When best-replies are not in equilibrium: understanding cooperative behaviour

To understand cooperative behaviour in social-dilemma experiments, we need to understand the game participants play not only in monetary but in preference terms. Does a Nash-prediction based on participants' actual preferences describe their behaviour in a public-good experiment well? And if not, where does the observed behaviour diverge from the prediction? This study provides an environment which allows to answer these questions: when making their contribution decision, participants are informed about their co-playersÕ priorly-elicited conditional contribution preferences. This induces common knowledge of preferences and thereby leads to direct experimental control over the game participants play. Results show that most people play best-responses to their beliefs. At the same time, beliefs in a third of the cases do not correspond to an equilibrium prediction that is based on the elicited conditional-cooperation preferences. Moreover, more often than not, beliefs are empirically inaccurate. This holds true even in a treatment that gives participants the option to look up the set of equilibria of their game.