Similarity in Human Behavior: A Geometric and Behavioral Rules-Based Approach to Games

We study partitions of games (called similarity classes) based on topological information of the games and how they are played. We apply formalism, for the first time, to empirical data and behavioral concepts such as level-k reasoning and fairness. We conduct a large-scale experiment on the comprehensive set of all one-shot 2x2 games with payoffs from {1, 2, 3, 4} without repetitions. The similarity classes derived from the experimental data are best explained by a variant of level-k reasoning and a rule that selects the fairest among the Pareto efficient outcomes. At the individual level, given the large number of decisions, most subjects' behavior appears to be driven by a combination of two rules.