Dynamics of Games: Aggregate Stability and Individual Learning Rules

The study of learning in games encompasses two research lines: (1) explaining the aggregate stability of classes of games when players use inductive reasoning, and (2) explaining the learning rules used by actual players. We mix approaches by considering both stability and learning rules in terms of a partition of the strategy space, where changes to the partition structure imply changes in the stability characteristics of the game. We present experimental results from a set of three-person games and find that: (1) stability is a function of the characteristic of the mixed-strategy equilibrium, and (2) since different learning rules imply different stability characteristics, we can empirically separate two important classes of learning rules, finding evidence of belief learning rather than performance-based learning in our subjects.