Biases in Beliefs

Many papers have reported behavioral biases in belief formation that come on top of standard game-theoretic reasoning. We show that the processes involved depend on the way participants reason about their beliefs. When they think about what everybody else or another "unspecified" individual is doing, they exhibit a consensus bias (believing that others are similar to themselves). In contrast, when they think about what their situation-specific counterpart is doing, they show ex-post rationalization, under which the reported belief is fitted to the action and not vice versa. Our findings suggest that there may not be an "innocent" belief-elicitation method that yields unbiased beliefs. However, if we "debias" the reported beliefs using our estimates of the different effects, we find no more treatment effect of how we ask for the belief. The "debiasing" exercise shows that not accounting for the biases will typically bias estimates of game-theoretic thinking upwards.