This paper reports experimental evidence from a series of a simple Dictator Games in which, randomly matched in pair, subjects choose repeatedly one out of four alternatives involving a pair of fixed monetary prizes, one for them and the other for an anonymously matched subject. While in some treatments player position (i.e. the identity of the best paid agent) is known in advance before subjects have to select their favorite option, in one treatment subjects choose under the veil of ignorance, only knowing that either role is equally likely. Finally, we also collect evidence from another treatment, in which the same options correponds to binary lotteries, in which subjects may win one prize or the other with equal probability.
Subjects' decisions are framed in the realm of a simple mean-variance utility maximization problem, where the parameter associated to the variance is interpreted (depending on the treatment) as a measure of risk or inequality aversion, or some combination of the two.