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

In many social situations, human behaviour differs from the Nash-equilibrium under selfish payoff-maximisation. Numerous social-preference models have been proposed, virtually all of them relying on the Nash-equilibrium concept. This paper determines the Nash-equilibrium sets that result given experiment participants’ elicited preferences, and tests the various aspects of a ‘revealed-preference Nash-equilibrium’ by inducing common knowledge of preferences, using a publicgood situation as an example. The data show that in a three-player public-good situation, multiple equilibria should be expected relatively often (in a third of the cases). Second, most participants’ individual behaviour is in accordance with aspects of Nash equilibrium: most people best-respond to their beliefs, choose equilibrium actions, and consider beliefs that correspond to an equilibrium. However, many participants predict others’ behaviour poorly, which also entails that behaviour rarely is in equilibrium. This points to models like level-k as potential components for better social-preference theories. The experimental findings are obtained using experienced participants and robust to giving participants the option to look up the set of equilibria of their game, and to reducing the number of players to two.