Alexander Sebald, University of Copenhagen: "Measuring nonlinear and heterogeneous guilt aversion"

Abstract

Recent experiments suggest that people’s preferences not only depend on their choices but also on their beliefs of other people’s choices and on their beliefs of other people’s beliefs. Guilt aversion is such belief-dependent preference that has received a lot of attention recently. An individual is considered to be guilt averse if he gets a negative utility from failing to live up to the beliefs of others. Typically, the sensitivity to guilt of a certain individual is taken to be constant, which implies that guilt aversion is linearly increasing in the shortfall between an individual’s second-order belief of another’s payoff and the latter’s actual payoff. We design experiments in order to test whether guilt aversion is indeed a linear function or has another functional shape. Specifically, we characterize the shape of the guilt aversion function at the individual level.

In our experiment we let subjects play binary dictator games in which the dictator chooses between a cooperative choice and a selfish choice. The “responder” is asked for his belief about the move of the dictator, and the dictator is elicited his second-order-belief-dependent strategy. Particularly, for all potential “responder” levels of beliefs, the dictator is asked whether to make the cooperative choice or the nice choice. This method allows measuring guilt sensitivity at the individual level and - given that each individual makes choices for three decision situations that would result in the same guilt sensitivity under the hypotheses the guilt aversion function is linear - allows identifying the guilt aversion function at the individual level.

Our experiment also includes control treatments in order to check whether the strategy elicitation method does not lead to different types of behavior and beliefs as the direct-response method. Our experimental results suggest there is a considerable amount of heterogeneity across individuals.