Disclosure of belief-dependent preferences in the trust game

We adopt a psychological games perspective to analyze behavior and beliefs in a Trust Game experiment. Subjects are randomly assigned to the role of "truster", A player, and "trustee", B player. Assuming that B subjects may be affected by guilt aversion and reciprocity, we try to elicit their belief dependent motivations with a set of hypothetical questions. We design the experiment so that subjects have no incentives to manipulate and we check that answers are reliable. We have two main treatments. In the No-Transmission (control) treatment, B's (belief dependent) preferences cannot be common knowledge, hence the game has incomplete information. In the Transmission treatment, B's answers to the hypothetical questions are transmitted and made common knowledge between the two matched subjects. In so far as such answers reveal the "psychological type" of B, this treatment approximates a psychological game with complete information. In this case, assuming that players coordinate their expectations on the efficient equilibrium, we should observe trust/cooperation when the revealed type of B is guilt averse (or reciprocal) and no-trust/defection when he is selfish. We also provide qualitative predictions for the incomplete information case, based on a simplified Bayesian psychological game. The main insight is that average behavior is intermediate. We analyze the set of answers of each B subject with a grid estimation algorithm. Most B subjects are not selfish and we observe a dominance of guilt aversion over reciprocity. Coherently with our theoretical insights, our experimental results show that in the Transmission treatment inducing a psychological game with (approximately) complete information behavior is more extreme: in the subpopulation of matched pairs where B is highly guilt averse there is more trust and cooperation than in the corresponding incomplete information setting without transmission; whereas in the subpopulation of matched pairs where B has low guilt aversion there is less trust and cooperation than in the corresponding incomplete information setting. In both information settings, we find that the B subjects' cooperation rate is positively related to guilt aversion.