Public goods games played on hypergraphs, by agents with bounded learning and planning

Public goods games between model agents with bounded rationality and a simple learning rule, which have been previously shown to represent experimentally observed human playing behavior, are studied by direct simulation on various lattices with different network topology. Despite strong coupling between playing groups, we find that average investments do not significantly depend upon network topology, but are determined solely by immediate local network environment. Furthermore, the dependence of investments on characteristic agent parameters factorizes into a function of individual cognitive budget, K, and a simple function $1/(1+c(0)/\beta )$, where $c\left(0\right)$ is the group centrality and $\beta =12.5$ for all networks investigated. Given the good agreement of agent behavior with available experiments, this seems to indicate that even complex societal networks of investment in public goods may be accessible to predictive simulation with limited effort.