Mixed Strategy Play and the Minimax Hypothesis

This paper reports on a set of experiments designed to discriminate among the possible sources of the failure of the unique mixed strategy minimax equilibrium of the O'Neill (1987) game. First, the experimental design allows one to identify the causes of the serial correlation in subjects' action choices. Second, the design allows one to discriminate among a set of mixed strategies which generate distributions over actions and wins similar to the distributions generated by the minimax strategy. This is accomplished by introducing a new methodology for eliciting mixed strategies, which overcomes some of the difficulties with previous experimental attempts to directly observe mixed strategies. The results clearly show that subjects' choices do not coincide with their minimax strategies, and that players' intended behavior is not minimax. Analysis of the data also reveals that subjects adopt a wide variety of strategies. It is shown that this heterogeneity can rationalize some of the features of the aggregate data that are consistent with the predictions made by the minimax hypothesis