Can observers predict trustworthiness?

We analyze experimental evidence on whether untrained subjects can predict how trustworthy an individual is. Two players on a TV show play a high stakes prisoner's dilemma with pre-play communication. Our subjects report probabilistic beliefs that each player cooperates, before and after communication. Subjects correctly predict that women, and players who voluntarily promise that they will cooperate, are more likely to cooperate. They are also able to distinguish truth from lies when a player is asked about his or her intentions by the host. In consequence, and in contrast with the psychology literature, our naive subjects are able to distinguish defectors from cooperators, with the latter inducing beliefs that are 7 percentage points higher. We also study Bayesian updating in the natural and complex context, and find mean reversion in beliefs, and reject the martingale property.

Keywords: trust, promises, Bayesian updating, detecting deception, martingale property of beliefs

JEL Classifications: C72, C93, D64, D83