Moral emotions as determinants of third-party punishment: Anger, guilt, and the functions of altruistic sanctions

Judgment and Decision Making, Vol. 4, No. 7, December 2009, pp. 543–553
Moral emotions as determinants of third-party punishment: Anger,
guilt, and the functions of altruistic sanctions
Rob M. A. Nelissen? and Marcel Zeelenberg
Tilburg University, the Netherlands
Abstract
Third-party punishment has recently received attention as an explanation for human altruism. Feelings of anger in
response to norm violations are assumed to motivate third-party sanctions, yet there is only sparse and indirect support
for this idea. We investigated the impact of both anger and guilt feelings on third-party sanctions. In two studies both
emotions were independently manipulated. Results show that anger and guilt independently constitute suf?cient but not
necessary causes of punishment. Low levels of punishment are observed only when neither emotion is elicited. We
discuss the implications of these ?ndings for the functions of altruistic sanctions.
Keywords: third-party punishment, social norms, emotions, decision-making.
1 Introduction
own (Haidt, 2003). We experience feelings like empathy,
anger, and guilt if we consider how others have been hurt,
People often defend the interests of others. They stand
wronged, or harmed (e.g., Batson, 2006; Haidt, 2003).
up for their friends if someone speaks ill about them in
The view that moral emotions also have functional behav-
their absence. They do not tolerate a colleague being bul-
ioral consequences is re?ected in evolutionary hypothe-
lied at work. They boycott consumer products that are
sis about their adaptive value (e.g., Tooby & Cosmides,
produced using child labor. Some even come to the aid
1990; Trivers, 1971).
of a stranger who is being physically harassed, in spite
Especially Robert Frank’s (2004) notion of emotions
of obvious personal danger. In general, people retaliate
as “commitment devices” seems relevant to understand
against injustice even if they are not directly victimized.
how moral emotions promote prosocial behavior, in spite
Sanctioning of norm-violations is vital for prosocial be-
of the associated costs. Frank argues that moral emotions
havior to be sustained (Fehr & Gächter, 2002; Yamagishi,
have been evolved as commitment devices that make peo-
1986). However, punishing norm-violations is costly in
ple forego their immediate self-interest, committing them
terms of time and energy. It may even impose physical
to a more rewarding long-term strategy. For instance, an
risks. Punishing injustice is therefore considered to be
angry individual retaliating a norm-violation may incur
a moral act, particularly when it is performed on behalf
an immediate cost, but may derive a greater bene?t in
of others (i.e., in case of third-party sanctions, Fehr &
the long run by deterring future exploitation. Similarly,
Fishbacher, 2004). This begs the question of what incites
Frank also argued that guilt feelings act as a commitment
third-party sanctions, as they usually oppose self-interest.
device because a guilty person may invest time and en-
ergy to make up for something (s)he did to another per-
1.1 Moral emotions and prosocial behavior
son, but may eventually bene?t thereof by saving a mutu-
ally rewarding and bene?cial relationship. Precisely this
Classic philosophical treatises on “moral sentiments” al-
effect was empirically supported (De Hooge, Breugel-
ready stressed the functional role of moral emotions as
mans, and Zeelenberg, 2007; Ketelaar & Au, 2003; Nelis-
elicitors of prosocial behavior (e.g., Hume, 1739; Smith,
sen, Dijker & de Vries, 2007). Moral emotions, par-
1759). Moral emotions are de?ned as feelings related
ticularly anger, have also been proposed as the proxi-
to the interest and welfare of others rather than one’s
mal mechanism underlying third-party sanctions (Fehr &
?We thank Urs Fischbacher for providing us with the instructions
Fishbacher, 2004; Fehr & Gachter, 2002). Nevertheless,
for the third-party punishment paradigm, Johan Karremans for useful
the proposed role for emotions in third-party punishment
comment, and Jon Baron for his contributions. Correspondence can
requires further exploration for two main reasons.
be addressed to Rob Nelissen, Department of Social Psychology and
First of all, anger has not yet been linked empiri-
TIBER (Tilburg Institute of Behavioral Economics Research), Tilburg
University, PO BOX 90153, 5000-LE Tilburg, the Netherlands, Email:
cally to third-party sanctions.
The associations may
r.m.a.nelissen@uvt.nl.
seem straightforward, as numerous studies have related
543

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
544
feelings of anger — either self-reported or at the phys-
dently manipulating the elicitation of anger and guilt. Our
iological level — to retaliation of personal ill treat-
data revealed that the elicitation of either emotion already
ment (e.g., Ben-Shakhar, Bornstein, Hopfensitz, & Van
resulted in punishment, as did the elicitation of both.
Winden, 2007; Bosman & Van Winden, 2002; Pillutla &
Murnighan, 1996; Sanfey, Rilling, Aronson, Nystrom, &
Cohen, 2003). Still, third-party sanctions are different, as
2 Experiment 1
perpetrators of norm-violations are punished not for what
they did to the punisher, but for what they did to someone
We investigated the impact of anger and guilt feelings in
else. Recent insights clearly show that anger about per-
a third-party punishment paradigm (Fehr & Fischbacher,
sonal harm is a distinct emotion from empathic anger at
2004). In this paradigm, participants witness an unfair
witnessing injustice or harm to someone else (Batson et
distribution of some valuable endowment between two
al., 2007). Since both types of anger have different elic-
other people. As a form of costly punishment, partici-
itors, it remains to be demonstrated whether (empathic)
pants then had the opportunity to assign reduction-points
anger also instigates third-party punishment.
to the allocator, out of their own endowment.
Second, moral emotions fall into two large categories
To independently manipulate the elicitation of anger
(Haidt, 2003). On the one hand, other-focused emotions
and guilt, we varied the extent to which the unfair distri-
like anger arise in response to perceiving someone else in-
bution was made intentionally by the allocator (as a proxy
tentionally causing harm to another person. On the other
for anger) and the participant’s responsibility for punish-
hand, self-focused emotions like guilt feelings arise in re-
ing the norm-violation (as a proxy for guilt). Research
sponse to or anticipation of oneself being responsible for
on ultimatum bargaining (e.g., Blount, 1995) shows that
another person’s misfortune. Several studies have shown
unfair proposals evoke angry reactions (resulting in in-
that feelings of guilt are associated with prosocial behav-
creased rejection) only in case the outcome is an inten-
ior (e.g., De Hooge et al., 2007; Ketelaar & Au, 2003;
tional act by another person, but not when people believe
Nelissen et al., 2007). Insofar as third-party punishment
that, for instance, a computer has randomly (i.e., unin-
is a form of “second-order cooperation” (Fehr & Gächter,
tentionally) generated the offer. Similarly, the elicitation
2002; Yamagishi, 1986, 1988), it may be fuelled not only
of guilt is related to perceptions of personal responsi-
by anger over norm-violations, but also by anticipated
bility (e.g., Smith & Ellsworth, 1985). The bystander
guilt with respect to not punishing this violation when it
effect (Latané & Darley, 1970) shows that the presence
would be one’s responsibility to do so.
of others diffuses personal responsibility and thereby de-
creases the willingness to help a victim. Simply put, this
means that, as the number of people that are able to allevi-
1.2 Anger, guilt, and third-party sanctions
ate someone’s misfortune increases, each individual feels
We thus predict that third-party sanctions could be
less guilty for not helping and is therefore less willing to
elicited not only by feelings of anger, but also by feelings
do so. Bystander effects have also been demonstrated for
of guilt. However, even though both emotions may under-
norm-enforcing behavior (Chekroun & Brauer, 2002).
lie third-party sanctions, their impact may occur through
Based on these well-documented effects, we predicted
different routes, and their contribution to future group-
that third-party punishment would be increased (1) if
level cooperation may also be of different nature. We
norm-violations were brought about intentionally, and
think that angry people punish because they perceive the
(2) if participants were solely responsible for norm-
unequal distribution to be morally unjust. Anger then in-
enforcement.
duces retribution “whatever the consequences”. This may
also serve to deter future transgressions against all group
2.1 Method
members, including the punisher (e.g., Carlsmith, Darley,
& Robinson, 2002). Guilt feelings, we think, are elicited
Participants and design. Thirty-eight male and 53 fe-
when people perceive themselves in some way respon-
male undergraduate students (Mage = 20.84 years) par-
sible for meting out punishment. When people antici-
ticipated in exchange for course credit. The experimental
pate guilt they punish in order to restore a sense of justice
design included two between-subject factors: Intention-
among group members in general and the victim in partic-
ality of norm-violation (intentional vs. unintentional) and
ular (e.g., Darley & Pitman, 2003). This ensures that we
Responsibility for norm-enforcement (high vs. low).
sometimes punish even if deterrence of future offenses is
Procedure. Participants were seated in individual cubi-
absent.
cles. Instructions and measures were presented on a com-
We thus expect that anger and guilt feelings are inde-
puter. Participants were told they would participate in a
pendent determinants of third-party sanctions. In two ex-
single-shot negotiation, and would be randomly assigned
perimental studies, we tested our predictions by indepen-
to one of the three possible roles (see Appendix A). In

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
545
fact, all participants were assigned the role of third-party
verify whether our manipulations had the intended effects
(denoted as Player C) and received an initial endowment
on participants’ emotions. Hereto, we presented a sepa-
of 50 points. Each point was worth one lottery ticket. Par-
rate sample of participants (N = 90) with four different
ticipants were told that they would ?rst witness a distri-
hypothetical scenarios depicting the experimental condi-
bution of 100 points between two other players, and that
tions of the present study. Participants were then asked to
they would subsequently have the opportunity to adjust
indicate the extent to which they expected that observing
that distribution by assigning reduction points to the allo-
the unequal distribution would make them feel angry or
cator (denoted as Player A). Each reduction point would
guilty over not assigning reduction points. Anger towards
reduce the endowment of the player to whom it was as-
the allocator was assessed by the items “angry”, “mad”,
signed by three points. Next, the negotiation started by
and “annoyed” (? = .94). Anticipated guilt over not pun-
the allocator dividing an initial endowment of 100 points
ishing was assessed by the items “guilty”, “feeling bad
with the receiver. All participants learned that the allo-
for what I did”, and “ashamed” (? = .85).2 Ratings were
cator kept 80 point and that the receiver received only
made on scales from 1 (not at all) to 7 (extremely).
20 points. Hence, the allocator clearly violated the equal
distribution norm.
Manipulation of intentionality. In the unintentional
2.2 Results
norm-violation condition participants learned that the
Pretest. We ?rst analyzed the data from the separate sam-
proposal from the allocator was in fact generated ran-
ple used to pretest the effect of our manipulations on emo-
domly by the computer. In order to avoid confounding
tions. A 2 (intentionality) × 2 (responsibility) ANOVA
of both manipulations, it was stressed to our participants
on anger showed only a main effect of intentionality, F(1,
that the receiver was unaware that the offer was generated
86) = 32.34, p < .001, ?2 = .27. As intended, unequal of-
randomly by the computer but still believed that another
fers by another participant aroused more anger (M = 3.29,
participant had made the offer. Hence, participants’ re-
SD = 1.40) than unequal offers made by a computer (M
sponsibility to step up for the receiver by punishing the
= 1.71, SD = 1.22). We also found only the expected
norm-violation was unaffected by the intentionality ma-
main effect of responsibility on anticipated guilt F(1, 86)
nipulation.
= 28.60, p < .001, ?2 = .24, showing that participants felt
Manipulation of responsibility. In order to manipulate
less guilty when there were other punishers (M = 2.68,
responsibility for norm-enforcement, participants were
SD = 1.31) than when they were the only person respon-
either told they were the only Player C in the interac-
sible for punishing the unfair distribution (M = 4.26, SD
tion (high-responsibility condition), or that there were
= 1.52).
two other participants who were also assigned this role
Manipulation checks. Subsequently, we turned to the
(low-responsibility condition).
data from the sample used in the main study. A 2 (inten-
Measures. The amount of reduction-points (0–50) as-
tionality) × 2 (responsibility) ANOVA on the perceived
signed to the allocator served as a measure of third-party
intentionality of the allocator revealed that our manipula-
punishment. To check the intentionality manipulation,
tion was successful. We found only the intended main
participants indicated the extent to which they perceived
effect of intentionality, F(1,91) = 7.12, p = .005, one-
the allocator as “acting intentionally”, “responsible”, and
tailed, ?2 = .08. Participants in the computer-generated
“accountable” (? = .59) for the proposed distribution. To
offer condition found that the allocator had acted less in-
check the responsibility manipulation, participants rated
tentionally (M = 4.64, SD = 1.01) than participants in the
the extent to which they themselves felt “responsible”,
person-generated offer condition (M = 5.25, SD = 1.14).
“accountable”, and “liable” (? = .84) for assigning re-
The same ANOVA on perceived personal responsibility
duction points. All measures were assessed on 7-point
to punish yielded only a main effect of responsibility,
scales from 1 (not at all) to 7 (completely).1
F(1,91) = 11.53, p < .001, ?2 = .57. Participants in the
Pretest. In addition, we also tested whether our manip-
low-responsibility (i.e., multiple punisher) condition felt
ulations of intentionality and responsibility affected feel-
less responsible (M = 3.06, SD = 0.88) than in the high-
ings of anger and guilt as we intended. However, because
previous studies (e.g., Keltner, Locke, & Audrain, 1993)
2Note, that the original text was in Dutch and that we used the fol-
have shown that explicitly rating emotional reactions may
lowing Dutch words (English in parentheses):
attenuate their effect, we did not measure emotions in this
boos ( angry), kwaad ( mad), geirriteerd (annoyed), schuldig (guilty),
voelde me slecht over wat ik gedaan had (feeling bad for what I did), and
study. Instead, we performed an independent check to
schaamte ( ashamed). Note moreover, that we assessed guilt by means
of three items, including on item tapping feeling of shame. We realize
1Note, that the original text was in Dutch and that we used the fol-
that shame is an emotion distinct from guilt, but at the same time, both
lowing Dutch words (English in parentheses): moedwillig (intention-
are self-conscious emotions that are felt when people feel responsible
ally), verantwoordelijk (responsible), aansprakelijk (accountable), and
for something bad happens to another person. In the present study, this
beslissend (liable).
is also the case, as is evident from the high reliability of the scale.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
546
responsibility (single-punisher) condition (M = 5.48, I =
in Experiment 2.
1.30). We wish to stress that perceived responsibility was
Secondly, participants’ willingness to punish a com-
not affected by whether the offer was made by a computer
puter (in the unintentional violation condition of Experi-
or a person. Hence, manipulating intentionality had not
ment 1) may seem surprising at ?rst. In our opinion, this
affected participants’ responsibility towards the receiver.
?nding clearly shows that people besides seeking retalia-
Punishment. A 2 (intentionality) × 2 (responsibility)
tion, also punish norm-violations to restore a sense of jus-
ANOVA on the number of reduction points yielded a
tice in the harmed person, particularly if this is the only
main effect of responsibility, F(1,89) = 7.61, p = .004,
means to alleviate the victim’s suffering. We acknowl-
one tailed, ?2 = .08, indicating that participants punished
edge however, that participants in our study may have
more if they were the only punisher (M = 14.16, SD =
considered punishing a computer an awkward way to ex-
5.84), than when there were two other punishers present
press their concern with the victim. Another reason for
(M = 10.56, SD = 6.42). The hypothesized effect of in-
conducting Experiment 2 was to alleviate doubts about
tentionality was also (barely) signi?cant (M = 13.42, SD
the validity of this effect, by using a different manipula-
= 6.78, when the offer was intentional; M = 11.13, SD =
tion of anger that did not require participants to punish a
5.82, when unintentional; F(1,89) = 2.90, p = .046, one-
computer.
tailed, ?2 = .03).3
3 Experiment 2
2.3 Discussion
In this experiment we manipulated the extent to which
To independently manipulate the elicitation of anger and
norm violations were brought about intentionality (as a
guilt without requiring our participants to punish an
proxy for anger) and people’s responsibility for sanction-
inanimate object, we applied a noise-manipulation (Van
ing that norm violation (as a proxy for guilt). A pretest
Lange, Ouwerkerk & Tazelaar, 2002). In social interac-
demonstrated that the manipulations of intentionality and
tions, noise can be de?ned as any kind of involuntary
responsibility affected feelings of anger and guilt as in-
disturbance from an intended outcome. Positive noise
tended. This results therefore support the notion that
causes the actual outcome to be better than originally in-
anger and guilt both elicit third-party sanctions of ap-
tended and negative noise leads to worse outcomes.
proximately 27 percent of the endowment. Levels of pun-
In the present study, we introduced noise by stating that
ishment were reduced to about 16% of their endowment
there was a possibility that the computer would randomly
only if people were not responsible for punishing an unin-
change the allocator’s offer to the receiver, without the
tentional norm violation. Because the actual punishment
receiver being aware of this possibility. In the positive
was three times the number of reduction points, the pun-
noise condition, an unfair division by the allocator was
ishments were quite substantial.
increased to a more or less equal offer. Notably, partic-
In order to further test the independent contribution of
ipants believed the receiver to be unaware of the orig-
anger and guilt on punishment, we ran another experi-
inal proposal and also of the possibility that the origi-
ment in which we investigated the effects of a manipu-
nal proposal could be randomly changed. So, feelings
lation that independently inhibited anger and guilt rather
of anticipated guilt should not be affected in the positive
than eliciting each emotion separately. The second ex-
noise condition. Hence, positive noise reduced the need
periment provided two other important extensions. First,
to restore justice to the receiver, but did not affect the de-
levels of anger and guilt were not directly associated to
terrence function of third-party sanctions. Consequently,
levels of punishment in Study 1 and therefore, their im-
participants should feel angry towards the allocator for
pact could only be inferred. Although we had good rea-
making an unequal offer.
sons to initially refrain from direct assessment of emo-
Negative noise was modeled by the reduction of a fair
tional reactions in the ?rst experiment, the independent
offer to an unfair one. This manipulation maintained the
causal in?uence of anger and guilt on third-party sanc-
need to restore a sense of justice to the receiver, but not
tions still requires empirical demonstration, which we did
to deter future norm violations by the allocator. We ex-
pected this to evoke anticipated guilt for not punishing
3We found a nearly signi?cant interaction, F(1,91) = 3.04, p = .085,
but to inhibit anger towards the allocator.
two-tailed, ?2 = .03. Simple-effects analysis revealed that the main
effect of responsibility on punishment was signi?cant only within the
Both conditions were compared to a control condition
unintentional violation condition, F(1,91) = 10.29, p = .002, ?2 = .11.
in which the allocator made an unfair offer that was not
Intentional norm violations were punished equally irrespective of the
changed. We expected this to elicit both feelings of anger
level of responsibility of the participants, F(1,91) = 0.53, p = .47, ?2
and anticipated guilt. Consequently we anticipated lower
= .01. The interaction term is not included in the main analysis just
reported; if it is included, the effect of intentionality is not quite signi?-
levels of punishment in both noise-conditions, showing
cant (p=.054, one-tailed).
the unique contribution of anger to punishment in the pos-

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
547
itive noise condition and of anticipated guilt in the nega-
Control
35
tive noise condition.
Positive noise
Furthermore, we predicted that the effects of noise
Negative noise
30
would be mediated by anger and anticipated guilt. Specif-
ically, the difference in observed levels of punishment be-
25
tween the control and the positive noise condition should
be mediated by feelings of anticipated guilt. Feelings
20
*
of anger should be equally high in the control and the
positive noise condition, which are similar in the sense
15
that a reaction is required to an unfair offer by the allo-
Reduction points (0?50)
cator. The difference in observed levels of punishment
10
between the control and the negative noise condition on
*
the other hand, should be mediated by feelings of anger.
5
*
*
Feelings of anticipated guilt should both be high in the
control and the negative noise condition, which are sim-
0
ilar in the sense that a reaction is required to an unfair
Punishment
Guilt
Anger
outcome to the receiver. Hence, we predict the effects of
positive and negative noise to be mediated by the feelings
Figure 1: Mean number of reduction points assigned in
that they are intended to reduce, not by the feelings that
the control, positive, and negative noise condition in Ex-
they should not affect compared to the control condition.
periment 2. * denotes a signi?cant difference in means
from the Control condition at p < .05.
3.1 Method
measure of third-party punishment.
Participants and design. Twenty-seven male and 103 fe-
male undergraduate students (Mage = 19.3 years) partici-
pated in exchange for course credit. Procedures and in-
3.2 Results
structions were identical to Experiment 1, except for the
Figure 1 shows the main results. As predicted, partici-
noise-manipulation (see Appendix B), which introduced
pants in the control condition (M = 9.86, SD = 7.19) pun-
the following three conditions: Control (no change of
ished more than participants in the positive noise condi-
unfair offer), Positive noise (unfair offer increased), and
tion (M = 6.59, SD = 5.75), t(82) = ?2.30, p = .012, one-
Negative noise (fair offer decreased).
tailed. Participants in the control condition also punished
Noise manipulation. As stated, participants were in-
more than participants in the negative noise condition (M
structed that in some instances, the computer would ran-
= 3.11, SD = 5.68), t(87) = ?4.93, p < .001.4
domly change the offer made by the allocator (Van Lange
Moreover, we also found the expected differences
et al., 2002). This change affected only the outcome to
between conditions in terms of self-reported anger to-
the receiver, whereas the allocator would still receive the
wards the allocator, and guilt over not assigning reduc-
payoff as originally proposed. In the no-change control
tion points. The control (M = 34.62, SD = 31.10) and
condition, the receiver received 20 (out of 100) points
the positive noise condition (M = 17.07, SD = 27.96) dif-
from the allocator. In the positive noise condition, the
fered signi?cantly only in terms of anticipated guilt over
allocator made the same unequal (i.e., 80/20) offer, yet
not punishing (p = .004, one-tailed) but not in terms of
the computer increased this offer to 52 points for the re-
anger towards the allocator (p = .396). The control and
ceiver. In the negative noise condition, the computer re-
the negative noise condition (M = 20.49, SD = 24.48)
duced an initially fair (i.e., 50/50) proposal to a mere 18
differed in terms of anticipated guilt (p = .010), but the
points for the receiver. We chose just off-round ?gures
control (M = 28.86, SD = 26.79) condition differed much
in the noise conditions to render the ostensibly random
more from the negative noise condition (M = 2.53, SD =
nature of changes more credible to participants.
10.61) in terms of anger towards the allocator (p < .001).
Measures. After participants learned the offer (and
A test of the canonical correlation between anger and
how this was changed in the noise conditions) they in-
guilt, on the one hand, and positive/control contrast and
dicated the extent to which they felt angry (? = .96) and
negative/control contrast, on the other, revealed that the
guilty (? = .91), using the same items as in the pretest
second canonical correlate was signi?cant (p=.004, using
of Experiment 1. Ratings were made by dragging a
Rao’s approximation; see Burns, 2009), which indicates
pointer on a 100 point visual analogue scale, anchored
not at all — extremely. Next, they indicated the number
4The difference between the positive and the negative noise condi-
of reduction-points (0–50) assigned to the allocator as a
tion was also signi?cant, t(85) = ?2.83, p = .01, two-tailed.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
548
Anger
sponsibility on punishment levels to those in a control
condition in which an unfair offer was made intention-
ally and participants were fully responsible for punishing
this norm violation. As predicted, we found that reducing
?.55***
.33**
intentionality and responsibility each resulted in a signi?-
cant reduction of punishment levels. These ?ndings com-
plemented those of Experiment 1 in which an increase
Negative
Punishment
rather than a decrease of intentionality and responsibil-
noise
?.47***/?.28*
ity was also found to exert an independent effect on third
party punishment.
Although a canonical correlation analysis suggested
Figure 2: In?uence of negative noise, direct and mediated
that the effects of our noise manipulations were best pre-
by anger on punishment in Experiment 2. Numbers rep-
dicted by two distinct emotion factors, we found direct
resent standardized regression coef?cients: * p < .05. **
con?rmation only of the hypothesized role of anger. Re-
p < .01. *** p < .001.
sults supported anger as a determinant of third-party pun-
ishment in the negative noise condition. Anger also ap-
that two distinct factors were required to predict emotion
parently mediated the effect of negative noise on punish-
from experimental condition.
ment. But guilt feelings did not mediate punishment in
The effects of negative noise on punishment were me-
the positive noise condition, contrary to our hypothesis.
diated by anger, as we hypothesized.5 Speci?cally, as
The failure of guilt to affect punishment directly does
shown in Figure 2, in regressions using standardized vari-
not necessarily contradict its role. The feeling of guilt
ables, negative noise was associated with reduced anger
might have been reduced by the intention to punish. Thus,
(? = ?.47, p < .001) and with reduced punishment
guilt could drive the desire to punish but then be reduced
(? = ?.55, p < .001). And anger predicted punishment
once this desire is translated into an intention to act. Such
when negative noise was included in the model (? = .33,
an interpretation is consistent with the effects of positive
p < .001). The effect of negative noise on punishment was
noise on guilt, and on punishment. Anger, by contrast,
reduced from ?.47 to ?.26 when anger was included in
would not dissipate so quickly, perhaps not until the pun-
the model.
ishment is actually implemented.
However, there was no mediation of the effect of pos-
itive noise by guilt feelings. Positive noise did reduce
punishment (? = ?.25, p=.024), and it reduced guilt
4 General discussion
(? = ?.28, p=.008), but guilt did not affect punish-
ment signi?cantly when positive noise was included in
We found in two experiments that third-party punishment
the model (? = .06, p=.614).6
has distinct determinants, that probably relate to distinct
emotional processes. Experiment 1 revealed that manipu-
lating the intentionality of the norm-violation (as a proxy
3.3 Discussion
for anger) and the responsibility for sanctioning (as a
In this experiment we aimed to inhibit feelings of anger
proxy for guilt) both increase levels of punishment. Ex-
and guilt independent from each other. To that end we
periment 2 demonstrated the involvement of anger and,
introduced a noise manipulation (Van Lange et al., 2002)
possibly, anticipated guilt more directly, by showing that
that entailed either an increase of an unfair offer to an
inhibiting each emotion reduced punishment. Together,
equal one (positive noise) or a decrease of a fair offer to
these ?ndings suggest that both emotions separately con-
an unequal one (negative noise). Consequently, the posi-
stitute suf?cient but not necessary causes of third-party
tive noise manipulation reduced the responsibility of the
punishment. Apparently, punishment can be extended by
participant for punishing a norm violation that was never-
third parties in response to unfair intentions and to un-
theless still intentional. The negative noise manipulation
fair outcomes. When either is missing, punishment is re-
maintained the participant’s responsibility to punish an
duced.
unequal offer that was however not made intentionally.
These results both support and extend current propos-
We compared the effects of reduced intentionality and re-
als about the functions of sanctioning norm-violations.
The impact of anger is in line with views that hold punish-
5See Baron and Kenney (1986) and MacKinnon et al., (2002) for
ment to primarily serve retaliatory purposes (Carlsmith
discussion of mediation.
et al. 2002; Darley & Pittman, 2003). Additionally, we
6In addition, anger did not mediate the effects of positive noise and
guilt did not mediate the effects of negative noise. These were not hy-
have provided initial support that feelings of anticipated
pothesized to occur.
guilt may underlie third-party sanctions as well. Feelings

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
549
of guilt over not punishing norm violations may them-
of envy could be evoked. Envy is the emotion that moti-
selves suf?ce to cause punishment. That is, anticipated
vates behavior aimed at reducing the differences between
guilt instigates punishment even when people know that
oneself and another who is better off (e.g., Van de Ven,
the norm was not (intentionally) violated. Identifying
Zeelenberg, Pieters, 2009). Hence participants may be
anticipated guilt as an additional determinant broadens
willing to incur a cost to punish the other, and hence re-
the functional perspective of third-party sanctions. Feel-
duce inequality. We agree that such effects of envy could
ings of guilt generally motivate behavior that is aimed
exist, also in our studies, in addition to the effects we
to restore transgressions towards others (Haidt, 2003),
have found. Future studies could investigate the extent to
guilt-induced punishment may imply that punishers seek
which envy has an additional effect here.
to compensate the victim by restoring a sense of justice
Furthermore, we documented independent contribu-
(e.g., Darley & Pitman, 2003) rather than by retaliating
tions of anger and guilt only in a single paradigm.
against the perpetrator of a violated norm.
Whether anger and guilt underlie third-party sanctions in
It may seem that our ?ndings contradict those of a re-
general remains to be seen. Similarly, we studied punish-
cent study that also investigated the consequences of a
ment only in response to violations of a single norm that
mismatch between intended and actually obtained out-
prescribes equal distribution. Behavior in other situations
comes in a dictator game (Cushman, Dreber, Wang, &
may be guided by different norms (e.g., courage, loy-
Costa, 2009). Whereas our ?ndings seem to imply that
alty, and modesty). Whether or not the violation of other
intentions matter more than outcomes, Cushman and col-
norms elicits feelings of anger and guilt, as well, also re-
leagues (2009) observed that unfair outcomes have a
mains to be seen. Some studies seem to suggest that dif-
strong effect on the level of punishment, even if alloca-
ferent types of violations evoke speci?c emotional reac-
tors apparently had fair intentions. Two important differ-
tions (Rozin, Lowery, Imada, & Haidt, 1999). This would
ences with our own study are that Cushman et al. inves-
suggest that the emotional basis for punishing norm vio-
tigated second-party sanctions (i.e., by the victim rather
lations is more diverse than the present study suggests.
than a third party), and that their manipulation allowed
Documenting speci?c emotional reactions may prove a
for idiosyncratic attribution of intentionality by the vic-
fruitful way to establish functional links between differ-
tim. This resulted in the victim having ample leeway
ent types of norm violations and the various goals un-
to ascribe unfair intentions to the allocator in case of an
derlying their punishment (Carlsmith et al., 2002; Darley
unequal distribution, which may have accounted for the
& Pitman, 2003; Zeelenberg, Nelissen, Breugelmans, &
?nding that outcomes were the primary cause of punish-
Pieters, 2008).
ment in their study. Our manipulation of noise in Study 2
Insight about multiple motives underlying sanc-
guaranteed a stricter differentiation of the effect of inten-
tions may also further our understanding of group-
tion and outcome on punishment levels, which we believe
identi?cation effects that have been reported for the pun-
explains the fact that outcome is not the strongest deter-
ishment of norm violations. For instance, punishment is
minant of punishment in our research.
more severe when in-group members suffer from norm-
Our results should be interpreted bearing some reser-
violations than when out-group members are victimized
vations in mind. One particular aspect of our studies may
(Bernard, Fischbacher & Fehr, 2006). On the other hand,
be of concern to perceptive readers. As participants’ out-
more severe punishment befalls in-group members vio-
comes in our studies earned them lottery tickets, and their
lating social-norms than when out-group members harm
chances of winning the lottery depended upon the number
each other (Shinada, Yamagishi & Ohmura, 2004).
of tickets they earned themselves but also on the number
It is likely that the motives underlying speci?c in-
of tickets earned by the other participants, one may ar-
stances of punishment depend upon the relations between
gue that they punished in order to increase their chances
punisher, perpetrator, and victim. Closer ties with the
of winning the lottery. We consider this unlikely how-
perpetrator, for instance, may attenuate the elicitation of
ever, as the levels of punishment in our studies did not
anger as punishers may be more inclined to justify vio-
exceed those reported in other third-party punishment ex-
lations from people with whom they have a close rela-
periments in which punishment actually cost money and
tionship. Guilt feelings are apt to be responsive to af?l-
therefore could in no way improve punishers’ outcomes
iations with the victim as they are likely to affect pun-
(e.g., Fehr & Fischbacher, 2004). We therefore do not be-
ishers’ perceptions of responsibility. Alternatively, it has
lieve that participants’ motivation to win the lottery has
been found that increased identi?cation with the victim
confounded the present ?ndings.
enhances feelings of anger (Yzerbyt, Dumont, Wigboldus
A similar line of reasoning may cause one to wonder
& Gordijn, 2003), which may imply that closer connec-
if our experiments induced only anger and guilt. For in-
tions with either party would instigate more intense feel-
stance, in Experiment 1, when participants learn that they
ings of anger in punishers. Although Yzerbyt and col-
receive 50 points while Player A has 80 points, feelings
leagues did not assess feelings of anticipated guilt, it may

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
550
be that feelings of guilt for not punishing norm-violations
Batson, C. D., Kennedy, C. L., Nord, L., Stocks, E. L.,
are not dependent upon ties with either victim or perpetra-
Fleming, D. A., Marzette, C. M., et al. (2007). Anger
tor. Rather, guilt-induced punishment in order to alleviate
at unfairness: Is it moral outrage? European Journal
a victim’s suffering may re?ect the violation of personal
of Social Psychology, 37, 1272–1285.
standards of fairness.
Ben-Shakhar, G., Bornstein, G., Hopfensitz, A., & van
The compound causation of third-party sanctions de-
Winden, F. (2007). Reciprocity and emotions in bar-
serves further inquiry as it may improve policies aiming
gaining using physiological and self-report measures.
to stimulate informal (i.e., non-institutionalized) sanc-
Journal of Economic Psychology, 28, 314–323.
tions as a means to instill norms that bene?t everyday
Bernhard, H., Fischbacher, U., & Fehr, E. (2006).
social interactions. Whereas the execution of actual third-
Parochial altruism in humans. Nature, 442, 912–915.
party punishment is arguably the function of govern-
Blount, S. (1995). When social outcomes aren’t fair: The
ments, and actual third party punishment is rare in every-
effect of causal attributions on preferences. Organiza-
day interactions (Sabini & Silver, 1982), people will not
tional Behavior and Human Decision Processes, 63,
refrain from extending mild forms of punishment in terms
131–144.
of remarks of passive sanctions, as the familiar examples
Bosman, R., & van Winden, F. (2002). Emotional hazard
in the opening paragraph illustrated. For instance, with
in a power to take experiment. The Economic Journal,
respect to phenomena like random violence, littering, and
112, 147–169.
proper codes of conduct in particular environments, infor-
Burns, C. T. (2009). yacca: Yet another canonical cor-
mal sanctions can improve social interactions. Regard-
relation analysis package, (R package version 1.1).
ing such issues, it should be noted that punishment is
http://CRAN.R-project.org.
still sensitive to concerns of self-interest, for increasing
Carlsmith, K. M., Darley, J. M., & Robinson, P. H.
costs of sanctions appear to reduce the level of punish-
(2002). Why do we punish? Deterrence and just
ment (Fehr & Fischbacher, 2004). Considering multiple
deserts as motives for punishment. Journal of Person-
motives may help to identify factors that maintain infor-
ality and Social Psychology, 83, 284–299.
mal sanctions even under conditions that are apt to under-
Checkroun, P., & Brauer, M. (2002). The bystander effect
cut other incentives. Angry reactions may dampen when
and social control behavior: The effect of the presence
norm-violations also bene?t the third-party, for example.
of others on people’s reactions to norm violations. Eu-
Still, it may well be that guilt-induced punishment pre-
ropean Journal of Social Psychology, 32, 853–867.
vails even under conditions where punishers do not feel
Cushman, F., Dreber, A., Wang, Y., & Costa, J. (2009).
angry.
Accidental outcomes guide punishment in a “Trem-
To summarize, the present study is the ?rst to empiri-
bling Hand” game. Plos One, 4, 1–7.
cally support the proposed link between anger and third-
Darley, J. M., & Pittman, T. S. (2003). The psychology of
party punishment. Moreover, we showed that angry reac-
compensatory and retributive justice. Personality and
tions to norm-violations are not necessarily required and
Social Psychology Review, 7, 324–336.
third-party punishment may also be motivated by previ-
De Hooge, I. E., Breugelmans, S. M. & Zeelenberg, M.
ously unexplored emotions like guilt. Either emotional
(2007). Moral sentiments and cooperation: Differen-
response may be in itself suf?cient to induce punishment.
tial in?uences of shame and guilt. Cognition and Emo-
This suggests that we not only punish norm violations to
tion, 21, 1025–1042.
deter future transgression by the perpetrator, but also to
Fehr, E., & Fischbacher, U. (2004). Third-party punish-
restore a sense of justice to the victim.
ment and social norms. Evolution and Human Behav-
ior, 25, 63–87.
Fehr, E., & Gächter, S. (2002). Altruistic punishment in
References
humans. Nature, 415, 137–140.
Frank, R. (2004). Introducing moral emotions into mod-
Barron, R. M., & Kenny, D. A. (1986). The moderator-
els of rational choice. In A. S. R. Manstead, N. Frijda
mediator variable distinction in social psychological
& A. Fischer (Eds.), Feelings and emotions: The Am-
research: Conceptual, strategic, and statistical consid-
sterdam symposium (pp. 422–440). Cambridge: Cam-
erations. Journal of Personality and social Psychology,
bridge University Press.
51, 1173–1182.
Haidt, J. (2003). The moral emotions. In R. J. Davidson,
Batson, C. D. (2006). Not all self-interest after all: Eco-
K. R. Scherer & H. H. Goldsmith (Eds.), Handbook of
nomics of empathy-induced altruism. In D. De Cre-
affective sciences (pp. 852–870). NY: Oxford Univer-
mer, M. Zeelenberg, & J. K. Murnighan (Eds), Social
sity Press.
psychology and economics (pp. 281–299). Mahwah,
Hume, D. (1739/1969). A treatise of human nature. Lon-
NJ: Lawrence Erlbaum Associates.
don: Penguin.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
551
Keltner, D., Locke, K. D., & Audrain, P. (1993). The
malicious envy. Emotion, 9, 419–429.
in?uence of attributions on the relevance of negative
Van Lange, P. A. M., Ouwerkerk, J. W., & Tazelaar, M. J.
feelings to personal satisfaction. Personality and So-
A. (2002). How to overcome the detrimental effects of
cial Psychology Bulletin, 19, 21–29.
noise in social interaction: The bene?ts of generosity.
Ketelaar, T., & Au, W. T. (2003). The effects of feel-
Journal of Personality and Social Psychology, 82, 768–
ings of guilt on the behavior of uncooperative individ-
780.
uals in repeated social bargaining games: An affect-
Yamagishi, T. (1986). The provision of a sanctioning sys-
as-information interpretation of the role of emotion in
tem as a public good. Journal of Personality and Social
social interaction. Cognition and Emotion, 17, 429–
Psychology, 51, 110–116
453.
Yamagishi, T. (1988). Seriousness of social dilemmas
Latané, B., & Darley, J. M. (1970). The unresponsive by-
and the provision of a sanctioning system. Social Psy-
stander: Why doesn’t he help? NY: Appleton-Century-
chology, 51, 32–42.
Crofts.
Yzerbyt, V., Dumont, M., Wigboldus, D., & Gordijn, E.
MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M.,
(2003). I feel for us: The impact of categorization
West, S. G., & Sheets, V. (2002). A comparison of
and identi?cation on emotions and action tendencies.
methods to test mediation and other intervening vari-
British Journal of Social Psychology, 42, 533–549.
able effects. Psychological Methods, 7, 83–104.
Zeelenberg, M., Nelissen, R. M. A., Breugelmans, S. M.,
Nelissen, R. M. A., Dijker, A. J., & de Vries, N. K.
& Pieters, R. (2008). On emotion speci?city in deci-
(2007). How to turn a hawk into a dove and vice
sion making: Why feeling is for doing. Judgment and
versa: Interactions between emotions and goals in a
Decision making, 3, 18–27.
give-some dilemma game. Journal of Experimental
Appendix A: Instructions for Experi-
Social Psychology, 43, 280–286.
Pillutla, M. M., & Murnighan, J. K. (1996). Unfairness,
ment 1
anger, and spite: Emotional rejections of ultimatum of-
fers. Organizational Behavior and Human Decision
Dear participant,
Processes, 68, 208–224.
Welcome to this study on social interactions. Please
Rozin, P., Lowery, L., Imada, S., & Haidt, J. (1999).
read the following instructions carefully. In case of ques-
The CAD triad hypothesis: A mapping between three
tions or uncertainties, please call the experimenter by
moral emotions (contempt, anger, disgust) and three
pushing the red button on the intercom.
moral codes (community, autonomy, divinity). Journal
You are going to participate in an interaction between
of Personality and social Psychology, 76, 574–586.
three parties. All participants will be randomly assigned
Sabini, J., & Silver, M. (1982). Moralities of Everyday
to either one of three possible roles, Player A, B, or C.
Life. New York: Oxford University Press.
In the course of this interaction, you and the other
Sanfey, A. G., Rilling, J. K., Aronson, J. A., Nystrom, L.
players can earn points. For each point that you have
E., & Cohen, J. D. (2003). The neural basis of eco-
earned you will get one lottery ticket. At the end of this
nomic decision-making in the ultimatum game. Sci-
week, when the experiment is over, we will draw three
ence, 300, 1755–58.
tickets that will win a price of C50,- each. This means
Shinada, M., Yamagishi, T., & Ohmura, Y. (2004). False
that your decisions in this interaction may increase or de-
friends are worse than bitter enemies: “Altruistic” pun-
crease your chances of winning money.
ishment of in-group members. Evolution and Human
Before we start, the computer will now randomly as-
Behavior, 25, 379–393.
sign each participant to the role of either Player A, B, or
Smith, A. (1759/1976). The theory of moral sentiments.
C.
Oxford: Clarendon Press.
[TIME LAG]
Smith, C. A., & Ellsworth, P. C. (1985). Patterns of cog-
nitive appraisal in emotion. Journal of Personality and
Conditions:
Social Psychology, 48, 813–838.
Single Punisher: You are assigned to the role of Player
Trivers, R. L. (1971). The evolution of reciprocal altru-
C.
ism. Quarterly Review of Biology, 46, 35–57.
Multiple Punishers: You are assigned to the role of
Tooby, J., & Cosmides, L. (1990). The Past Explains the
Player C. Apart from yourself; two other partici-
Present: Emotional Adaptations and the Structure of
pants have also been assigned the role of Player C.
Ancestral Environments. Ethology and Sociobiology,
11, 375–424.
Player A = Participant: In this interaction you are
Ven, N. van de, Zeelenberg, M., & Pieters, R. (2009).
paired with two other participants who have been as-
Leveling up and down: The experiences of benign and
signed the roles of Player A and Player B.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Effect of moral emotions on third-party punishment
552
Player A = Computer: The role of Player A in this in-
You are going to participate in an interaction between
teraction will be played by the computer. The com-
three parties. All participants will be randomly assigned
puter will randomly allocate points to Player B.
to either one of three possible roles, Player A, B, or C.
Player B, however, is not aware of this and assumes
In the course of this interaction, you and the other
that another participant makes the decision to allo-
players can earn points. For each point that you have
cate points.
earned you will get one lottery ticket. At the end of this
week, when the experiment is over, we will draw three
Everyone will remain completely anonymous during
tickets that will win a price of C50,- each. This means
and after the experiment. So, you will never know which
that your decisions in this interaction may increase or de-
other participants were the Players A and B in this in-
crease your chances of winning money.
teraction, nor will they know that you were the Player C
Before we start, the computer will now randomly as-
with whom they had interacted.
sign each participant to the role of either Player A, B, or
If you press “Continue” you will receive information
C.
about the two phases of the social interaction. Players A
[TIME LAG]
and B will receive the same information.
You are assigned to the role of Player C.
In phase 1 only the participants that have been assigned
Everyone will remain completely anonymous during
to the role of Player A will make a decision. At the be-
and after the experiment. So, you will never know which
ginning of phase 1, Player A has 100 points. Player B has
other participants were the Players A and B in this in-
no points, and Player C has 50 points.
teraction, nor will they know that you were the Player C
Player A may allocate a voluntary number of points to
with whom they had interacted.
Player B. It is up to Player A to decide how many points
If you press “Continue” you will receive information
he or she will allocate to Player B. Player A may even
about the two phases of the social interaction.
decide not to allocate any points at all to Player B, or to
In phase 1 only the participants that have been assigned
allocate all points to Player B. [EXAMPLES].
to the role of Player A will make a decision. At the be-
In phase 2 only the participants that have been assigned
ginning of phase 1, Player A has 100 points. Player B has
the role of Player C will make a decision. As soon as
no points, and Player C has 50 points.
Player A has decided how many points to allocate to
Player A may allocate a voluntary number of points to
Player B, Players B and C will be informed of this de-
Player B. It is up to Player A to decide how many points
cision.
he or she will allocate to Player B. Player A may even
Next, Player C can assign reduction points to Player A.
decide not to allocate any points at all to Player B, or to
Every reduction point that Player C assigns to Player A
allocate all points to Player B. [EXAMPLES].
reduces the total number of points from Player C by 1, but
In phase 2 only the participants that have been assigned
the total number of points from Player A by 3. Player C
the role of Player C will make a decision. As soon as
is free to assign any number of reduction points to Player
Player A has decided how many points to allocate to
A, either zero of all 50 points that were initially assigned.
Player B, Players B and C will be informed of this de-
[EXAMPLES].
cision.
So, Player A will ?rst decide how to distribute the 100
Next, Player C can assign reduction points to Player A.
points. As soon as Player A has made a decision, Players
Every reduction point that Player C assigns to Player A
B and C will be informed thereof. Next, Player C may
reduces the total number of points from Player C by 1, but
decide to assign reduction points to Player A.
the total number of points from Player A by 3. Player C
Finally, player B will be informed of how much reduc-
is free to assign any number of reduction points to Player
tion points Player C has assigned to Player A.
A, either zero of all 50 points that were initially assigned.
Please press “Continue” to start with the interaction.
[EXAMPLES].
So, Player A will ?rst decide how to distribute the 100
points. As soon as Player A has made a decision, Players
Appendix B: Instructions for Experi- B and C will be informed thereof. Next, Player C may
decide to assign reduction points to Player A.
ment 2
Finally, player B will be informed of how much reduc-
tion points Player C has assigned to Player A.
Dear participant,
To Player C: Please not that in some instances, the
Welcome to this study on social interactions. Please
computer will change the number of points allocated by
read the following instructions carefully. In case of ques-
Player A to Player B. The computer will randomly sub-
tions or uncertainties, please call the experimenter by
tract or add points to the number of points allocated to
pushing the red button on the intercom.
Player B. It may also be the case that the number of points

---