Despite decades of research on heuristics and biases, evidence on the effect of large incentives on cognitive biases is scant. We test the effect of incentives on four widely documented biases: base-rate neglect, anchoring, failure of contingent thinking, and intuitive reasoning. In laboratory experiments with 1,236 college students in Nairobi, we implement three incentive levels: no incentives, standard lab payments, and very high incentives. We find that very high stakes increase response times by 40% but improve performance only very mildly or not at all. In none of the tasks do very high stakes come close to debiasing participants.

We study a dynamic game in which players compete for a prize. In a waiting game with two-sided private information about strength levels, players choose fighting, fleeing, or waiting. Players earn a “deterrence value” on top of the prize if their opponent escapes without a battle. We show that this value is a key determinant of the type of equilibrium. For intermediate values, sorting takes place, with weaker players fleeing before others fight. Time then helps to reduce battles. In an experiment, we find support for the key theoretical predictions and document suboptimal predatory fighting.

Psychologists and economists have argued that rewards often have hidden costs. One possible reason is that the principal may have incentives to offer higher rewards when she knows the task is difficult. Our experiment tests if high rewards embody such bad news and if this is correctly perceived by their recipients. Our design allows us to decompose the overall effect of rewards on effort into a direct incentive and an informational effect. The results show that participants correctly interpret high rewards as bad news. In accordance with theory, the negative informational effect coexists with the direct positive effect.

We investigate whether experimental subjects can predict behavior in a prisoner's dilemma played on a TV show. Subjects report probabilistic beliefs that a player cooperates, before and after the players communicate. Subjects correctly predict that women and players who make a voluntary promise are more likely to cooperate. They are able to distinguish truth from lies when a player is asked about her intentions by the host. Subjects are to some extent able to predict behavior; their beliefs are 7~percentage points higher for cooperators than for defectors. We also study their Bayesian updating. Beliefs do not satisfy the martingale property and display mean reversion.