Guilt aversion revisited: An experimental test of a new model

• We develop a new model of “guilt aversion”.
• The new model is consistent with extant experimental results and cannot be rejected by any of them.
• Also, it provides a more refined equilibrium prediction than a previously defined version does.
• We experimentally test the new model in the laboratory by using a trust game with communication.
• The experimental result provides a case against the guilt aversion hypothesis.

In this paper, we experimentally investigate the guilt aversion hypothesis by using a trust game with pre-play communication. For this purpose, we first develop a new version of guilt aversion called personal guilt aversion. It is shown that the new version is consistent with extant experimental results in the literature and thus cannot be rejected by any of them. Given this observation, we then design an experiment that can test the new as well as original versions of the guilt aversion hypothesis. In contrast to the prediction of both versions of the hypothesis, we find that the correlation between elicited beliefs and (trustful or trustworthy) behavior is almost zero even in an environment with pre-play communication. Thus, our experimental result provides a case against the guilt aversion hypothesis.