We study experimentally whether heterogeneity of behavior in the Centipede game can be interpreted as the result of a learning process of individuals with different preference types (more and less pro-social) and coarse information regarding the opponent's past behavior. We manipulate the quality of information feedbacks provided after each play. If subjects rely only on their personal database, long run behavior resembles a Self-confirming equi- librium whereby less pro-social types take at earlier nodes due to prediction errors. Aggregate information release decreases heterogeneity of behavior by increasing the passing rates of pro-selfs and play moves towards Bayesian Nash equilibrium.
social preferences learning Self-confirming equilibrium experiment