Humans exhibit a remarkable capacity for large-scale cooperation, a phenomenon that has long puzzled evolutionary social scientists. While kin selection models explain cooperation based on inclusive fitness and kin discrimination, they fail to account for cooperation among non-kin. Reciprocity, reputation, and status differentiation play significant roles in shaping cooperative networks beyond kinship ties. Early models of direct reciprocity, while documenting reciprocal exchange in humans and animals, prove insufficient to explain the breadth of human cooperation, particularly in larger groups or transient interactions. The limitation of direct reciprocity becomes apparent as group size increases, and the need for alternative mechanisms arises. Third-party punishment, although effective, faces a "second-order free-rider problem." Indirect reciprocity emerges as a potential solution, where individuals cooperate or punish to maintain their standing or reputation, influenced by observable past behaviors. This study focuses on both positive (cooperation based on good reputation) and negative (exploitation based on bad reputation) indirect reciprocity. Existing models suggest that targeted exploitation of individuals with selfish reputations can incentivize cooperation. However, few empirical studies have examined the interplay of dyadic perceptions and community-wide reputations in shaping cooperation, exploitation, and punishment networks in non-WEIRD contexts. This research uses network-structured economic games and dyadic peer ratings to investigate the influence of positive and negative reputations on these behaviors in four rural Colombian communities.

The literature extensively explores the evolution of cooperation, beginning with kin selection models emphasizing inclusive fitness [Hamilton, 1964; Hamilton, 1970]. However, the widespread cooperation observed beyond kin relationships necessitates alternative explanatory frameworks. Direct reciprocity, initially proposed by Trivers (1971) and further developed by Axelrod and Hamilton (1981), posits that cooperation evolves through repeated interactions and mutual benefit. Numerous studies have demonstrated direct reciprocity in various human and animal contexts [e.g., Apicella et al., 2012; Carter & Wilkinson, 2013], yet debates persist regarding its defining characteristics [Carter, 2014]. Limitations in direct reciprocity models, particularly concerning scalability to larger groups and the absence of direct reciprocation in some instances [Boyd, 1989; Boyd & Richerson, 1988], led to the development of models incorporating indirect reciprocity [Nowak & Sigmund, 2005; Panchanathan & Boyd, 2004; Boyd & Richerson, 1989]. These models emphasize the role of reputation and social standing in shaping cooperative behavior, focusing on both positive indirect reciprocity, where cooperation is directed towards individuals with good reputations [Johnstone & Bshary, 2004; Roberts et al., 2021], and negative indirect reciprocity, where exploitation is targeted at those with bad reputations [Bhui et al., 2019]. The present study builds upon these theoretical advancements, investigating the empirical evidence for both forms of reciprocity in naturally occurring social networks.

This study employed a suite of three network-structured economic games, termed "RICH" games, alongside dyadic peer ratings to investigate cooperation, exploitation, and punishment in four rural Colombian communities. The RICH games measured participants' preferences for allocating resources, exploiting others, and imposing costly punishments. The games were recipient identity-conditioned, allowing participants to know the identity of other players, but decider-confidential; individuals' choices remained unknown to those they affected. The games included: (1) a RICH allocation game, where participants could allocate a fixed amount of coins among community members, including themselves; (2) a RICH exploitation game, where participants could anonymously take coins pre-allocated to others; and (3) a RICH costly reduction game, where participants could use coins to purchase tokens to reduce others' payouts. The stakes varied across communities, reflecting local economic conditions (roughly $5-$25 USD equivalent). Dyadic peer ratings assessed individuals' perceptions of others' generosity and selfishness, providing measures of social standing independent of game-play behavior. Data collection involved presenting participants with randomized photo arrays of community members. The data (generosity, selfishness ratings, and economic game outcomes) comprise multi-layer networks, reflecting the interactions between individuals across different contexts. A multiplex Bayesian generalization of the social relations model (SRM) [Kenny & La Voie, 1984; McElreath, 2020] was used to analyze the multiplex network data. This model allowed for investigation of dyadic and generalized reciprocity within and between network layers. The model accounts for sender effects, receiver effects, dyadic effects, and the correlations between these parameters across different layers. The analysis focused on testing whether individual behavior (giving, exploiting, punishing) was influenced by personal perceptions of social standing (dyadic reciprocity) and community-wide reputations (generalized reciprocity).

The analysis revealed strong support for the predictions of direct and indirect reciprocity. At the dyadic level, individuals exhibited reciprocal giving, taking, and punishment, indicating direct reciprocity. Furthermore, individuals' behaviors were significantly influenced by their personal perceptions of others' social standing, exhibiting positive direct reciprocity by giving more to those perceived as generous and negative direct reciprocity by exploiting and punishing those perceived as selfish. Specifically: Individuals reciprocally gave to one another (high positive correlations across all communities); they reciprocally took from one another (positive correlations across all communities, stronger in lowland and coastal communities); and they reciprocally punished one another (positive correlations across three out of four communities, with highland community showing less reliable reciprocity due to low frequency of punishment). The correlations between behaviors and perceptions further demonstrated that individuals were more likely to give to those they rated as generous and less likely to give to those they rated as selfish. Similarly, participants were less likely to exploit generous individuals and more likely to exploit selfish individuals. Punishment behaviors also followed this pattern. At the individual level, generalized reciprocity was evident, whereby community-wide reputations influenced behavior. Individuals with generous reputations received more allocations and were less likely to be exploited, although this association with reduced punishment was less consistent. In contrast, individuals with selfish reputations were more likely to be punished, although the prediction of increased exploitation was not consistently supported.

The findings strongly support the role of both direct and indirect reciprocity in shaping cooperative networks. The tight link between dyadic perceptions and behavior underscores the importance of individual-level assessments of social standing in guiding resource allocation, exploitation, and punishment decisions. The extension to community-level reputations shows that these localized perceptions scale to broader social dynamics, influencing overall cooperation patterns. The unexpected finding that generous individuals were sometimes more likely to be punished merits further investigation, possibly reflecting economic need or divergent perceptions within communities. These results highlight the complexity of reputation and its impact on behavior, highlighting the importance of analyzing dyadic relationships alongside aggregated reputation measures. The study contributes to the broader understanding of how humans achieve and maintain cooperation in diverse social contexts. The use of ecologically valid economic games in a non-WEIRD context improves upon prior laboratory-based research, enhancing the generalizability of findings.

This study provides robust empirical evidence supporting the theoretical models of direct and indirect reciprocity as mechanisms underpinning human cooperation. The findings demonstrate how individual perceptions of social standing and community-wide reputations influence cooperative, exploitative, and punitive behaviors. Future research could investigate the causal links between these factors more definitively through experimental interventions, explore longitudinal dynamics, and examine the influence of additional reputation types on cooperation networks. The methodological advancements of employing ecologically valid network-structured economic games in diverse settings enhance the applicability of these findings beyond laboratory settings.

The study's correlational design limits causal inferences. While the strong associations between perceptions and behaviors are suggestive of causal links, experimental manipulations would be necessary to definitively establish causality. The focus on generosity and selfishness as reputation measures omits other potentially influential factors, such as status and different interpersonal sentiments, that warrant future research. Finally, the cross-cultural comparison is limited to four communities in a specific region of Colombia, and generalizability to other contexts may vary. Further research is needed to determine how these findings hold up in various cultural settings.