Social image concerns promote cooperation more than altruistic punishment

Introduction

According to evolutionary and economic theories, humans, similar to other animals, are expected to behave selfishly, maximizing material gains for themselves1,2,3. Nevertheless, human cooperation occurs in all known societies and is common between genetically unrelated individuals and where repeated interactions may be uncommon. This is enigmatic and demands explanation3. Several mechanisms have been proposed to explain the evolution of cooperation4. Among them, indirect reciprocity has attracted the attention of many researchers5,6,7,8. Indirect reciprocity can take many forms, although one that seems to be particularly effective for cooperation builds on an individual’s desire to maintain a reputation as a cooperator in the social group. Reputation-building is possible, because information about an individual’s cooperative behaviour in social interactions diffuse through the social network. Others can then use such information to either extend cooperation back to a cooperator or refrain from doing so vis-à-vis a defector. Evolutionary models assume that individuals accrue image scores based on the frequency with which they cooperated in the past. An individual has an incentive to maintain a reputation as a cooperator, because in future interactions she will in turn benefit from others’ cooperation in the group. An extensive body of empirical and experimental evidence confirms that maintaining a reputation as a cooperator indeed commands material rewards from other group members9,10,11. Compared with direct reciprocity12, indirect reciprocity expands the scope of cooperation, because an altruistic action will be reciprocated not only by the direct beneficiary of the action but also by whomever has knowledge of the cooperator’s positive social image.

An alternative account singles out costly punishment as a factor that can also increase cooperation. This relies on individuals’ propensity to punish, even at a cost to themselves, people who defect from cooperation or who deviate from norms of fairness13,14,15. Such punishment has been frequently observed in ephemeral ‘one-shot’ interactions, or is performed by ‘third parties’16,17,18,19. As in both cases the punisher cannot receive any material gain from punishment, this has been labelled ‘altruistic’20. Punishment appears widespread in human groups ranging from small-scale traditional societies to large-scale complex societies17. It has been argued that altruistic punishment is not a mechanism for the evolution of cooperation per se, but rather is a proximate factor that can enhance cooperation when latched onto by other evolutionary mechanisms including direct or indirect reciprocity, or group selection4,21.

There is however a profound disavantage to using altruistic punishment to maintain cooperation; because punishment entails a cost both for the punisher and the punished, it is expensive and inefficient, resulting in considerable monetary losses22,23,24 or, at worst, destructive acts of vengeance25,26. Moreover, the act of punishment entails a second-order public goods problem whose solution may require strategic coordination, asymmetries or the insurgence of a centralized authority27,28. By contrast, building a positive social image is potentially a cheaper, automatic and more efficient means of enforcing cooperation22,29,30. Arguably, maintaining a positive social image is motivationally less demanding than altruistic punishment and furnishes a different basis for the moral norms that become established in a community. Rather than relying on altruistic motivations, maintaining a good social image is in the individual’s long-term self-interest13. The main drawback of reputational mechanisms is that they are only effective in communities where there is high ‘broadcast efficiency’, that is, reliable information about individuals’ past behaviours can diffuse rapidly through the community31. We expect this to be most common in small, tight-knit communities where gossip travels quickly or where there exists an authority or institution that occupies a central position amid its social networks and serves as a conduit for disseminating information about people’s social images. Nonetheless, observability of one’s actions can have extensive consequences even in contemporary societies32 and it has been argued that online reputation systems and social media have extended broadcast efficiency even to large populations33. In addition, individuals are likely to be much more concerned about their social image in the presence of other members of their group rather than outsiders34,35,36. Indeed, preferential concern for social image among ingroup compared with outgroup members and the need to rapidly assess group membership may have favoured the elaboration of ethnic markers in, for instance, dress, dialect and bodily adornment over the course of human evolutionary history37.

Social image and altruistic punishment are not mutually exclusive motivators of cooperation, but may be activated, separately or jointly, in different situations38. It is likely to be that cultural factors and specific social structures may favour one or the other factor. It is nonetheless interesting to investigate their relative efficacy in promoting social welfare, to shed light on the respective role that they can play to sustain human cooperation. The evidence on this topic is scant and limited to few experiments where one’s social image is artificially created and maintained in the laboratory22,39,40,41. Moreover, these studies were conducted in Western, industrialized contexts and among strangers where reputation effects are ephemeral. A compelling argument has been made42 that these conditions are relatively novel within evolutionary history, and that data from non-Western small-scale societies are essential for testing hypotheses about human psychology, especially in the domains of preferences and decision-making.

Here we provide data from the first-ever study of the relative roles of indirect reciprocity and alruistic punishment, alone and in combination, in promoting cooperation among the Teop, a small-scale society located in Bougainville, Papua New Guinea. In doing so, we are among the first to study the effects on cooperation of real-life social image rather than artificial, experimentally created, social image33,43,44,45. We do so by using persons of authority as observers. Social relationships in Teop revolve around the figure of the ‘Big Man’ (BM). Big men possess exclusive knowledge and ‘impose discipline, uphold the traditional way of life and give executive directions’46 to other community members. Social disputes or problems of coordination between clans are normally dealt with under the supervision or explicit intervention of Big Men. Big Men have informal authority and also act as ‘guardians of morality’ within the society. Arguably, they are figures towards whom individuals strive to keep a positive social image. Moreover, they are ‘hubs’ of the social network; hence, they are central in disseminating social image information through the society. Running our study in Teop gives us the opportunity to examine the impact of one’s social image in a context presumably closer to that characterizing human societies for the majority of our evolutionary history35.

Our experiments contrast concerns for social image vis-à-vis the BM and punishment as factors promoting cooperation, either in isolation or combined. All these factors raise cooperation above baseline levels. Nevertheless, only when social image alone is at stake do average economic gains rise significantly above baseline. Punishment, either alone or combined with social image building, yields gains even lower than the baseline. We also show that when a BM from an external group acts as the observer, again payoffs do not rise above baseline. We conclude that individuals’ desire to establish a positive social image within their community emerges as a more decisive factor than punishment in promoting human cooperation.

Results

The experimental game

We conducted a series of anonymous, one-shot, prisoner’s dilemma (PD) games involving two participants. In the baseline condition, each participant received 10 Kina (K10) to be used in the game (Endowment I), plus K4 (Endowment II) that were not used in the game but were cashed in at the end of the game. Participants had to decide whether they wanted to keep K10, or give K10 to the other participant. If a participant kept the K10, this person would receive the K10 at the end of play. If a participant gave K10, the other person would receive K20 at the end of play because, as it was explained, the researcher would add K10 to the exchange. The payoff structure of the game, in its simplicity, resembles a ‘tragedy of the commons’ scenario4,5,9. Mutual cooperation—namely, both players giving to the counterpart—ensures the highest payoff for the group, but mutual defection—namely, both players keeping their Endowment I—is the rational strategy for individual payoff maximization.

To test for effects of social image concerns and altruistic punishment in motivating cooperation, we implemented four additional experimental treatments (summarized in Table 1). In treatment ‘BM’ participants played the PD as in the baseline, with the key difference that a BM from the same village as each participant was present in the room and observed his or her decision. In ‘Big Man External’ (BM EXT), each decision was witnessed by a BM from a different ethno-linguistic group. These treatments capture differences in experimental outcomes when Big Men having clear and different social distances from the participants are present, as well as varying informal authority. The extant literature demonstrates that being observed by one’s peers leads individuals to modify their behaviours in significant ways43,44,45,47, although this may lead to less pro-sociality when ingroup–outgroup relationships are made salient36.

Table 1: Experimental design.

Full size table

We modelled altruistic punishment as is standard in the literature18,20,22,23,25 by introducing a punishment stage after all participants played the PD game described above. Either player had the option of spending the K4 from Endowment II to reduce the other participant’s payoff. Each participant could spend K0, K2 or K4 to reduce the counterpart’s payoff by K0, K10 or K20, respectively. We used the ‘strategy method’17,35 to investigate punishment patterns. Each participant had to make two decisions under the assumption that the other participant had either kept K10 or given K10. This allows us to examine patterns of what have been named ‘altruistic’ punishment—that is, punishment when the other player defects—and ‘anti-social’ punishment—that is, punishment when the other player cooperates25,30. To study the interaction between social image concerns and altruistic punishment, we added a ‘Big Man+Punishment’ (BM+PUN) treatment in which the local BM observed both the PD and punishment choices, in the same way as in the BM treatment.

Participants made their decisions privately and anonymously in all conditions, never knowing who was their co-player. The only information they received was that the other player was from either their own or a neighbouring village. Unlike other research conducted in small-scale societies17, experimenters and local assistants left the experimental room when participants made their choices. This was done to maximize the saliency of the BM alone, rather than the experimenters, in participants’ concerns with social image building and their consequent choices. Handing out the K4 ‘Endowment II’ in all treatments guarantees the absence of income effects across treatments (net of the punishment decision in the punishment treatments). Before participants made their decisions, they had to pass a thorough comprehension check (see Methods, Supplementary Methods sections 2.3 and 3.2, and Supplementary Table 1 for demographic characteristics of the sample).

Cooperation is highest in the ‘BM’ treatment

Figure 1 plots mean cooperation rates per treatment. Cooperation is highest in the BM treatment. 63.9% of participants gave their Endowment I to their counterpart in the BM condition, whereas cooperation rates in PUN and BM+PUN are 3–4% lower than in BM. However, such cooperation rates are not statistically different from each other (Wald’s tests derived from logit regression; P=0.60, N=272 for difference between BM and PUN; P=0.67, N=272 for difference between BM and BM+PUN; P=0.92, N=272 for difference between PUN and BM+PUN; all tests being reported are two-tailed; seeSupplementary Tables 2 and 3 and ‘Statistical methods’) in a logit model that controls for village effects, experimenter identity effects, gender and comprehension. The same conclusion holds if adding additional demographic controls for age, education and an index of household wealth (see Supplementary Tables 3 and 4), or if any control is omitted (see Supplementary Table 3, column 1). Among demographic controls, we note that age is positively associated with cooperation (P=0.018), which is in line with findings from Western societies48. Education, too, exerts a positive effect on cooperation (P=0.029 for ‘Years of Education’ longer than 10 years).

Figure 1: Mean cooperation rates by treatment.

Error bars reflect ±1 s.e.m. Top horizontal bars show results of pairwise Wald’s tests over the existence of significant treatment differences in a logit regression, as per model reported inSupplementary Table 3, column 3. *P<0.05. Only significant tests from such regressions analyses are reported. Cooperation rates are highest in the BM treatment (n=61). These are statistically significantly higher than those in baseline (n=70), while cooperation rates in BM EXT (n=27), BM+PUN (n=57) and PUN (n=57) are not.