Equilibrium PLay and Best Response to (Stated) Beliefs in Constant Sum Games

In a laboratory experiment, subjects played ten two-person 3x3 constant sum games and stated beliefs about the frequencies of play by their opponents. Contrary to previous experimental evidence, game-theoretical predictions work well: 80% of actions coincided with Nash equilibrium, subjects were good at predicting the action which was played with highest frequency and 73% of actions taken were best responses to stated beliefs. Complexity, measured by the necessary number of rounds of iterated deletion of dominated strategies to reach the equilibrium, did not affect behavior, although whether games were dominance solvable had an effect. We discuss possible reasons why results differ when the games and the experimental procedures are changed.