Realistic gossip in Trust Game on networks: the GODS model

TL;DR

The paper tackles the realism gap in gossip-driven reputation for cooperation on networks by introducing GODS, an agent-based model that couples localized triadic gossip on empirical signed networks with Trust Game dynamics. It explicitly models direct experiences via $I$ and gossip-based reputations via $R$, and compares parallel versus triadic information diffusion under well-mixed, static, and dynamic network regimes, using parameters such as $ abla$ (gossip weighting) and $\\omega$. Key findings show that while perfect information maximizes cooperative success for Cs, realistic triadic gossip often yields growth in resources but amplifies inequalities and requires favorable thresholds; dynamic networks provide the strongest boost to cooperation and Cs' resource share. Overall, the work demonstrates that information structure—local, selective gossip versus global information—profoundly shapes cooperation, growth, and inequality, with practical implications for reputation systems and organizational design.

Abstract

Gossip has been shown to be a relatively efficient solution to problems of cooperation in reputation-based systems of exchange, but many studies don't conceptualize gossiping in a realistic way, often assuming near-perfect information or broadcast-like dynamics of its spread. To solve this problem, we developed an agent-based model that pairs realistic gossip processes with different variants of Trust Game. The results show that cooperators suffer when local interactions govern spread of gossip, because they cannot discriminate against defectors. Realistic gossiping increases the overall amount of resources, but is more likely to promote defection. Moreover, even partner selection through dynamic networks can lead to high payoff inequalities among agent types. Cooperators face a choice between outcompeting defectors and overall growth. By blending direct and indirect reciprocity with reputations we show that gossiping increases the efficiency of cooperation by an order of magnitude.

Figures (3)

• Figure 1: Panel A: comparison of three density plots (Y axis) of gossip mechanisms (parallel, triadic and simple) in their support for cooperation measured by difference in mean resources between two groups divided by the standard deviation for all of the resources (X axis). Panel B: breakdown of influence of amount of Ds in the system (colors) for each gossip mechanism (X axis) measured in difference in SD (Y axis). Panel C: difference in SD (Y axis) for reputation thresholds (colored bars) grouped for proportion of Ds in the system (X axis).
• Figure 2: Panel A: mean resource level of cooperators (Y axis) for different information conditions and thresholds (colored boxes). Scatterplots with fitted regression lines for parallel (B, D) and triadic gossip conditions (C, E) comparing relative difference in resources (X axis) and overall variability (SD on Y axis in B and C) for different TG interaction regimes (red: dynamic networks, green: static networks, blue: well-mixed). Histograms compare the densities of absolute difference in resources (X axis in D and E) in situations when Cs win (blue) and Ds win (red) for different thresholds (Y axis).
• Figure 3: Density (A,B) and boxplots (C-F) comparing action rules (red: 1, green: 2, blue: 3) in different TG interaction regimes (wm: well-mixed, sn: static networks, dn: dynamic networks). Panels A and B: breakdown of interactions between TG interaction and action rules for triadic (A) and parallel (B) gossip mechanisms. Panels C-F: contribution of TG interaction and action rules for triadic (D, F) and parallel (C, E) gossip mechanisms measured with relative difference (C, D) and mean resources of Cs (E, F).