Biases in Belief Reports

Belief elicitation is important in many different fields of economic research. We show that how a researcher elicits such beliefs-in particular, whether the belief is about the participant's opponent, an unrelated other, or the population of others-affects the processes involved in the formation of belief reports. We find a clear consensus effect. Yet, when matching the opponent's action would lead to a low payoff and the researcher asks for the belief about this opponent, ex-post rationalization kicks in and beliefs are re-adjusted again. Hence, we recommend to ask about unrelated others or about the population in such cases, as 'opponent beliefs' are even more detached from the beliefs participants had when deciding about their actions in the corresponding game. We find no evidence of a hindsight bias or wishful thinking in any of the treatments.