Reputation systems are crucial for fostering cooperation in human societies, enabling individuals to identify reliable partners and avoid those with a history of selfish behavior. These systems incentivize prosocial actions, as individuals with positive reputations are more likely to be chosen for beneficial collaborations and less likely to be excluded from social networks. In dynamic networks, where individuals can adjust their ties, reputations lead to cooperative clusters, enhancing the overall welfare of prosocial actors. However, the impact of reputation systems in societies marked by strong, identity-based divisions, such as the politically polarized contemporary US, remains less understood. In highly polarized environments, while maintaining a positive reputation within one's own political party is beneficial, the value of being known as someone who cooperates with members of the opposing party is less certain. Different reputational information – whether reputations track behavior towards outgroup members (e.g., opposing political party) or only ingroup members – can lead to different patterns of cooperation and network formation. This research investigates how these varied reputation systems affect cooperation and network segregation based on political affiliation, contributing to a broader understanding of reputation effects on cooperation and segregation across social identities. The study focuses on a context highly conducive to cooperation, where participants can sever unwanted ties and form new ones, in order to analyze the effects of different reputational information under such circumstances. Early differences in cooperation are anticipated to influence tie formation and dissolution, resulting in varying levels of network segregation. This study directly contributes to understanding the social and informational underpinnings of polarization, a crucial issue with significant social and democratic implications. The experiment employs four conditions: a control condition with hidden political identities, and three treatment conditions (undifferentiated, parochial, and intra/intergroup reputations) where political identities are visible and reputation systems differ in how they track behavior towards ingroup and outgroup members.

The study draws upon existing research on reputation systems and their role in promoting cooperation. Existing literature establishes the crucial role of reputation in fostering cooperation and influencing network structure. The authors review studies showing the positive correlation between reputation systems and cooperation levels, the formation of cooperative clusters, and the avoidance of selfish actors. The paper also references research on social identity theory, highlighting its implications for intergroup cooperation and network segregation. Social identity theory posits that individuals strive to maximize their ingroup's relative advantage over outgroups. This perspective suggests that a reputation system that differentiates between ingroup and outgroup treatment might reduce intergroup cooperation and increase network segregation. Conversely, the literature on unbounded indirect reciprocity proposes that distinct reputations for ingroup and outgroup interactions could encourage cooperation with both, thereby promoting integration. The study's theoretical framework integrates these competing perspectives to predict the outcomes of different reputation systems in a politically polarized environment. This integrated approach helps to frame the specific research questions and hypotheses that guide the experimental design and analysis.

The study employed a large-scale web-based experiment (N=1073 participants) using a custom application built on Breadboard. Participants were recruited through Prolific and randomly assigned to one of four conditions, manipulating the visibility of political identities and the type of reputation system. The initial networks were random Erdos-Renyi graphs with a density of 0.167. Each tie in the network represented an opportunity to play a dyadic iterated prisoner's dilemma, where participants decided how many of 50 monetary units to send to their partners (0=defection, 50=maximal cooperation). Sent amounts were doubled for the recipient. Participants made independent decisions for each alter. Reputation scores were based on objective information: the average number of monetary units donated to alters in the previous three rounds, without distinguishing between cooperation with cooperative versus uncooperative alters. The four conditions were: (1) Control: political identities hidden, reputation based on average monetary units given; (2) Undifferentiated reputations: political identities visible, reputation based on average monetary units given regardless of ingroup/outgroup status; (3) Parochial reputations: political identities visible, reputation based on average monetary units given to ingroup members only; (4) Intra/Intergroup reputations: political identities visible, separate reputations for average monetary units given to ingroup and outgroup members. Every four rounds, participants could sever one tie and propose a new one; proposed ties required approval by the other participant. The study lasted 18 rounds, with participants earning money based on their accrued monetary units. After completing the interaction phase, participants completed a post-study questionnaire. Statistical analyses included linear mixed models, conditional logistic regressions, and OLS regressions to analyze cooperation rates, tie severing and formation, and network clustering and segregation. Non-parametric inference methods were used to address network dependence and temporal autocorrelation.

The results revealed that when participants knew each other's political affiliations, early cooperation patterns exhibited ingroup favoritism, leading to network segregation based on political identities. However, this effect was significantly reduced in the intra/intergroup reputation condition, where participants had distinct reputations for how they treated ingroup and outgroup members. Overall cooperation rates were consistently highest in the intra/intergroup reputation condition throughout the study. Analyzing cooperation in the first eight rounds, before network dynamics reached equilibrium, showed the strongest ingroup favoritism effect in the parochial reputation condition and the weakest in the intra/intergroup condition. This difference suggests that the type of reputational information available influenced cooperation patterns, not just learning over time. In terms of network dynamics, participants in the intra/intergroup condition were less likely to sever ties, and when they did, the decisions were primarily based on cooperation levels, not political identity. While participants in all three conditions where political identities were visible tended to preferentially propose ties to ingroup members, this tendency was weakest in the intra/intergroup condition. Network clustering increased over time in all conditions, but network segregation was significantly higher in the undifferentiated and parochial conditions compared to the intra/intergroup and control conditions. These findings highlight that having separate reputations for ingroup and outgroup treatment mitigated the tendency for networks to segregate based on political identities. Participants in the intra/intergroup condition tended to sort primarily on cooperation levels rather than political affiliation.

The findings address the research question of how different reputation systems influence cooperation and segregation in politically polarized environments. The results demonstrate that the design of reputation systems significantly impacts cooperation and network segregation. Specifically, a reputation system that allows for distinct reputations for ingroup and outgroup treatment is crucial in mitigating identity-based segregation. This contradicts predictions from social identity theory, which suggests that such a system might intensify outgroup discrimination. Instead, the results align with the unbounded indirect reciprocity perspective, showing that providing information on outgroup interactions can promote cooperation and integration. The study's strength lies in its experimental design, rigorously manipulating reputation systems to isolate their effects on cooperation and network structure. However, the findings are specific to contexts where cooperation is incentivized by monetary gains, and the generalizability to other scenarios where political or ideological stakes are higher remains to be investigated. The study contributes to understanding the interplay between reputation, identity, and network dynamics in shaping social interactions, which holds significant implications for bridging social divides and promoting cooperation in diverse societies.

This study demonstrates the crucial role of reputation system design in shaping cooperation and network segregation in politically polarized contexts. The finding that distinct reputations for ingroup and outgroup treatment promote cross-party cooperation contrasts with predictions from social identity theory and highlights the importance of considering the information content of reputation systems. Future research should investigate the generalizability of these findings to other social identities and contexts with higher stakes and examine the interplay between reputation systems and other interventions aimed at reducing political polarization. Exploring the effects of subjective versus objective reputation evaluations is another important direction for future work.

The study's findings are primarily based on an online experiment using monetary incentives, which may not perfectly reflect real-world political interactions. The use of a specific iterated prisoner's dilemma game and the experimental control over network formation might limit the generalizability of the results to naturally occurring networks. Also, the measurement of political identity using a simple six-point scale might not fully capture the complexity of political affiliations and beliefs. Finally, the study does not directly measure changes in attitudes or beliefs towards the outgroup, focusing instead on network structure and cooperation levels.