Optimal network structure for collective performance with strategic information sharing

Researchers have long known that information sharing boosts collective performance, but real-world competition often makes individuals reluctant to share. A new paper on arXiv (2605.00758) tackles this by modeling strategic information sharing in a collective estimation task. The authors — Ye Wang, Andrea Civilini, Anzhi Sheng, Xiaojie Chen, Long Wang, and Vito Latora — place individuals in a network where each must guess the distribution of ball colors in a box, sampling a set number of balls and deciding whether to share that data with neighbors.

Using evolutionary game theory, they derive analytical results showing that the optimal network is a trade-off between the sharing rate and how information is integrated. Surprisingly, collective performance is maximized at an intermediate average degree, regardless of network type (e.g., random, regular, or scale-free). When individual sample sizes vary, the best performance occurs when those with more connections sample fewer balls (inverse proportionality to degree). This framework provides a rigorous foundation for designing organizational or digital networks where strategic sharing is unavoidable.