Evidence based rules and learning in symmetric normal-form games

We put forth a general theory of boundedly rational behavior and learning for symmetric normal-form games with unique symmetric Nash equilibria. A class of evidence-based behavioral rules is specified, which includes best-responding to a prior and Nash play. A player begins with initial propsenities towards the rules, and given experience over time adjusts his/her propensities in proportion to the past performance of the rules. We focus on scenarios in which the past distribution of play is revealed to all players. Confronting this theory with experimental data, we find significant support for rule learning and heterogeneity among participants.