Nonmetric test of the minimax theory of two-person zerosum games.

As an experimental test of the minimax theory for two-person zerosum games, subjects played a game that was especially easy for them to understand and whose minimax-prescribed solution did not depend on quantitative assumptions about their utilities for money. Players' average relative frequencies for the moves and their proportions of wins were almost exactly as predicted by minimax, but subject-to-subject variability was too high. These results suggest that people can deviate somewhat from minimax play since their opponents have limited information-processing ability and are imperfect record keepers, but they do not stray so far that the difference will be noticed and their own payoffs will be diminished.