Justice among strangers. On altruism, inequality aversion and fairness

We present an axiomatic model of choice involving two agents, motivated by the experimental evidence on non-selfish preferences. We distinguish two classes of social preferences, depending on whether they are or not separable. Altruism and spite (Andreoni & Miller, 2002; Cox et al., 2007) are separable, while the various forms of inequality aversion are not (Fehr & Schmidt, 1999; Bolton & Ockenfels, 2000; Charness & Rabin, 2002). Separable and non-separable preferences give very close predictions when only sure outcomes are involved, but they make opposite predictions in choices involving lotteries. We show this by proposing a generalization of expected utility that accounts for preferences for �fair procedures�, which violate the independence axiom. An experimental test of the model reveals little evidence of ex-post inequality aversion, even when non-expected utility preferences are accounted for.