We propose an new experimental design to measure the distribution of guilt sensitivity in a population and to relate this distribution to the level of stakes (payoffs) in a game. The experimental design implements a sequence of mini-dictator games where information about the beliefs of players varies exogenously across games.
The design rules out reciprocity motives and can separate behavior consistent with various levels of guilt sensitivity from behavior consistent with distributional concerns.
We use our choice data to estimate a structural model of behavior which assigns players to one of two classes. The first class consists of stake insensitive players whose guilt sensitivity does not vary with the level of stakes. The second class consists of players whose sensitivity to guilt varies with the stakes. The model allows to recover the distribution of preferences within each class as well as the proportion of players belonging to each class. Our model estimates suggest that slightly more than half of subjects are stake insensitive. We further measure modest heterogeneity of guilt sensitivity within this class. We find significantly more heterogeneity of preferences in the stake sensitive class. In the latter class, we find that the average level of guilt sensitivity significantly decreases with the stakes. Finally, we find a positive and significant correlation between our economic measure of guilt sensitivity and a psychological measure of guilt sensitivity measured using a TOSCA questionnaire.