The Fundamental Properties, the Stability and the Predictive Ability of Distributional Preferences

Social preferences play an important role in a large variety of domains, yet knowledge about their distribution in the broader population is still scarce. In this paper, we study the fundamental properties, the stability and the predictive ability of distributional preferences in four different samples drawn from the general population. We characterize preference heterogeneity using a Bayesian nonparametric clustering method that allows to segregate individuals into distinct types based ontheir behavioral similarities. The method (i) does not ex-ante assume the existence of specific preference types nor does it require characterization of the preference functions, it (ii) allows for the full spectrum of behavioral heterogeneity in the population (ranging from a single representative agent to all individuals being different), and it (iii) makes the tradeoff between parsimony and descriptive accuracy explicit by penalizing the increased complexity from adding new types. In all our four data sets, we identify three fundamentally distinct and empirically relevant behavioral types: a large group of inequality averse individuals, a somewhat smaller, but still large group of altruistic individuals, and a minority of predominantly selfish individuals. This preference characterization is strikingly stable over time, and it is robust across all data sets. We also demonstrate that our type characterization of behavioral heterogeneity has excellent predictive ability: it predicts behavior out-of-sample substantially better than a representative agent model, but it does only perform slightly worse than a model that permits all individuals to differ. Finally, we contrast these results with predictions obtained from a machine learning algorithm,and show that the out-of-sample predictive ability of our type-based model is considerably better than the predictive ability of the machine learning approach.