Punishment and Compensation in Children
THE Consequences of Trivial Group Affiliation on Punishment and Compensation in Children
This study looked at how elementary school children (N = 38) responded to unfairness in a group context. Children were assigned trivial groups and witnessed unfair sharing in a dictator game. Given the option to make reparations, punish, or do nothing, children had a strong preference for compensation over punishment, regardless of group membership. Children were more likely to make reparations when the person harmed in the dictator game was a member of the in-group rather than the out-group. Similarly, they were less likely to punish when the person harmed was a member of the in-group than the out-group. Interactions between actor group and recipient group were insignificant.
The ability to distinguish between groups of people is a key concern for developing children. They engage with a variety of social actors and must be able to differentiate those who are similar to them and those who are not. Social space is thus divided into the in-group, and assortment of actors like them, and the out-group, a collection of dissimilar others (Dunham et al., 2011).
Early in life, children learn how moral obligations hinge on social membership. For young children, knowing group membership indicates an individual’s rights and obligations more than it does their preferences or behavior (Kalish & Lawson, 2008). Children understand social categories as marking people who are intrinsically obligated to one another (Rhodes and Chalik, 2013). In some sense, people have an obligation to treat members of their in-group more positively than out-group members. Dunham showed that even minimal group assignment caused children to be more favorable towards their in-group in terms of implicit and explicit measures of attitude and resource allocation (Dunham et al., 2011).
While there is a large body of research that shows that children have a general concern for fairness, this might be mitigated by group relationships. Many studies have found that people are willing to incur a personal cost to avoid unfairness among other people (Dawes, Fowler, Johnson, McElreath & Smirnov, 2007), and this willingness to assume costs to obtain fairness develops in childhood, with three-year olds being relatively unwilling to incur costs to avoid unfairness and eight-year olds being much more willing to do so (Fehr, Bernard & Rockenbach, 2008). In a study related to ours, children were willing to give up some of their candy to punish an unfair actor in a third-party punishment game on around 30% of the unfair trials they viewed (J. Jordan, personal communication, 2014). But children’s concern for fairness consists of two components: a desire to be fair and a desire to be perceived as being fair. Children are more likely to act unfairly when they could be unfair without others knowing (Shaw et al., 2012), and so we see that a child’s sense of morality is not absolute. Children are not pure egalitarians. Their judgments of morality and fairness depend on social contexts, and this idea extends to intergroup relations as well.
Our study aims to look at how children respond to immoral acts, specifically unfair acts, in a group context. Gummerum attempted to look at the effect of group membership on altruistic behavior in children but found that group membership did not serve as a salient social cue that influenced altruistic behavior (Gummerum, 2009). But Gummerum openly admits that perhaps the young children in her study did not interpret the trivial social categories of color groups as salient enough markers to distinguish an in-group from an out-group. Her study can be critiqued on that front: it used a dot estimator task to randomly assign children to color groups. Since children were randomly assigned to colors, they might not have felt much attachment to their group. If the child’s color preference did not line up with the color of their group membership, you would not expect the intergroup differentiation to be as prevalent. When Jordan assigned children to color groups based on their own color preference, she found a main effect of actor group (children were less likely to punish when the actor was in the in-group) and a main effect of recipient group (children were more likely to punish when the recipient was in the in-group)(J. Jordan, personal communication, 2014).
Gromet and Darley (2009) argue that people’s default reaction to injustice is to punish the perpetrator. Punishment is norm enforcing, teaching unfair actors that their behavior is unacceptable and attempting to dissuade them from behaving similarly in the future. However, creating equality need not be restricted to punishment; making reparations, or compensating the victim, also levels the playing field. Making reparations might come out of feelings of deservingness or a desire to reaffirm community values of equality (Leliveld & Beest, 2012). In our study, we seek to understand the relationship between these punishment and reparations by pitting them against each other in a third-party dictator game.
We look at overall rates of punishment and reparations. Jordan’s study found that participants punished an unfair act in around 30% of the trials (Jordan, personal communication, 2014). However, her participants had only two options: punish or do nothing. We wish our analysis to be similar, so we translated our three choices into a binary decision – punish or not punish, make reparations or not make reparations.
Our equations are the following, where p = proportion of punishment, r = proportion of reparations, n = proportion of trials in which participant does nothing.
p + r + n = 1
p = 1 – r – n
r = 1 – p – n
If our results are similar to Jordan’s, but the introduction of reparations has no effect, we would expect p = .3, r = 0, n = .7. This would mean that children did not use reparations as a way to try to impact the behavior of the actors. This is not likely to be the case; children will likely make reparations, in which case, r will come out of p, n, or both.
If the rate at which children decide to act after an unfair trial is similar to that found in Jordan’s study, n will stay at around .7. Perhaps their decision between reparations and punishment will be completely random, as both punishing and making reparations creates equality after an unfair trial. Thus we would expect p = .15, r = .15, n = .7. However, punishment and reparations differ in terms of the implicit moral message and the social context. The motivation to punish might come from a desire to teach the actor to not share unfairly. By making reparations, the actor is off the hook. In some ways, the actor “got away with” their unfair act even if the end scenario is equality. However, there might be something intrinsically pleasant about introducing more resources into the game instead of removing them. There is a social cost to punishing if others around you are not exposed to the moral reason why you are punishing. If others did not see that you were punishing an unfair act, you seem like the unfair actor. The social cost of reparations is much smaller because you appear generous in any case, and for this reason, compensation will likely be more prevalent than punishment.
Since we’ve added another option – making reparations – the overall rate at which a child does nothing (n) will be lower than what was found in Jordan’s study, thus I predict n < .7; p + r > .3; r > p.
We also look at the effect of group membership to investigate whether group affiliation affects a child’s moral sense and changes how the rates of punishment and reparations. In one study, children were more likely to make reparations for (or help fix) a broken toy if the person who broke the toy was in the in-group (Over, 2014). In a similar vein, we expect that there might be some interactions between the actor group and the receiver group for reparations and punishment.
We look at the interactions of group membership according to the 2 x 2 matrix design as shown in the diagram below:
In Situation I, where both the actor and receiver are in the in-group, we predict higher rates of reparations: children will likely choose to increase the number of resources of the in-group as opposed to taking them away via punishment.
In Situation II, where an in-group member acts unfairly towards an out-group member, we predict higher rates of reparations. Children might be less likely to punish the in-group because that means taking resources away from the in-group. Reparations would be akin to making an apology for the in-group, thus creating equality and allowing the in-group to keep resources at the same time. However, it could also be possible that people are more likely to punish in Situation II because of the black sheep effect. We do not want to be associated with an unfair in-group, so we harshly punish their unfair acts (Marques & Yzerbyt, 1988).
In Situation III, where an out-group member acts unfairly towards an in-group member, we predict higher rates of punishment and lower rates of doing nothing. Children might feel more strongly about defending their in-group, especially when the unfair actor is in the out-group, so they would be more likely to act after seeing this interaction. But they might want retribution. They might want to get the out-group member back for harming one of their own, in which case, the defense would likely take the form of punishment.
In Situation IV, an out-group member acts unfairly towards another out-group member. In this case, we expect third party observers to be more likely to do nothing. This interaction might be seen as an out-group problem that a member of the in-group need not be involved in: if it’s not your in-group, you might not feel the need to take responsibility for their actions or teach them to act otherwise.
We also look at the number of resources distributed to in-group and out-group members. Perhaps people subtly punish and make reparations in a way such that the in-group gets more resources. Children might want to create the appearance of being fair and even-handing while giving a slight but significant advantage to their in-group.
The sample consisted of 42 children (22 boys and 20 girls; mean age = 6.38, SD = 1.08). Early participants were recruited via telephone from a laboratory maintained database at Yale University and tested in the laboratory. Later participants were recruited and tested at the Yale Peabody Museum in New Haven, Connecticut. Parental consent was secured before all testing, and participants were compensated with stickers.
Of the original sample of 42 children, four were excluded from the results because they failed to identify with their minimal group when choosing their avatars (see below). The final sample thus consisted of 38 children (19 boys and 19 girls; mean age = 6.47, SD = 1.06).
Participants were tested individually. For early of the trials, only one experimenter was involved. However, for most of the trials, there were two experimenters, one who interacted with the children and read from the script and another who helped with logistical operations (acting out the child’s choice using stickers and envelopes, removing stickers and envelopes for following trial, etc).
Color Preference Task: To establish membership in a trivial social category, participants were presented a green crayon and an orange crayon and asked to pick which they liked better. Group membership hinged on crayon color preference. Those who liked the green crayon were placed in the green group and given a green sticker to wear on their chest. Those who liked the orange crayon were placed in the orange group and given an orange sticker to wear on their chest.
Avatar Task. The avatar task was designed to test whether participants understood their group membership. Participants were presented with two printed and laminated characters. The gender of the characters matched that of the participant, and the only difference between characters was the color of the circle on their shirt. We asked participants to choose an avatar that looked like them. We excluded the results of those who failed to choose the avatar of the same color group.
Once participants chose an avatar, they were given a bucket of 22 identical stickers. Experimenters explained that they were allowed to take home the stickers left in their bucket at the end of the game. Participants were then asked to attach their avatar to their bucket of stickers.
Participants then were introduced to the third-party dictator game, played using a mixed display of physical and technological elements. The exact setup can be observed in Figure 1. Participants were told that other children had come in earlier; some had four stickers, some had none. Those with four stickers had a chance to give away some of their stickers to the children with none. Participants were shown how the other children decided to share using avatars on an iPad display, and were told that they were going to be the judge and decide what to do about the sharing.
In each trial, the experimenters placed two envelopes in front of the iPad, one on the left and one on the right, corresponding to the avatars displayed on the left and right side of the screen on the device. The envelope on the left had four stickers because the avatar on the left represented the child who starts with four stickers. The envelope on the right had no stickers because the avatar on the right represented the child who starts with no stickers. The experimenters then removed the stickers from the left envelope and showed how the stickers were shared. In the fair 2-2 sharing trials, two of the stickers stayed on top of the left envelope, and experimenters moved the other two stickers on top of the right envelope. In the unfair 4-0 sharing trials, all of the stickers stayed on top of the left envelope.
In front of the iPad display, there were three buckets: the “gets none” bucket, the “gets four” bucket, and the bucket of their stickers with their avatar attached. For each interaction they saw, the participants had three choices: they could do nothing, keeping all of their stickers in the bucket with their avatar; they could “punish” by putting one of their stickers in the “gets none” bucket to take away all of the stickers from the actor, leaving both the actor and receiver with no stickers; or they could “make reparations” by putting one of their stickers in the “gets four” bucket to give the receiver four, ensuring that both actor and receiver had four stickers. Or they could do nothing.
Experimenters enacted the results of the participants’ decisions. If participants chose to put a sticker in the “gets none” bucket, all of the stickers from the left envelope and the sticker placed in the “gets none” bucket were thrown in the trash, and envelopes were removed. If participants chose to put a sticker in the “gets four” bucket, the experimenters placed four new stickers on the right envelope, packed up and removed both envelopes.
All participants were tested on their comprehension of the game before they started making decisions. They were shown two practice trials. In the first trial, the actor kept all four of the stickers, leaving the receiver with none. We then asked the subjects to make the actor have none. In the second practice trial, one of the characters kept all four of the stickers, leaving the other with none. We asked subjects to make the receiver have four. The sample scenarios were also accompanied with cartoon avatars with no group affiliation.
After indicating their understanding of the rules in the comprehension task, participants viewed twelve interactions between actors and respondents. In these trials, the avatars were randomized both in terms of group membership (indicated by a large colored circle on the avatar’s shirt) and appearance (hair color, glasses vs. no glasses, etc.) such that each avatar was a novel character. Group membership of the actor and respondent was randomized. Actors either allocated the four resources fairly, giving the receivers two stickers, or they allocated the resources unfairly, keeping all four stickers. Children responded to the twelve dictator game scenarios while data was recorded on the iPad using Qualtrics software.
As reported earlier, our analysis excluded participants who failed the group check measure. A set of preliminary analyses was carried out to examine gender and age group effects (5-6 year-olds compared to 7-8) and found no significant effect of either, thus the data were collapsed across gender and age group. The findings are reported accordingly.
(1) Overall Rates of Punishment and Reparations
We first collapsed across actor and receiver group and compared the overall rates of punishment and reparations using a paired-samples t-test. Overall rates are presented in Figure 2. The analysis revealed that participants were much more likely to make reparations (M = .523, SD = .265) than punish (M = .220, SD = .200) in unfair trials, t(37) = 4.724, p < .001. When children saw an unfair interaction where an actor gave 0 of 4 stickers to the receiver, they were more than twice as likely to make reparations than to punish. As predicted, the rate of doing nothing (M = .257, SD = .253) was lower than what Jordan found.
(2) Rates of Reparations versus Actor and Receiver Group
We then looked at the rates of punishment depending on actor and receiver group using a repeated-measures ANOVA with two within-subjects factors (actor group: in-group or out-group; receiver group: in-group or out-group) and two between-subjects factors (gender: male or female; age group: 5-6, 7-8). Rates of reparations versus actor and receiver group are displayed in Figure 3. The predicted interactions between actor group and receiver group did not play out in our study results. There were no significant interactions of actor and receiver group. However, there was a medium to large significant main effect of recipient group with participants being more likely to make reparations when the recipient was a member of the in-group (M = .590, SD = .358) than no (M = .461, SD = .375), F(1,34) = 5.436, p = .026, partial eta-square = .138. Regardless of who was acting unfairly, children were more likely to make reparations to an in-group member. If an in-group member gets harmed, participants were more likely to make up for the initial unfairness by giving that in-group member their proper share.
(3) Rates of Punishment versus Actor and Receiver Group
We used a similar repeated-measures ANOVA with two within-participant factors (actor group: in-group or out-group; receiver group: in-group or out-group) and two between-subjects factors (gender: male or female; age group: 5-6, 7-8) to examine the rates of punishment depending on actor and receiver group. Rates of punishment versus actor and receiver group are displayed in Figure 4. Much like the analysis of rates of reparations, there were no significant interactions of actor and receiver group. However, there was a large, significant main effect of recipient group with participants being less likely to punish when the recipient is a member of the in-group (M = .162, SD = .276) than not (M = .267, SD = .291), F(1, 34) = 5.034, p =.031, with a partial eta-square = .175. Regardless of who was acting unfairly, children were more likely to punish if the person harmed was a member of the in-group. This is likely related to the rates of reparations because if a participant made reparations, they could not also punish. Thus if they were more likely to make reparations when the receiver was in the in-group, they would also be less likely to punish when the receiver was in the in-group.
(4) Difference between Punishment and Reparations versus Group Membership
A similar repeated-measures ANOVA was performed to look at the difference between punishment and reparations depending on actor and receiver group membership. The difference in rates of punishment and reparations versus actor and receiver group are displayed in Figure 5. The predicted interactions of actor and receiver group were not supported by the data. However, there was a large significant main effect of recipient group, with much higher rates of reparations than punishment when the recipient was a member of the in-group (M = -.422, SD = .528) than not (M = -.194, SD = .620), F(1,34) = 7.234, p = .011, partial eta-square = 175. When children saw an unfair interaction, they were much more likely to make reparations than punish if the person harmed was a member of their own group. Rates of punishment were closer to rates of reparation when the recipient was a member of the out-group.
(5) Overall Sticker Count
A paired-samples t-test was used to examine the total number of stickers allotted to the in-group and the out-group. Total stickers was calculated by summing the number of stickers given to the groups when children made reparations and then subtracting the number of stickers removed from the group when children punished. There was no significant difference in the average number of stickers the in-group had at the end of the trials (M = 25.95, SD = 1.50) compared to the out-group (M = 26.47, SD = 1.18), t(37) = -.436, p = .665. In terms of sticker totals, children did not show a preference towards their in-group.
In many ways, moral obligations depend on social membership, even minimal group membership (Dunham et al., 2011). In this study, we were interested in how group dynamics affected how children responded to unfair acts. We gave children the option to either punish the actor by taking away resources or compensate the victim by making reparations. Offering both options allowed us to compare two strategies for creating equality.
There is a large body of research that suggests that people are willing to incur a personal cost to punish unfair actors. But, as our study suggests, perhaps that is only the case when the options are limited to punishing or doing nothing. We found that children vastly preferred to make reparations than punish in unfair trials, and this is consistent with other adult research about compensation (Chavez & Bicchieri, 2013). Punishment may have larger social costs than making reparations. Imagine the following social scenario in which you are observing others’ behavior, and they are observing yours. If you witness an unfair interaction, you might altruistically punish at a cost to yourself to provide a public good (Fehr & Gacheter, 2002). If other observers see both the original unfair act and your altruistic punishment response, your action might be seen as in the right. However, if they did not see the original unfair act that you are punishing, you might appear as the unfair agent. This social cost is lessened if you compensate rather than punish because in either scenario, you would be seen as generous. Thus, the preference for reparations over punishment might have been the result of a worry about the reputational cost of punishing.
We found a main effect of recipient group for both compensating and punishing. Children were more likely to make reparations in response to an unfair act when the receiver was in the in-group. Children were more likely to compensate a victim in the in-group than a victim in the out-group. Perhaps seeing an in-group member get harmed was more emotionally salient than seeing an out-group member get harmed, so children were more generous when receiver was in the in-group. This finding is consistent with other displays of in-group preference found in Dunham’s study (2012).
There was also a similar main effect of recipient group on rates of punishment, with children being less likely to punish when the recipient was in the in-group. This makes sense given the preference to make reparations when the recipient was in the in-group. Since there were three options (punish, make reparations, or do nothing), an increase in likelihood of reparations would also mean a decrease in one or both of the other two options. Considering that rates of doing nothing were stable across actor and receiver group membership, an increase in the rates of reparations would be accompanied by a decrease in the rates of punishment.
However, while we predicted that there would be interactions between group membership of the actor and recipient for reparations and punishment, we found none. According to Gummerum (2009), young children “develop in-group bias for social categories that are commonly used to sort people into groups (e.g., gender, ethnicity, caste) in their specific culture.” Possibly our group manipulation was too subtle or not salient enough to detect more precise group differentiations. We assigned children to groups based on their color preference. We chose not to randomly assign them to color groups to avoid any sort of pre-existing color bias affecting their decisions. But in defining groups in this way, our study suffered from a relatively imbalanced group membership, with children choosing to be in the orange group more than twice as often as the green group. And while our samples were balanced in terms of gender and age, we had a relatively small sample size. When analyzing reparations and punishment across actor and receiver group, we had a sample size of 38. However, when looking at between-subjects variables such as gender and age, our sample was too small to make strong conclusions. For example, we could only compare nine boys to compare to ten girls in the 5 to 7 age range. Future studies would want a larger and more balanced sample size.
Whereas our study establishes that children would rather incur a cost to compensate than to punish, we did not look at how strong that preference was. Perhaps when reparations costs X times as much as punishment, children will punish and compensate equally. Future research could vary the cost to punish and make reparations in order to examine this relationship more closely.
One could also look at the between subjects factors that affect reparations and punishment in children. Leliveld (2012) argues that the willingness to altruistically compensate might be related to empathetic concern. Empathy is defined as including affective capacity to share in another’s feelings and the cognitive ability to understand another person’s feelings and perspective (Garton & Gringart, 2005). “Whereas people low in empathetic concern may disregard the feelings of others, people high in empathetic concern seem more inclined to react to how others feel” (Leliveld, 2012).
In Leliveld’s study, highly empathetic people were more likely to compensate the victim in a third-party punishment game, whereas low empathetic people were more likely to punish the offender (Leliveld, 2012). Future studies could look at whether there are dispositional characteristics that make a child more likely to choose one option over the other. More specifically, they might want to evaluate whether empathetic concern affects how children respond to unfairness.
Chavez, A.K., Bicchieri, C. (2013). Third-party sanctioning and compensation behavior:findings from the ultimatum game. Journal of Economic Psychology, Vol. 39, 268-277.
Dawes, C., Fowler, J., Johnson, T., McElreath, R., & Smirnov, O. (2007). Egalitarian motives in humans. Nature, Vol. 446, 794-796.
Dunham, Y., Baron, A.S., & Carey, S. (2011). Consequences of “minimal” group affiliations in children. Child Development, Vol. 00, 1-19.
Fehr, E., Bernhard, H., & Rockenbach, B. (2008). Egalitarianism in children. Nature, Vol. 454, 1079-1083.
Fehr, E. & Gachter, S. (2002). Altruistic punishment in humans. Nature, Vol. 415, 137-140.
Gromet, D. M., & Darley, J. M. (2009). Punishment and beyond: Achieving justice through the satisfaction of multiple goals. Law & Society Review, Vol. 43, 1–38. Retrieved from http://onlinelibrary.wiley.com/doi/10.1111/j.1540-5893.2009.00365.x/full
Gummerum, M., Takezawa, M., Keller, M., (2009). The influence of social category and reciprocity on adults’ and children’s altruistic behavior. Evolutionary Psychology. Vol. 7(2), 295-316.
Marques, J.M., Yzerbyt, V.Y. (1988). The black sheep effect: judgmental extremity towards in-group members in inter- and intra-group situations. European Journal of Social Psychology. Vol. 18, 287-292.
Kalish, C.W., & Lawson, C.A. (2008). Development of social category representations: early appreciation of roles and deontic relations. Child Development. Vol. 79(3), 577-593.
Leliveld, M.C., Vandijk, E., Vanbeest, I. (2012). Punishing and compensating others at your own expense: the role of empathetic concern on reactions to distributive justice. European Journal of Social Psychology, Vol. 42, 135-140.
Over, H. 2014. A preliminary investigation of collective responsibility. Unpublished Manuscript.
Rhodes, M., & Chalik, L. (2013). Social categories as markers of intrinsic interpersonal obligations. Psychological Science. doi:10.1177/0956797612466267
Shaw, A., Montinari, N., Piovesan, M., Olson, K.R., Gino, F., & Norton, M.I. (2012). Children develop a veil of fairness. (Working Paper No. 13-003). Retreived from Harvard Business School website: http://www.hbs.edu/faculty/Publication%20Files/13-003.pdf.
 The gender of the avatar always matched the gender of the participant to avoid gender bias.