An Empirical Model of Strategic Choice with an Application to Coordination Games

This paper develops structural estimation techniques which can be applied to experimental game data to consistently estimate and test models of strategic choice. I assume that the true game is a Bayesian game of incomplete information and that the observed actions are supported by an equilibrium of this game. The implied structure permits estimation of the distribution of player types. This method provides a unified framework for rigorously testing hypotheses about behavior in games within the theory. Behavior observed in coordination game experiments is inconsistent with the joint hypotheses of rational play and complete information of the game. One possible explanation is that some players are altruistic; another is that subjects are learning as they play the game. Neither can be clearly rejected or supported with usual empirical approaches. The structural approach adopted in this paper does give clear evidence regarding the importance and validity of the alternative explanations.