Schedule - Parallel Session 6 - Intelligence and Strategic Behaviour 2

Several past experiments have sought to elicit participants’ beliefs about other players’ strategic behaviour, and then to test the best response hypothesis that players maximize their expected payoffs given these beliefs. To date, these beliefs were restricted to be independent of a player’s own action, in which case no dominated strategy is ever a best response. To accommodate observed choices of dominated strategies, this paper discusses an alternative experiment allowing participants to state beliefs that are action-dependent ” i.e., they may depend on the player’s own action. We allow beliefs to be action-dependent. Such beliefs made their first significant appearance in Newcomb’s Problem, named after the physicist William Newcomb who originally devised it as a thought experiment. The problem was analyzed in depth by the philosopher Robert Nozick (1969), and discussed a few years later in Scientific American (e.g., Gardner 1974). The hypothesis to be investigated here is whether similar action-dependent beliefs could help to explain observed behaviour in experimental games better than is possible with the more restrictive action-independent beliefs that feature in earlier work. To test this hypothesis, we have conducted an incentivized laboratory experiment in three stages. Throughout participants were faced with a variety of strategic games that had previously been used to elicit players’ beliefs and to test the best response hypothesis. The key difference in our experiment was that participants were explicitly allowed to report both action-dependent and action-independent beliefs, along with their decisions in these strategic games. We compared participants’ actual decisions with: (i) Nash equilibrium predictions; (ii) their best-responses given their own stated action-independent beliefs (henceforth, AIB); as well as (iii) their best responses given their own stated action-dependent beliefs (henceforth, ADB). Our major finding is that allowing action-dependence explains the data better than other prominent alternatives. Our results (1) suggest that action-dependent beliefs could play an important role in explaining observed behavior, as well as beliefs, in strategic games; and (2) call for further research about the comparative descriptive and predictive power of both types of beliefs.