For instructors

Teaching notes

A tax authority decides which returns to audit while firms decide whether to report in full. Two companies prefer different technical standards, and both prefer a common standard to none. An incumbent threatens a price war against an entrant. Each of these situations can be written as a table of payoffs, and this game lets a class play any such table.

What the game shows

The game tests the predictions of Nash equilibrium in strategic form. Depending on the table, it shows whether people randomize as a mixed equilibrium requires, which of several equilibria a population selects, and whether a threat that is an equilibrium on paper is believed. Because the instructor writes the payoffs, a single change in one cell can be used to test a comparative static of the theory.

The theory

Nash (1950) proved that every finite game has an equilibrium once mixed strategies are allowed, and Nash (1951) developed the concept. In a mixed equilibrium each player's probabilities are set so that the other player is indifferent among the actions the other uses. A player's own payoffs therefore determine the opponent's mixture and leave the player's own mixture unchanged, which is the least intuitive prediction of the theory and the easiest to test. In games of pure conflict such as matching pennies the equilibrium coincides with minimax play.

Games such as the battle of the sexes, hawk and dove, and the stag hunt have several equilibria, and the theory does not say which will be played. The entry game in strategic form has an equilibrium in which the entrant stays out because the incumbent would fight, a threat that would not be carried out if entry occurred. That equilibrium motivates the refinement to subgame perfection. As an account of the systematic deviations from equilibrium, McKelvey and Palfrey (1995) propose that players choose better responses more often than worse ones without always choosing the best.

What to expect

Aggregate frequencies in zero-sum games tend to lie near the minimax prediction, while individual play does not look random. O'Neill (1987) found average frequencies and win rates close to the prediction, and Brown and Rosenthal (1990) showed in the same data that players' choices depended on earlier rounds. Professional tennis shows the same pattern in Walker and Wooders (2001), where win rates did not differ significantly between serve directions and players nevertheless switched direction too often for their sequences to be random.

The prediction about own payoffs fails in a pronounced way. In the matching pennies game of Goeree and Holt (2001), 48 percent of row players chose the first action when the table was symmetric. When the row player's payoff in one cell was raised from 80 to 320, 96 percent chose that action, although the equilibrium mixture of the row player is the same in both tables, and 84 percent of column players moved to the action that defeats it. Games with a unique mixed equilibrium under repeated play are studied further by Ochs (1995).

In the battle of the sexes, Cooper et al. (1989) observed coordination in 41 percent of plays without communication and in 95 percent when one player could announce an intention. For rock, paper, scissors, Wang, Xu and Zhou (2014) report cyclic movement in the choices of a laboratory population where the equilibrium predicts none.

Tips and things to look out for

References