Poking Holes and Adding Points in Dictator Games

Deviations from choices predicted by self-regarding preferences have regularly been observed in standard dictator games. Such behavior is not inconsistent with conventional preference theory or revealed preference theory, which accommodate other-regarding preferences. By contrast, experiments in which giving nothing is not the least generous feasible act produce data that is inconsistent with conventional preference theory including social preference models and suggest the possible relevance of reference point models. Two such models are the reference-dependent theory of riskless choice with loss aversion and choice monotonicity in moral reference points. Our experiment includes novel treatments designed to challenge both theoretical models of reference dependence and conventional rational choice theory by poking holes in or adding to the dictator’s feasible set along with changes to the initial endowment of the players. Our design creates tests that at most one of these models can pass. However, we do not find that any of these models fully capture behavior. In part this result is due to our observing behavior in some treatments that differs from previous experiments for reasons attributable to implementation differences across studies.