Abstract. Experimentalists frequently claim that human subjects in the laboratory violate game-theoretic predictions. It is here argued that this claim is usually premature. The paper elaborates on this theme by way of raising some conceptual and methodological issues in connection with the very definition of a game and of players ’ preferences, in particular with respect to potential context dependence, interpersonal preference dependence, backward induction and incomplete information.

This paper reports an experiment that elicits subjects ’ initial responses to 16 dominance-solvable two-person guessing games. The structure is publicly announced except for varying payoff parameters, to which subjects are given free access. Varying the parameters allows very strong separation of the behavior implied by leading decision rules. Subjects ’ decisions and searches show that most subjects understood the games and sought to maximize payoffs, but many had simplified models of others ’ decisions that led to systematic deviations from equilibrium. The predictable component of their deviations is well explained by a structural nonequilibrium model of initial responses based on level-k thinking. (JEL C72, C92, D83)... professional investment may be likened to those newspaper competitions in which the competitors have to pick out the six prettiest faces from a hundred photographs, the prize being awarded to the competitor whose choice most nearly corresponds to the average preferences of the competitors as a whole; so that each competitor has to pick, not those faces which he himself finds prettiest, but those which he thinks likeliest to catch the fancy of the other competitors, all of whom are looking at the problem from the same point of view. It is not a case of choosing those which, to the best of one’s judgment, are really the prettiest, nor even those which average opinion genuinely thinks the prettiest. We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practice the fourth, fifth, and higher degrees.

Combining a strategy model, an inference procedure and a new experimental design, we map sequences of observed actions in repeated games to unobserved strategies that reflect decision-makers ’ plans. We demonstrate the method by studying two institutional settings with distinct theoretical predictions. We find that almost all strategies inferred are best responses to one of the inferred strategies of other players, and in one of the settings almost all of the inferred strategies, which include triggers to punish non-cooperators, are consistent with equilibrium strategies. By developing a method to infer unobserved repeated-game strategies from actions, we take a step toward making game theory a more applied tool, bridging a gap between theory and observed behavior. JEL classification: C72, C80, C90

Because human decisions are the result of cognitive processes, theories of human behavior rest at least implicitly on assumptions about cognition. Neuroeconomics reflects the belief that using evidence on neural correlates of cognition will lead us to better theories of decisions. Gul and Pesendorfer (2005, henceforth “GP”) argue that, on the contrary, because economic

Backward induction is the most widely accepted principle for predicting behavior in dynamic games. In experiments, however, players frequently violate this principle. An alternative is a 2-parameter “dynamic level-k ” model, where players choose a rule from a rule hierarchy. The rule hierarchy is iteratively defined such that the level-k rule is a best-response to the level-(k − 1) rule and the level-∞ rule corresponds to backward induction. Players choose rules based on their best guesses of others ’ rules and use past plays to improve their guesses. The model captures two systematic violations of backward induction in the centipede game, limited induction and time unraveling. The dynamic level-k model can be considered as a tracing procedure for backward induction because the former always converges to the latter in the limit.

This paper reports experiments that elicit subjects ’ initial responses to 16 dominancesolvable two-person guessing games. The structure is publicly announced except for varying payoff parameters, to which subjects are given free access, game by game, through an interface that records their information searches. Varying the parameters allows strong separation of the behavior implied by leading decision rules and makes monitoring search a powerful tool for studying cognition. Many subjects ’ decisions and searches show unambiguously that they understand the games and seek to maximize their payoffs, but have simple non-equilibrium models of others ’ decisions, which lead to systematic deviations from equilibrium.

There is a tendency for some economists who are less familiar with game theory to comment that it has never fulfilled its early promise, that it essentially has little to offer as a tool to understand, predict and guide human social and economic behaviour. For instance in Sylvia Nasar’s book ‘A Beautiful Mind ’ she recounts the remarkable controversy that surrounded the award of a Nobel prize for game theory, including one committee member apparently asking for ‘a single major example that game theory had any empirical validity whatsoever ’ (p.371). An excellent reply to such sceptics would be to present them with a copy of Colin Camerer’s new book, ‘Behavioral Game Theory’. This is not to say that his book catalogues an unbroken string of successes for the theory; as we shall see neither game theory nor human behaviour emerges unscathed from the body of work reported therein. Camerer’s book is an up-to-date and comprehensive look at a relatively new but fastgrowing area of economics concerned with using experimental methods to investigate strategic interaction. Grouped into nine chapters he summarises several hundreds of studies on different aspects of game theory, from mixed-strategy equilibrium to