Introduction

Punishment is pervasive throughout many societies1. Decades of research in behavioral economics and psychology empirically demonstrate that humans exhibit a strong desire to punish when deciding how to restore justice, even when it is costly2. However, more recent work reveals that if non-punitive alternatives are made available, punishment is not always systemically endorsed3,4,5. This suggests that preferences for punishment may be limited, perhaps to contexts where it is the only available option for restoring justice. Although the desirability of punishment has been questioned6,7,8, it remains unclear what features of a moral transgression motivate a punitive versus non-punitive response. Here, we explore this question, investigating when—and why—punishment is preferred as an instrument for restoring justice.

The desire to punish appears to lie either in its ability to change unjust behavior through deterrence9 or alleviate negative emotions triggered from being treated unfairly10,11. In either case, punishing a transgressor can reap positive benefits. There are, however, several negative consequences associated with responding punitively. For example, punishment can erode cooperation by turning prosocial cooperators into antisocial punishers2,12,13. In other cases, perceived unjustified punishment leads individuals to retaliate with retributive behavior, and this cycle of retribution collectively reduces the welfare of the group14,15. These findings together suggest that punishment may have mixed desirability depending on the context, and thus may not be a uniformly preferred method for restoring justice.

In contrast, non-punitive options have shown to be powerful motivators for restoring justice16,17,18. These non-punitive options typically address the victim’s needs through compensation (operationalized within the laboratory by endowing the victim with money) instead of focusing on punishing the perpetrator. Compensating the victim has many real-world analogues. For example, insurance companies pay reimbursement for stolen or damaged goods, and programs such as the Office for Victims of Crime allocate money (e.g., more than $2 billion in 2016) to help more than 5 million victims who have been impacted by crime19. Furthermore, monetarily rewarding good behavior—compared to sanctioning norm violators—appears to be one effective route for promoting cooperation20,21,22. Other research examining how rewards and sanctions differentially encourage cooperation reveals that rewards can produce better outcomes for the group compared to punishment alone3. Although these results suggest that, depending on the context, non-punitive measures may be viable alternatives, the boundary conditions (i.e., is there a tipping point for how much compensation is required before punishment is no longer desired?) remain unknown. This inconsistent evidence naturally begs the question of when punishment is the preferred avenue for restoring justice. One possibility is that the decision to punish hinges on the context of the moral infraction23. One naturally-occurring context that may shape punishment preferences is when punishment is decided by a victim versus an impartial third-party5,24. In these cases, it is conceivable that different moral signals are generated depending on the perspective of the person deciding to punish, such that the act of punishing can either reap reputational benefits or damage one’s reputation. For example, when deciding to restore justice as a victim, punishment (such as gossiping about or publicly blaming norm violators, deciding to ostracize transgressors, etc.) may be perceived as retributive or vindictive while non-punitive responses may embody the doctrine of “turning the other cheek”, conveying that one values forgiveness25. If so, the reputational information gleaned from a victim deciding not to punish may signal that an individual has a positive moral character. In contrast, if a third-party member fails to punish, it may be interpreted as condoning the transgression, which would result in missing out on potential reputational advantages associated with punishing26,27. Simply put, punishment may act as a conduit for signaling different moral values depending on the context in which it arises and the perspective of the punisher. To test these hypotheses and identify the social contexts in which preferences for punitive responses (as opposed to non-punitive alternatives such as compensation) are preferred, we leverage both behavioral economic games and crime vignettes. We fully parameterize the decision space to precisely measure the point at which people make trade-offs between punishment and compensation. If punishment is not a preferred method for restoring justice, then it should be less appealing as the victim’s needs are met through compensation. Indeed, it is possible that once the victim’s needs are sufficiently met, no amount of punishment—however small—will be preferred. In contrast, only when the victim’s monetary needs are not met should punishment become the preferred response. Additionally, we predict that the reputational value associated with decisions to punish will be modulated by the situation (i.e., deciding as a victim or a third-party). If punishment provides a greater positive reputational moral signal when deciding as a third-party, then third-party punishers should be preferred compared to punitive victims. Experiment 1: Compensating the Victim Abolishes Preferences to Punish Perpetrator Method Participants 150 participants (48 women, 1 participant whose gender was unknown; age = 33.1 years, SD ± 10.6) were recruited through Amazon’s Mechanical Turk28 (AMT) to play a modified Justice Game5 (JG) and were paid$1.50 for completing the task, as well as a bonus determined from their decision on one randomly selected trial.

For all experiments: (1) participants played anonymously over the Internet and were not allowed to participate in more than one experimental session; (2) all studies were approved by Brown University’s Institutional Review Board; (3) all the methods were performed in accordance with the relevant guidelines and regulations of the institution, and informed consent of all participants was obtained; (4) we only analyzed the data of subjects who had unique Internet protocol addresses to avoid duplicate respondents; (5) all analyses were conducted in R29 and mixed-effects models were made with lme430. (6) the desired sample sizes recruited for all studies were calculated using G*Power 3.131. For all studies employing an economic game (Experiments 1, 2, and 4), our sample sizes were based on effect sizes (Cohen’s d = 0.28) observed in previous studies using a similar design5, which suggest than an approximate sample size of 110 participants (alpha = 0.05, beta = 0.90) was sufficient to detect preferences for compensation over punishment; (7) We collected a variety of post-experimental questionniares that were not analyzed for this manuscript. These include: Interpersonal Reactivity Index32, Social Value Orientation33, Delayed Discounting34, Intolerance of Uncertainty Scale35, and demographic data. For all studies we have reported all measures, manipulations, and participant exclusions. Finally, (8) we used Amazon’s Mechanical Turk to recruit all participants, whose subject pool has been shown to be representative of the broader population in the United States36.

Participants played a modified version of a well-vetted economic paradigm known as the Justice Game5,37 (JG). In the JG, participants are paired with a unique anonymous player every trial and the two players must agree on how to split a sum of money. As the first mover, Player A is endowed with one dollar and proposes a division of this money with the participant, Player B (Player A: $1-x, Player B: x), in increments of$0.10 (prior work reveals incentive size—$1 versus$10—does not affect behavior5). We restrict offers from Player A (unbeknownst to participants, predetermined offers from a computer) to varying levels of unfairness, equally ranging from moderately unfair ($0.60,$0.40) to highly unfair splits ($0.90,$0.10). On each trial, participants, as Player B, respond by choosing between two options: (1) compensation: a non-punitive option where the victim deserves recompense—such that Player B is compensated some monetary amount while Player A’s payout remains unaffected; and (2) punishment: a ‘just deserts’ option where the perpetrator deserves punishment for the wrong committed—such that Player A is punished through monetary reduction while Player B is compensated38.

Accordingly, there are two trial types: partial compensation and partial punishment. In partial compensation trials, participants (Player B) must choose between maximal punishment to Player A and some amount of compensation to oneself (Fig. 1a). As an example, after receiving a highly unfair offer ($0.90,$0.10), the partial compensation option could increase Player B’s payout to some amount greater than $0.10, while not changing Player A’s payout ($0.90). This option is always pitted against maximal punishment, which fully reduces Player’s payout (to $0.10) and fully enhances Player B’s payout (to$0.90). In other words, participants choose between maximally punishing Player A at no cost to themselves, versus accepting some amount of compensation to avoid punishing Player A. Intuitively, when presented with these options, maximal punishment should be treated as the default, and thus this first experiment seeks to answer the question of how much compensation is required to shift people away from punishing?

Conversely, in partial punishment trials, Player B has the option to enhance their own payout while only partially reducing Player A’s payout. Effectively, they can choose to punish Player A for an unfair offer (partial punishment) or receive maximal compensation and forgo any punishment towards Player A. Critically, Player B’s payout is constant in these trials regardless of their choice and the only aspect that differs is how much punishment could be applied to Player A (ranging from minimal to maximal). This task structure allows us to determine the preference placed on applying punishment to Player A versus compensating oneself. Because partial punishment trials vary the amount of punishment available, the question is simply how much punishment is required (if any) to restore justice?

Participants completed 80 fully randomized, one-shot, self-paced trials: 20 partial compensation trials and 20 partial punishment trials deciding as the victim (Player B) and the same 40 trials deciding as a third-party (a within-subject design; see supplement for task instructions, full methodological details, and results of the third-party condition which show a similar pattern of behavior).

Results

Preferences shift from punishing the perpetrator to compensating the victim

Given that punishment has been shown to be a highly preferred response to severe fairness violations1, we initially restricted our analysis to the most unfair offers ($0.90,$0.10). Using a logistic mixed-effects model, we tested our hypothesis based on previous work5,37 that increasing compensation for the self would result in less punitive behavior towards Player A (we report maximal models for all analyses39,40). In partial compensation trials with minimal compensation, participants unsurprisingly prefer maximal punishment since that option maximizes one’s own payout (Fig. 1b). However, participants become significantly less punitive as compensation to themselves increases ($$Odds\,Rati{o}_{compensation}$$ = 0.40; Table 1). In fact, once maximal compensation is available, the endorsement of punishment drops by almost 50 percentage points (Fig. 1b). This illustrates that participants’ preferences to punish a perpetrator attenuates as their own compensation increases. Replicating previous work5,37, once participants are fully compensated, punishment is no longer the most preferred response. This effect was observed for all fairness levels (see supplement).

Preferences to punish do not increase if there is ample compensation

We next tested whether manipulating the degree of punishment in partial punishment trials can shift participants’ preferences in the wake of a fairness violation. That is, if a victim’s needs are already met through the maximal compensation option, how much punishment (if any) is endorsed? Contrary to the idea that there are appropriate amounts of punishment specific to certain degrees of infraction38 (i.e., a $0.90,$0.10 offer), we observed no increased desire to punish so long as participants received maximal compensation ($$Odds\,Rati{o}_{punishment}$$ = 0.91; Fig. 2b; Table 2). Simply put, given a specific infraction, participants were insensitive to the degree of punishment toward Player A, even in the face of highly unfair offers ($0.90,$0.10; see supplement for the results for other fairness levels, which all mirror the same behavioral pattern). These results indicate that the preference for punishment is non-dominant as long as one is sufficiently compensated, even when punishment is free and easy to implement.

Experiment 3: Preferences for Compensation Generalize to Severe Moral Transgressions

In Experiment 2, when participants were given full control over the final redistribution outcomes, we again found evidence that they preferred to compensate themselves more than punish the perpetrator. However, it remains unclear if these preferences exist only in contexts where there are relatively minor violations (i.e., unfair splits of one dollar in one-shot economic games), or if they generalize to other more egregious transgressions. In Experiment 3, we explore this question, investigating whether preferences to punish and compensate scale with increasingly severe moral transgressions. While past research illustrates that people endorse the doctrine of “proportional punishment” (such that punishment scales with the severity of the crime committed38), it remains unknown whether compensation to the victim also scales as moral violations become increasingly pernicious. Given our findings in Experiments 1–2 that compensation can rebalance the scales of justice, we posited that people should increasingly endorse compensation as transgressions become more severe.

Method

Participants played a two-stage economic game, which began with a variant of the JG. In this version of the JG, after Player A makes an offer, Player B can reapportion the money by choosing between two randomly selected options drawn from the following six options (Fig. 5a): (1) Accept: agree to the proposed split ($1-x, x); (2) Equity: split the monetary pie equally so that both players receive half of the initial endowment ($0.50, $0.50); (3) Compensate: increase Player B’s own payout to equal Player A’s payout, thus enlarging the pie to maximize both players’ monetary outcomes ($1-x, $1-x); (4) Punish: reduce Player A’s payout to the original amount offered to Player B (x, x); (5): Reverse: reverse the proposed split, the ‘just deserts’ motive where Player A deserves punishment proportional to the unfairness of the offer; and finally, (6): Reject: both players receive nothing ($0, $0). The first three options can be considered more prosocial in nature, with little or no punishment of Player A; while the last three options can be considered more antisocial in nature, where punishment acts as a key ingredient for restoring justice. By adding the Reject option (which was not available in the original JG version5), we ensure that there is an equal probability of being presented with a prosocial option (Accept, Equity, and Compensate) or antisocial option (Punish, Reverse, and Reject). In other words, because Player A does not know which two options will be given on any given trial, having an equal number of prosocial and antisocial options buffers against any interpretation that Player A could be strategically giving unfair offers. Here, we use these labels—prosocial and antisocial—to differentiate options not involving punishment versus those that do (although these labels were never presented to the subjects). Participants are only given two options on any trial, such that each option is randomly paired with one alternative option, resulting in every combination pair for a total of 15 unique pairs. In this first phase of the game, we have participants play the JG so they can learn all the psychological motivations behind each redistribution option. In a subsequent partner selection phase (Fig. 5b), participants are informed they will be choosing their partner for two future economic games (unknown to participants, the experiment concludes after the partner selection phase ends). The only information participants are given to inform their choice is their potential partners’ past behavior in the JG. For example, participants might be informed that a partner’s most endorsed option was “Reverse”, while another partner preferred “Equity”. Unlike the previous experiments, here, we label each of the options (i.e., Accept, Equity, Compensate, Punish, Reverse, and Reject) to reduce the cognitive load on having to infer the option name from monetary redistributions. Using a between-subjects design, participants either completed the two-stage economic game from the perspective of the victim, in which participants play the game as Player B, or from the perspective of an impartial third-party, termed Player C. When playing as Player C, participants are asked to make decisions on behalf of an anonymous Player B. Unlike decisions as the victim, third-party decisions are non-costly and non-beneficial since the final choice only influences the monetary outcomes of Players A and B. The critical test is whether those who punished would be selected as a social partner at the same rate regardless if they made the decision as a victim or a third-party. In addition, we were interested in examining the boundary conditions for when punishment would be perceived as a positive moral signal. We posited that third-party punishers may not always be systematically preferred as social partners, and that this preference will depend on the demands of the social situation. Accordingly, participants were tasked with selecting partners for both the Trust Game43 (TG) and the Dictator Game44 (DG). For the Trust Game, participants were selecting partners who would be the second mover, which indexes the perceived trustworthiness of their partner. When selecting partners for the Dictator Game, participants chose partners who would be the first mover, which indexes the perceived generosity of their partner. We chose these two games on the assumption that the reputational benefits associated with punishing might shift between these different social contexts. Take for example previous work illustrating that third-party decisions to punish are taken as a positive signal of one’s trustworthiness23: those who are willing to punish norm violators are seen as more likely to observe the norm of reciprocity (i.e., reciprocating trust) in the Trust Game. There may be other social contexts, however, in which punishing is not perceived as a desirable behavior. It is possible that punitive partners would not be preferred in contexts that require generosity or altruism because the original punishment may construed as spiteful45. Rather, someone who behaved prosocially in the Justice Game—for example, by equally redistributing the money or compensating without punishing—would be more highly sought after than someone who punished the perpetrator. To test this, we utilized a similar pairwise comparison design used in Experiment 1 and fully parameterized the space by comparing each partner type against all others. All participants completed 15 self-paced partner selection trials for the TG and 15 self-paced partner selection trials for the DG, and the order of these games was counterbalanced. Results When deciding as a victim, punishment is a poor signal of trustworthiness We compute partner selection by the number of times a partner who preferred a particular option (e.g., a Compensator) is chosen for a Trust Game compared to every possible alternative partner (i.e., Accepter, Equity Maximizer, Punisher, Reverser, and Rejecter). Participants showed a strong preference for engaging with victims who responded prosocially (i.e., selected Accept, Equity, or Compensate options, compared to the other three antisocial partners; Friedman’s test, χ2(1) = 88.36, p < 0.001, Wilcoxon Signed-Rank test post-hoc effect size r = 0.93; Fig. 6a), with an especially strong desire for engaging with those who Compensated in the wake of a fairness violation (80%). In contrast, participants least preferred to play with partners who Rejected offers as a victim (11%). These findings offer evidence that choosing prosocial actions as a victim confers strong reputational benefits and offer one possible explanation for why non-punitive decisions may be preferred. When deciding as a third-party, punishment is a relatively better signal of trustworthiness Although participants still showed an overall preference for third-party Compensators (82%), when directly comparing preferences for partners who responded antisocially (collapsing Punish, Reverse and Reject options) as the victim versus as a third-party, we observed a stark difference in how punitive responses shaped subsequent engagement (Fig. 6c). Partners who selected the antisocial, punitive option as a third-party were more likely to be selected as partners for a Trust Game than those who responded punitively as the victim (Welch’s t-test, t(147.32) = 6.98, p < 0.001, g = 0.98). Interestingly, there was a steep and significant reduction in preferring third-parties who ‘Accepted’ an unfair offer on behalf of another (45%; Fig. 5b) compared to victims who did the same (73%; Fig. 5a: Kruskal Wallis test, χ2(1) = 38.54, p < 0.001, r = −0.44). While there is a dominant preference for prosocial third-party partners who compensate, these findings suggest that punishment can send a relatively better moral signal of one’s trustworthiness when that decision is enacted by a third-party compared to a victim. Context modulates perception of third-party punishment We next examined whether the reputational benefit for third-party punishment is stable across different social contexts. While a similar overall preference for prosocial over antisocial partners was also observed in contexts valuing generosity (i.e., in the Dictator Game; see supplement for DG results), there were significant differences in how frequently antisocial third-parties were selected in the TG compared to the DG. Third-parties who chose either to Punish, Reverse, or Reject in the JG (the antisocial options) were less preferred as partners in the DG compared to the TG (paired t-test on combined data from all three antisocial partners, t(99) = −4.47, p < 0.001, g = −0.38; Fig. 6d). When examining each option separately, Bonferroni-corrected post-hoc analysis showed that only third-parties who punished were significantly less preferred in the DG than the TG (paired t-test, t(99) = −8.11, p < 0.001, g = −0.82; Fig. 6d), and there were no contextual differences for those who chose Reverse or Reject. These results reveal that the reputational value of third-party punishment is not stable but shaped by the context of the situation. In other words, there is a significant decrease in preferring punitive partners when the situation values altruistic tendencies (i.e., generous offers in the DG), suggesting that punishment is not always interpreted as a signal of good moral character. In contexts which have strong social norms of reciprocity, such as the TG, it is possible that those who punish send a signal that they would be willing to uphold and enforce these norms. Experiment 5: Lenient Victims and Punitive Third-Parties are Perceived as Moral Experiment 4 demonstrated that victims who punish are not preferred as social exchange partners, especially compared to victims who refrain from punishing. For those who do decide to punish, third-parties are preferred to victims. However, it is unclear whether third-parties and victims who punish are actually perceived as more or less moral, respectively. Accordingly, we ran a final experiment to probe whether those who punish as third-parties and those who refrain from punishing as victims are perceived as moral. In contrast, we posited that third-parties who fail to punish and victims who desire punishment should be perceived as similarly immoral. Method Participants 189 participants (117 women; age = 31.3 years, SD ± 10.2) were recruited through Amazon’s Mechanical Turk. Participants were paid$0.50 for completing the task. Our sample size was based on the effect size (Cohen’s d = 0.26) observed in previous studies examining third-party punishment5, which indicated that a sample size of N = 129 is sufficient to observe this effect with 90% power (α = 0.05).

Similar to Experiment 3, we used a crime vignette to test the reputational effects of a victim or third-party deciding to punish (or not punish) a perpetrator. We adapted a fictional case about vigilantism which details a shoplifting crime46. Using a 2 × 2 between-subjects design, participants read one of the following four vignettes: (1) the victim (store owner) monetarily punished the perpetrator for shoplifting; (2) the victim chooses not to punish; (3) a third-party (judge) monetarily punishes the perpetrator; or (4) the third-party chooses not to punish (see supplement for full descriptions of each vignette). Participants only made a single judgment where they rated the moral character of the victim or third-party, ranging from −50 (Immoral) to 50 (Moral) on a continuous Likert scale.

Results

To examine how participants’ perceptions of moral character differed as a function of who was deciding to punish, we conducted a linear regression where moral character is predicted by role (victim/third-party) and action (punish/did not punish). Results reveal an interaction between role and action: A judge who punishes is perceived to be significantly more moral compared to a victim who punishes, and victims who refrain from punishing are perceived as significantly morally superior to judges who fail to punish (Cohen’s d = 0.60; Fig. 7; Table 5). Both judges who do not punish and victims who do punish are perceived as having neutral moral characters (statistically not different from 0 = ’neutral moral character’, one sample t-test against 0 for third-party: t(48) = 1.39, p = 0.17; for victim: t(47) = −0.18, p = 0.86). Furthermore, judges who punish and victims who refrained from punishing—both of whom were perceived to be high in moral character—were indistinguishable from one another (Welch’s t-test, t(92.26) = 1.04, p = 0.30, g = 0.21). These results accord with the hypothesis that the perception of one’s moral character hinges on whether they are deciding as a victim or third-party: a victim who turns the other cheek, when they could alternatively enact vigilante justice, signals a positive moral character, while a judge gets a positive moral boost by punishing.

Open practices

The data and analysis scripts that support the findings of these studies are avaliable online at the Open Science Framework at the following URL: https://osf.io/ft9bv/

Discussion

Until now, it was unknown when—and under what conditions—people preferred punishment as a means of restoring justice. Our findings are twofold. First, dovetailing with previous work5,37 we demonstrated that once a victim’s needs are sufficiently met through monetary compensation, there is little desire to punish (Experiment 1). Even in a non-constrained decision space, participants freely decide to compensate themselves and administer relatively low amounts of punishment to the perpetrator—despite being able to fully reduce the perpetrator’s payout (Experiment 2). Furthermore, preferences for compensation appear to be a viable method of justice restoration for even the most egregious crimes, and is equally preferred to punishment when the infraction is not too severe (Experiment 3). Second, we observed that the reputational benefits associated with non-punitive and punitive measures are modulated by the perspective of the deciding agent (Experiment 4). Choosing not to punish as a victim sends a positive moral signal that one is likely to be both trustworthy and altruistic. In contrast, punitive responses are preferred when made by a third-party compared to similar punitive responses made by a victim. However, punishing as a third-party is not ubiquitously perceived as a positive moral signal: when a social situation emphasizes generosity, preference for third-party punishment declines. Moreover, lenient victims and punitive third-parties are perceived as having similar positive moral character, while third-parties who fail to punish are perceived as being more immoral (Experiment 5). Together, these results elucidate the boundary conditions of when justice preferences shift between punitive and non-punitive responses, while also providing a mechanism for why punishment is not systematically endorsed across all social contexts.

Our data suggest that there are only some situations where punishment is preferred over non-punitive options. Even if punishment is free and easy to enact, people are less motivated to punish if they are already amply compensated. Moreover, so long as compensation is available, decisions (but not judgments) to punish are largely not proportional to the degree of infraction. While this stands in stark contrast to the large amount of evidence illustrating that people have a strong desire to punish11,38, it is possible that previous work may have inflated punitive preferences because participants were not able to select non-punitive options for restoring justice. Our results suggest that when the transgression is relatively minor, punishment may not be preferred nor needed when non-punitive alternatives are available to rectify the fairness violation. However, future research—such as field experiments—can explore whether preferences for compensation extend to severe moral transgressions that are not hypothetical in nature.

Dovetailing with evolutionary accounts about reputation47,48, we find that how one decides to restore justice can signal information about one’s moral (or immoral) character. Individuals who endorse prosocial responses to moral infractions are perceived as more valued social partners than those who express antisocial preferences for restoring justice. Indeed, endorsing prosocial responses may have an effect on positive emotions and overall wellbeing49. However, the reputational impact of these decisions is modulated by the perspective of the deciding agent. When people decide to punish on behalf of another individual, they are more desired as social partners and are perceived as having better moral character than individuals who punish as a victim. This likely because third-party observers are adhering to and upholding fundamental moral norms, in this case punishing fairness violations23,26,27. Future studies leveraging manipulations can more directly examine whether this relationship is causal in nature.

In contrast, failing to punish or taking no action as a third-party (e.g., accepting the unfair offer without any attempt at restoring justice) may be interpreted as condoning the transgression, such that the individual is neither aware nor willing to uphold moral norms. In these cases, failing to act is perceived as a signal of a questionable moral character. In a similar vein, not wanting to engage with victims who punished perpetrators suggests that such antisocial behaviors are perceived in a negative moral light. For instance, it is possible that victims who punish are perceived as trying to get ‘payback’ or acting on vindictive motives. Together, these results demonstrate that the reputational value associated with punishment is malleable, and critically changes depending on the perspective of the individual enacting justice.

Even the reputational signal of third-party punishment, however, seems to hinge on the social context. In situations where different moral values such as altruism and generosity are prioritized, we find that third-party members who punish are preferred less compared to those who responded in a prosocial manner. One possibility for why third-party punishers are not selected as social partners in a Dictator Game is because decisions to punish provide little information about whether a person values altruism. For example, there does not seem to be a relationship between real world altruistic generosity and decisions to punish50, and recent work illustrates that punishing can either be construed as altruistically enforcing a social norm or spitefully responding to a perpetrator45. Thus, in these cases, punishment may be sending an ambiguous signal about an individual’s moral character. In contrast, endorsement of other choices in the Justice Game likely sends more informative signals about a person’s altruistic tendencies and what they care to uphold. For example, choosing compensation or equity could signal that the person clearly, and unambiguously, cares about prosocial values, including altruism or generosity. These results provide compelling evidence that third-party punishment is only perceived as a positive moral signal in limited social contexts.

Decades of research have examined people’s preference for punishment and its suggested role in enforcing normative behavior. And yet, our data suggests there is a strongly held preference for non-punitive justice restoration. These findings mirror evidence from outside the lab. A 2016 National Survey of Victims’ Views found that victims overwhelmingly prefer that the United States criminal justice system focus more on rehabilitation than punishment51. Together, these findings have implications for criminal justice systems; if we want justice restoration to reflect the values and preferences of its citizens, then non-punitive alternatives for alleviating wrongdoing should be promoted more than they are now.