Motivations to reciprocate cooperation and punish defection are calibrated by estimates of how easily others can switch partners

Motivations to reciprocate cooperation and punish defection are calibrated by estimates of how easily others can switch partners. Sakura Arai, John Tooby, Leda Cosmides. PLoS One, April 19, 2022. https://doi.org/10.1371/journal.pone.0267153

Abstract: Evolutionary models of dyadic cooperation demonstrate that selection favors different strategies for reciprocity depending on opportunities to choose alternative partners. We propose that selection has favored mechanisms that estimate the extent to which others can switch partners and calibrate motivations to reciprocate and punish accordingly. These estimates should reflect default assumptions about relational mobility: the probability that individuals in one’s social world will have the opportunity to form relationships with new partners. This prior probability can be updated by cues present in the immediate situation one is facing. The resulting estimate of a partner’s outside options should serve as input to motivational systems regulating reciprocity: Higher estimates should down-regulate the use of sanctions to prevent defection by a current partner, and up-regulate efforts to attract better cooperative partners by curating one’s own reputation and monitoring that of others. We tested this hypothesis using a Trust Game with Punishment (TGP), which provides continuous measures of reciprocity, defection, and punishment in response to defection. We measured each participant’s perception of relational mobility in their real-world social ecology and experimentally varied a cue to partner switching. Moreover, the study was conducted in the US (n = 519) and Japan (n = 520): societies that are high versus low in relational mobility. Across conditions and societies, higher perceptions of relational mobility were associated with increased reciprocity and decreased punishment: i.e., those who thought that others have many opportunities to find new partners reciprocated more and punished less. The situational cue to partner switching was detected, but relational mobility in one’s real social world regulated motivations to reciprocate and punish, even in the experimental setting. The current research provides evidence that motivational systems are designed to estimate varying degrees of partner choice in one’s social ecology and regulate reciprocal behaviors accordingly.

4 Discussion

4.1 Evidence that motivational systems are designed for social ecologies with varying levels of partner choice

Ancestral variation in the availability of cooperative partners would have favored the evolution of motivational systems that treat partner choice as a continuous variable. Motivations to keep valuable cooperative partners and abandon unrewarding ones should be up-regulated in response to the perception that others can easily switch partners.

Here we tested the hypothesis that an individual’s motivations to reciprocate and punish are calibrated by that person’s estimate of the degree to which others in their local social ecology can exercise partner choice. This estimate is captured by measures of relational mobility [50]. The higher an individual’s relational mobility score, the more opportunities they believe others have to leave unsatisfying relationships for better ones.

We assessed motivations to trust, reciprocate, defect, punish, and switch partners by allowing people to cooperate for mutual benefit with a new individual. The results showed that motivations to reciprocate and punish tracked participants’ perceptions of relational mobility. The more partner choice they thought others in their social ecology could exercise, the more they reciprocated their partner’s trust and the less they paid to punish their partner—even when that partner had defected.

Providing incentives for desirable partners to stay in the relationship is the proposed function of these motivational calibrations. If that is correct, then people who have the opportunity to switch partners will be more likely to stay with a partner who reciprocates their trust and more likely to leave one who punishes them. After two rounds, half the participants were asked if they wanted to keep their current partner or switch to someone new. Holding all else equal, having been defected on more than quadrupled the odds that they wanted to switch and having been punished tripled the odds they would choose to leave. These were the two biggest independent predictors of switching decisions. The desire to leave a partner who punished was especially strong for participants who returned 40%—a response that creates a positive payoff for both parties that is almost equal. These individuals were almost 10 times more likely to want a new partner.

4.1.1 Are priors about social ecology updated by information about the situation or the person?

Perceptions of relational mobility are based on a huge database of experiences in a local social ecology—sometimes a lifetime’s worth. For this reason, we proposed that relational mobility serves as an estimate of the prior probability that others in one’s social ecology can exercise partner choice. It is a best guess before you learn what your partner is like—the situation participants faced in round 1.

If relational mobility in your social ecology is used to estimate a partner’s outside options when you know nothing else about that person, then its effect on cooperative motivations should be reduced (or eliminated) by data about that specific person’s value as a cooperative partner—to yourself and others. The evidence indicates that participants in both societies updated this prior based on first-hand knowledge of their partner’s willingness to cooperate and reluctance to punish. Once participants had experienced how their partner behaved in round 1, relational mobility no longer predicted how much they trusted, reciprocated, or punished in round 2, in either the US or Japan. The behavior of the sham partner in round 1 (and, of course, in round 2) did predict their responses. The only behavior that relational mobility continued to influence was antisocial punishment. The belief others in your social ecology can easily switch partners tempered—but did not eliminate—antisocial punishment. (See S5 Appendix.)

The results suggest that estimates of partner choice based on social ecology are updated based on properties of the person with whom one is interacting. But are these estimates updated in response to cues about a temporary situation one is facing—ones unrelated to the partner’s value as a cooperator? It is not clear that they should be.

Delton et al. [35] examined the evolution of motivations to cooperate in Bayesian agents who knew the base rate of one-shot interactions in their population and updated this prior based on a cue about the immediate situation they were facing. The cue reflected the probability that they would never interact again with their current partner. These Bayesian agents evolved a strong disposition to cooperate even when they rationally believed the interaction was one-shot. Selection favored agents who behaved as if they would repeatedly interact with their current partner even when they knew this was unlikely. Agent-based models also show that meeting a new individual once was a good cue that you will meet them again in ancestral social ecologies [64]. Every participant in our study was exposed to this ancestrally-reliable cue to a shadow of the future: They interacted with their partner for two rounds.

We did, however, provide a verbal cue relevant to partner choice in the temporary situation that they were facing. Half the participants were told they would be interacting with the same partner in every round (i.e., they were engaged in a repeated interaction with this person). The other half were told they could change partners after two rounds (i.e., their current partner can refuse to interact with them repeatedly). If this verbal cue is used to (temporarily) update their prior probability that a newly encountered person can exercise partner choice, their motivations to cooperate or punish might shift in response.

There was little evidence that participants in round 1 used this situational cue to update a prior that was based on their social ecology. Being told whether they would have the opportunity to switch partners had no effect on how much participants punished defections by their partner: Higher relational mobility in their local social ecology predicted less punishment, regardless of condition or society. The cue did have an effect on how much American participants reciprocated their partner’s trust, however. Although average levels of reciprocation were similar in both conditions, higher relational mobility predicted more reciprocation when Americans were told they and their partner could part ways after two rounds, but not when they were told that all of their interactions would be with the same partner.

Japanese participants did not respond to this cue at all: Their estimates of relational mobility predicted more reciprocation (and less punishment) to the same extent in both conditions. That is, there was no evidence that people in Japan updated their prior hypothesis about relational mobility based on the situational cue we provided. If they did, the change was too small to influence their willingness to reciprocate or punish.

If this result generalizes to other cues about a temporary situation, it suggests that the benefits of opportunistic behavior in the short term were generally outweighed by the risk of losing a valuable, long-term cooperative partner.

4.2 What is the function of punishment in dyadic reciprocal cooperation?

What, if anything, is the adaptive function of motivations to pay a cost to punish a defecting partner? This was not a rare response: Of participants who were trusters in round 1, 44% punished when the responder defected. It is usually assumed that the function of punishing defectors is to elicit more cooperation from them in the future—especially when they do not have the option to change partners.

People who believe others in their social ecology have fewer options to switch partners did pay more to punish defectors: Low relational mobility scores predicted paying more to punish. But there was no evidence that punishment succeeded in eliciting greater cooperation from participants. Quite the contrary: Participants who were punished for returning 0–40% in round 1 did not respond by sending more points as truster in round 2. Indeed, they returned fewer points as truster (β = -.22, p = .0002), and this effect was particularly pronounced for those who had provided a positive payoff by returning 40% in round 1, β = -.41, p = .0001 (vs. β = -.12, p = .099 for those who provided a negative payoff in round 1; see S5 Appendix). Moreover, those who were punished in round 1 were more likely to retaliate by punishing their partner in round 2 (S5 Appendix; for similar results, see [65, 66]).

Not only did punishment fail to elicit more cooperation from punished partners, but it also drove them away. When partner switching was possible, having been punished was one of the biggest independent predictors of wanting to change partners. Driving away defectors might be a function of punishment, of course—when they were not punished, ~70% of people who returned 0–40% wanted to remain with their accommodating partner (~68% of those who returned 0–30%; ~71% of those returning 40%). Although participants in this study could prevent future interactions at lower cost by simply deciding to switch after round 2, avoiding unrewarding partners may be more difficult in real life, especially when they want to continue cooperating with you.

Krasnow et al. [57] suggest that punishing defection signals a willingness to continue cooperating with your current partner, but on more favorable terms. Using a paradigm similar to the TGP, they found that participants who punished a defecting partner in the first round were 11 times more likely to cooperate than defect in the second one (switching was not an option). This pattern was not apparent in our study: Participants who punished a defecting partner did not return more in round 2 than those who did not (39.28% vs. 35.18%, t (226.99) = 1.41, p = .160), and they were not more likely to want to remain with their partner—indeed, the more points participants paid to punish the partner, the more—although slightly—they wanted to switch (OR = 1.02; 95% CI = [1.01, 1.03]). (Note, however, that a participant’s decision to stay did not ensure a continuing interaction in our study; the partner also had the option to leave, and punished ones were likely to do so.)

Our results showing that retaliatory punishment was common—~45% of those who were punished in round 1 retaliated in round 2—suggest an alternative explanation. In Krasnow et al. [57], participants who punished defectors in round 1 may have cooperated in round 2 to avoid (very costly) retaliatory punishment by their partner. Those who did not punish partners who succumbed to the temptation to cheat in round 1 may have assumed their partner would “reciprocate” by not punishing when them when they did the same in round 2.

Motivations to punish did not reflect the participant’s own commitment to stay in the relationship, but they were up-regulated by estimates that partners might have few outside options: Lower relational mobility in one’s social ecology did predict amount paid to punish defectors. The results are consistent with the hypothesis that motivations to punish evolved to deter bad treatment in the future by partners who do not seem to value your welfare [67]. Defecting now may be a reliable cue that this partner does not value your welfare sufficiently, and punishment was overwhelmingly directed at defectors. In ancestral social ecologies, partners who part ways now may nevertheless have to cooperate again in the future [64, 67, 68]. Punishment may have evolved as a warning, to deter bad treatment by defectors who may darken your door in the future.

4.3 Micro and macro effects of social ecology

We measured two variables regarding participants’ real-life social ecology of partner choice. First, we measured participants’ perceptions of their partner choice ecology with the relational mobility scale [50]. Second, we recruited participants from two societies in which average relational mobility scores are typically high (US) versus low (Japan). This lets us see whether behavior at the individual level scales up to explain differences between nations.

Within each society, the motivations of individuals were calibrated by their perceptions of other people’s relational mobility: the number of opportunities they believe that others have to form new relationships. Moreover, the pattern of calibration was universal: Within each society, higher relational mobility scores predicted more reciprocation and less punishment. Individual-level effects tracked individual perceptions of the local social ecology.

What about group-level differences? The concept of relational mobility was built from Yamagishi’s seminal work on general trust: a cognitive bias to assume that newly encountered people will treat you with benevolence rather than exploitation [69, 70]. General trust varies across nations; scores on the standard survey measure are higher in the US than Japan, for example. Where general trust is higher, people are more willing to risk cooperating with strangers who could, if untrustworthy, profit at their expense. The benefit of trusting strangers is that it allows people to discover better cooperative partners, giving them more outside options. The resulting increase relational mobility then tempers the risk of trusting strangers: The threat that a good partner will leave for a better outside option can deter exploitive behavior and increase benevolence.

With this in mind, we compared average behavior in the US and Japan. As in other studies, perceptions of relational mobility were higher in the US than Japan (RM others: 4.12 vs. 3.57, t (1028.2) = 13.76, p = 10−16; RM self: 4.20 vs. 3.37, t (1030.8) = 18.71, p = 10−16). That is, the average American believes others have more outside options than the average person from Japan does. Moreover, as Yamagishi’s view of general trust predicts, when participants had no prior experiences with their partners, American trusters risked more points on a stranger than Japanese participants did (Trust: 59 vs. 50.6, t (502.55) = 2.9, p = .004). And trusting strangers usually paid off: Most responders delivered a positive payoff in both societies (US 67%, JP 76%).

Did the perception that others have more outside options lead the average American to reciprocate more and punish less than the average person from Japan? No. Not only did Americans return less, on average, than Japanese participants, but more of them exploited their partner’s trust by delivering a negative payoff (US 33% vs. JP 24%). Americans were also more punitive, not less: They paid more to punish their partners, even when controlling for all other factors (including whether their partner defected). And, despite less reciprocation and more punishment at the macro-level, Americans were more likely to stay with their partner than Japanese participants (all else equal).

Within each society, individual differences in reciprocation and punishment were associated with individual differences in perceptions of relational mobility, but this did not translate into group-level differences between the US and Japan. Assuming that individual differences fully explain group-level differences is called the ecological fallacy [71–73]. The data clearly show that the micro-level effect of individuals’ perceptions of relational mobility and the macro-level effect of society were independent of one another. The individual-level psychological calibrations and the group-level differences between nations coexist, rather than one producing the other.

Features of the social ecology other than relational mobility could be responsible for the differences in group-level calibrations between the US and Japan (see e.g., [56, 74]). That Japanese participants were less punitive than Americans is contrary to findings that Japan (or East Asian countries in general) has “tighter” norms than the US which, when broken, elicit great censure [75, 76], but perhaps consistent with studies showing greater motivations to avoid rejection in people in Japan than the US [74]. Our data cannot speak to these explanations of the group-level differences we found.

4.4 Limitations and future directions

Motivations responded when participants learned how the partner treats them, but the partner switching instructions influenced Americans only (and not much at that). This could be because repeated interactions—with interruptions between—were common ancestrally, making long-run estimates of social ecology a more reliable basis for calibration than cues about a fleeting situation. The other possibility is that a cue delivered online was too divorced from real life, devoid of psychophysical cues typical of social isolation versus community. Future studies might enhance the salience of the situational cue, perhaps by including visual displays showing many versus few alternative partners (avatars or faces), or by giving participants prior experiences of a desirable partner leaving for a better one or an unrewarding partner staying.

A person with fewer outside options than others in their local ecology may feel they need to reciprocate more and punish less. We did adapt the relational mobility scale to ask about the self; although self and other scores were correlated r (515) = .60 (p = 10−16) in the US and r (516) = .50 (p = 10−16) in Japan, we calculated whether RM self < RM other for each participant. In Japan, 67% of participants felt their outside options were worse than those of other people, compared to 44% in the US. And, in both countries, those who felt they have fewer outside options returned more points than those who felt their options were better than or equal to others, but the difference in points returned was not significant. A better measure in the future might be to ask, for each RM question, whether people feel they have more, the same, or fewer options than others in their society.

Dyadic cooperation may be affected by other aspects of the social ecology as well, such as how likely others will be to take advantage of you [69]. Punishment as a deterrent may be up-regulated in ecologies where the probability of being exploited are higher, as they were in the US in this study. Perceptions of these probabilities would be a fruitful variable to assess.

Lastly, our participants were from either the US or Japan, two populous, large-scale industrialized societies. Objectively speaking, most people in these countries are free to associate with anyone they like, and they are surrounded by strangers, each of whom is a potential new partner. It would be fruitful to extend the current line of research to smaller societies in which the actual—not only perceived—possibility of partner choice is more limited.