Segregating socioeconomic classes leads to an unequal redistribution of wealth

The study investigates how socioeconomic class segregation and asymmetric punishment power shape cooperation and income distribution. Using iterated Prisoner’s Dilemma settings, the authors test whether allowing only one class (dominant/rich or subordinate/poor) to punish the other (segregated society) versus allowing mutual punishment under random matching (integrated society) alters cooperation, defection, punishment behavior, and payoffs. Conducted in China, a high-inequality context, the central hypothesis is that segregation with unidirectional punishment amplifies inequality, whereas integration with reciprocal punishment opportunities fosters wealth redistribution.

The paper situates its question within extensive work on social hierarchies, inequality, and cooperation. Prior research links the emergence of hierarchies to societal complexity and agriculture and documents power asymmetries affecting cooperative behavior and payoffs. The role of punishment in sustaining cooperation remains debated across species and human laboratory games (public goods and prisoner’s dilemma), with evidence that sanctioning can both promote cooperation and exacerbate payoff differences. The authors note China’s relatively high Gini coefficients and limited behavioral economic evidence on reducing income disparity via game-theoretic mechanisms. They build on studies manipulating punishment power and contribute by tying punishment ability to measured socioeconomic background rather than exogenously assigning it.

Laboratory experiments: Conducted at Yunnan University of Finance and Economics (Kunming, China) across 9 sessions/days between Dec 2014 and Apr 2015. Participants: 348 undergraduate students (164 males, 184 females), ages 18–22, with no prior exposure to game theory/behavioral experiments; each participated once. Informed consent obtained; ethics approval by YUFE Ethics Committee. Sessions averaged ~38 participants, ~82.5 rounds per session, with repeated interactions (up to 9 repeated partners in the same day with 75% continuation probability). Average duration ~1 hour (unknown to players); average payment 53.64 CNY. Design: Iterated Prisoner’s Dilemma with costly punishment option. Three treatments, each with two phases; subjects assigned randomly to treatment. Players were divided into two gameplay classes: CDP (allowed to choose Cooperate, Defect, or Punish) and CD (allowed only Cooperate or Defect). Text option to terminate interaction was disabled. Payoff values for defection and punishment were set relatively lower than in prior studies to avoid confounds between PD and public goods structures. Treatments: T0 (integrated society model) with random matching across socioeconomic classes, both classes could punish each other in random pairings; T1 and T2 (segregated society models) with predetermined cross-class matching and asymmetric punishment power (in one, richer could punish poorer; in the other, poorer could punish richer), such that only one class could punish in a given treatment. Socioeconomic classification: Before play, participants completed an anonymous electronic questionnaire on family wealth indicators: number of houses and cars, rooms in first house relative to family size, parents’ occupations and whether parents have academic degrees (full list in Supplement). An index was constructed with weighted components (greater weight to property ownership, then cars, room count/space per family member, managerial jobs, and parental university education). Subjects were split into “rich” vs “poor” classes by median split of the index. Validation: Individuals with higher socioeconomic index scores earned more in the experiment, consistent with the hypothesis that financial background correlates with more effective strategies (F(1,26)=4.84, p=0.029). Agent-based simulations: Post-experiment agent-based model mirrored the human game payoff matrix and treatment structures, with 350 agents split into rich and poor classes. Baseline hypothesis encoded a class-dependent initial cooperation propensity (rich p=0.7, poor p=0.3). Additional scenarios: (i) no class difference with high initial defection propensity for all (p=0.95); (ii) variation in memory size (short memory 5 vs 10 rounds). Strategies included Tit-for-Tat and memory-based updating rules. Statistical analysis: Behavioral choice frequencies (C, D, P) and final earnings analyzed using generalized linear mixed models (GLMMs) with class-by-treatment as fixed effects and subject as random effect. Analyses conducted in R 3.2.4 with lmerTest; experimental interface Z-Tree. Socioeconomic index computed in SPSS Statistics 24; figures produced in SPSS and supplements.

Behavioral frequencies: - Class membership significantly explained variation in behavioral strategies (GLMM: F(1,347)=49.15, p<0.001). - Treatment condition (integrated vs segregated) did not significantly affect overall strategy selection in the reduced model (F(2,345)=0.74, p=0.47). - Integrated condition (T0): Classes cooperated at similar levels; punishers restrained defection while punishing (F(2,518)=4.21, p=0.015). - Segregated conditions (T1, T2): Cooperation levels significantly differed across classes (F(2,510)=266, p<0.001), while defection levels did not (F(2,510)=2.026, p=0.79). Punishers decreased cooperation when choosing to punish. Earnings (human experiments): - Across conditions, total incomes differed between CDP (punishers) and CD (non-punishers), indicating the class manipulation produced sizable payoff differences. - Integrated condition (T0): The ability to punish yielded a modest earnings increase (~11 points), marginally significant (t(1,008)=199, p=0.048); earnings differences between classes were weak, implying reduced inequality. - Segregated conditions (T1, T2): Punishers earned almost double the earnings of the CD class regardless of which class held punishment power (F(1,229)=147, p<0.001; no treatment interaction F(1,229)=0.064, p=0.8). Agent-based simulations: - With short memory (5 rounds) and general defection predisposition (ρ=0.95), agents displayed higher defection relative to humans but reproduced treatment patterns. - T0: Equal earnings across classes. - T1: Rich agents earned about double the poor agents. - T2: Rich agents earned more than double the poor agents. - Varying memory sizes and strategies did not alter the qualitative conclusion; the simulation outcomes closely matched human data when using short memory and high initial defection propensity. Overall: Segregating classes with asymmetric punishment concentrates earnings in the punishing class (increasing inequality), while integration with reciprocal punishment opportunities diminishes payoff disparities.

The findings address whether socioeconomic segregation and asymmetric punishment power shape cooperation and income distribution. In segregated settings where only one class can punish, punishers reduce cooperation while punishing yet earn substantially more, producing large payoff disparities. In integrated settings with random cross-class matching and mutual punishment opportunities, subjects curb defection to enable punishment, leading to only slight earnings differences and reduced inequality. These dynamics suggest that integration mechanisms, rather than segregation, facilitate more equitable outcomes without necessarily increasing cooperative behavior per se. Agent-based simulations corroborate the empirical results across many rounds and parameter variations, indicating robustness of the mechanism: when punishment power aligns with segregation, wealth accrues to the punishing group; when punishment is symmetric in integrated interactions, earnings converge. The behavioral patterns are consistent with short-memory strategies and a high initial tendency to defect, congruent with prior evidence on lower generalized trust in the study context. Policy-wise, reducing structural segregation and enabling balanced enforcement across classes may promote redistribution and attenuate wealth inequality.

This study shows that socioeconomic segregation combined with asymmetric punishment power produces strong payoff inequalities: the punishing class earns about twice as much as the non-punishing class. By contrast, integrating classes and enabling mutual punishment opportunities reduces earnings disparities, effectively redistributing wealth. The results are consistent across human laboratory data and agent-based simulations. The main contribution is linking punishment ability to measured socioeconomic background, demonstrating how social structure (segregation vs integration) mediates the redistributional consequences of enforcement in repeated dilemmas. Future research could (i) test larger and more diverse populations beyond students; (ii) explore intra-class punishment dynamics and richer payoff structures; (iii) vary transparency of partners’ socioeconomic status; (iv) analyze long-term dynamics with alternative memory and learning rules; and (v) examine policy interventions that promote cross-class integration and balanced sanctioning.

- Sample restricted to university students in China, limiting generalizability to broader populations and other cultural contexts. - Socioeconomic status (SES) index was constructed with arbitrary weights; although validated against earnings, measurement error or construct validity concerns remain. - The design imposed unidirectional punishment in segregated treatments, a simplification that may not capture all real-world enforcement mechanisms. - Participants were unaware of partners’ SES, which may differ from settings where status is observable. - Payoff matrices and reduced punishment/defection values were chosen to avoid confounds; different parameterizations might yield different quantitative effects. - No explicit option to terminate interactions, which could influence strategic behavior. - Some reported statistics and treatment labels are complex/partly inconsistent in the text; detailed parameter values and full payoff tables are in supplements, limiting reproducibility from the main text alone.