The Dynamics of Individual Preferences in Repeated Public Good Experiments

We investigate the stability of individual behavior in a repeated public good experiment over time by reinviting subjects back to the lab up to four times in one week intervals. We exclude effects due to learning about others' behavior and reputation building by employing a non-learning and non-reputation environment: subjects are neither told nor paid their earnings until the very end of their participation and thus deprived of any feedback information and strategic possibilities to signal their intentions. This experimental design thus leaves unstable preferences as the most likely source for unstable behavior. We observe that, in the first wave of the experiment, subjects contribute to the public good in accordance to other-regarding preferences, but become more selfish in the latter waves of the experiment and consequently contributions to the public good decrease over time. The decline is mainly caused by initially conditional cooperators who turn into free riders over the course of the experiment.