Thirty-six subjects participated in two-person zero-sum games played against an experimenter in an examination of the effect of different types of game matrices and different mixed strategies on behavior. Results showed that when the opponent played nonoptimally, subjects were able to detect this nonoptimality and to exploit it to their own benefit. When the opponent played according to the minimax prescription, subjects' performance was not optimal but was sufficiently and consistently close to it that arguments in terms of differences between perceived and objective probabilities provide an attractive explanation for the differences.
F.*J
ARIE THEORY has provided a model G which serves as a basis for human decision malting experiments typified by two or more individuals making decisions which jointly affect their respective outcomes. The prescriptions of this model are unambiguous in the restricted case of twoperson zero-sum games. The distinguishing characteristic of such games of pure conflict is that one player's winnings are exactly cquivalcnt to the other's losses. The minimax theorem of Von Neumann and Morgenstern (1944) provides a strategy for the advantaged player guaranteeing him an expected win of no less than a certain amount, called the value of the game. Similarly, a strategy exists for the disadvantaged player guaranteeing him an expected loss of no more than the value of the game. I n a fair game the value is equal to zero.
Games to which the minimax theorem applies are best described in the normal form (Luce & Raiffa, 1957), where the rows and columns of a matrix represent the alternatives, or pure strategies, available to the respective players. In such a representation the cell entries denote the outcome of a joint choice of row and column. Central to the concept of an optimal minimax strategy is a mixed strategy, defined as a probability vector across a player's set of pure strategies.
Empirical studies have compared the normative prescriptions of the minimax