Learning by imitation in games: theory, field, and laboratory

We exploit a unique opportunity to study how a large population of players in the field learn to play a novel game with a complicated and non-intuitive mixed strategy equilibrium. We argue that standard models of belief-based learning and reinforcement learning are unable to explain the data, but that a simple model of similarity-based global cumulative imitation can do so. We corroborate our findings using laboratory data from a scaled-down version of the same game, as well as from three other games. The theoretical properties of the proposed learning model are studied by means of stochastic approximation