. J. Econ. Behav.
Org. 55, 31–44. (doi:10.1016/j.jebo.2003.07.004)
Cooper, D. & Kagel, J. H. 2010 Other-regarding preferences:
a selective survey of experimental results. In Handbook
of experimental economics (eds A. Roth & J. H. Kagel).
Princeton, NJ: Princeton University Press.
Croson, R. 2007 Theories of commitment, altruism and
reciprocity: evidence from linear public goods games.
Econ. Inquiry 45, 199–216.
Dawes, C. T., Fowler, J. H., Johnson, T., McElreath, R. &
Smirnov, O. 2007 Egalitarian motives in humans.
Nature 446, 794–796. (doi:10.1038/nature05651)
Dufwenberg, M. & Kirchsteiger, G. 2004 A theory of
sequential reciprocity. Games Econ. Behav. 47, 268–298.
(doi:10.1016/j.geb.2003.06.003)
Engelmann, D. & Fischbacher, U. 2009 Indirect reciprocity
and strategic reputation building in an experimental helping
game. Games Econ. Behav. 67, 399–407. (doi:10.
1016/j.geb.2008.12.006)
Falk, A. & Fischbacher, U. 2006 A theory of reciprocity.
Games Econ. Behav. 54, 293–315. (doi:10.1016/j.geb.
2005.03.001)
Fehr, E. & Fischbacher, U. 2003 The nature of human altruism.
Nature 425, 785–791. (doi:10.1038/nature02043)
Fehr, E. & Schmidt, K. M. 1999 A theory of fairness, competition,
and co-operation. Q. J. Econ. 114, 817–868.
(doi:10.1162/003355399556151)
Fehr, E., Fischbacher, U. & Ga¨chter, S. 2002 Strong reciprocity,
human co-operation, and the enforcement of social
norms. Human Nat. 13, 1–25. (doi:10.1007/s12110-
002-1012-7)
Fehr, E., Brown, M. & Zehnder, C. 2009 On reputation: a
microfoundation of contract enforcement and price rigidity.
Econ. J. 119, 333–353. (doi:10.1111/j.1468-0297.
2008.02240.x)
Fischbacher, U., Ga¨chter, S. & Fehr, E. 2001 Are people
conditionally co-operative? Evidence from a public
goods experiment. Econ. Lett. 71, 397–404. (doi:10.
1016/S0165-1765(01)00394-9)
Ga¨chter, S. & Falk, A. 2002 Reputation and reciprocity:
consequences for the labour relation. Scand. J. Econ.
104, 1–27. (doi:10.1111/1467-9442.00269)
Ga¨chter, S. & Fehr, E. 1999 Collective action as a social
exchange. J. Econ. Behav. Org. 39, 341–369. (doi:10.
1016/S0167-2681(99)00045-1)
Gintis, H. 2000 Strong reciprocity and human sociality.
J. Theoret. Biol. 206, 169–179. (doi:10.1006/jtbi.2000.2111)
Gintis, H., Bowles, S., Boyd, R. & Fehr, E. 2003 Explaining
altruistic behaviour in humans. Evol. Hum. Behav. 24,
153–172. (doi:10.1016/S1090-5138(02)00157-5)
Hagen, E. H. & Hammerstein, P. 2006 Game theory and
human evolution: a critique of some recent interpretations
1322 E. Fehr & F. Schneider Eye cues and strong reciprocity
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from
of experimental games. Theoret. Popul. Biol. 69, 339–348.
(doi:10.1016/j.tpb.2005.09.005)
Haley, K. J. & Fessler, D. M. T. 2005 Nobody’s watching?
Subtle cues can affect generosity in an anonymous economic
game. Evol. Hum. Behav. 26, 245–256. (doi:10.
1016/j.evolhumbehav.2005.01.002)
Henrich, J. & Boyd, R. 2001 Why people punish defectors:
weak conformist transmission can stabilize costly enforcement
of norms in co-operative dilemmas. J. Theoret. Biol.
208, 79–89. (doi:10.1006/jtbi.2000.2202)
Kocher, M. G., Cherry, T., Kroll, S., Netzer, R. & Sutter, M.
2008 Conditional co-operation on three continents.
Econ. Lett. 101, 175–178. (doi:10.1016/j.econlet.2008.
07.015)
Kurzban, R. & Houser, D. 2005 Experiments investigating
cooperative types in humans: a complement to evolutionary
theory and simulations. Proc. Natl Acad. USA 102,
1803–1807. (doi:10.1073/pnas.0408759102)
Kurzban, R., DeScioli, P. & O’Brien, E. 2007 Audience
effects on moralistic punishment. Evol. Hum. Behav. 28,
75–84. (doi:10.1016/j.evolhumbehav.2006.06.001)
Milinski, M., Semmann, D., Bakker, T. C. M. & Krambeck,
H. J. 2001 Co-operation through indirect reciprocity:
image scoring or standing strategy? Proc. R. Soc. Lond.
B 268, 2495–2501. (doi:10.1098/rspb.2001.1809)
Milinski, M., Semmann, D. & Krambeck, H. J. 2002 Reputation
helps solve the ‘tragedy of the commons’. Nature
415, 424–426. (doi:10.1038/415424a)
Rabin, M. 1993 Incorporating fairness into game theory and
economics. Am. Econ. Rev. 83, 1281–1302.
Rege, M. & Telle, K. 2004 The impact of social approval and
framing on co-operation in public good situations.
J. Public Econ. 88, 1625–1644. (doi:10.1016/S0047-
2727(03)00021-5)
Rigdon, M., Ishii, K., Watabe, M. & Kitayama, S.
2009 Minimal social cues in the dictator game.
J. Econ. Psychol. 30, 358–367. (doi:10.1016/j.joep.2009.
02.002)
Rockenbach, B. & Milinski, M. 2006 The efficient interaction
of indirect reciprocity and costly punishment.
Nature 444, 718–723. (doi:10.1038/nature05229)
Simpson, B. & Willer, R. 2008 Altruism and indirect reciprocity:
the interaction of person and situation in prosocial
behavior. Soc. Psychol. Quart. 71, 37–52.
Smith, F. G., Debruine, L. M., Jones, B. C., Krupp, D. B.,
Welling, L. L. M. & Conway, C. A. 2009 Attractiveness
qualifies the effect of observation on trusting behavior in
an economic game. Evol. Hum. Behav. 30, 393–397.
(doi:10.1016/j.evolhumbehav.2009.06.003)
Spitzer, M., Fischbacher, U., Herrnberger, B., Gron, G. &
Fehr, E. 2007 The neural signature of social norm compliance.
Neuron 56, 185–196. (doi:10.1016/j.neuron.
2007.09.011)
Wedekind, C. & Milinski, M. 2000 Co-operation through
image scoring in humans. Science 288, 850–852.
(doi:10.1126/science.288.5467.850)
Eye cues and strong reciprocity E. Fehr & F. Schneider 1323
Proc. R. Soc. B (2010)
on February 23, 2017http://rspb.royalsocietypublishing.org/Downloaded from