Models of other-regarding preferences attempt to increase predictive power by assuming a utility function that includes other-regarding terms, such as altruism, inequity aversion and reciprocity. The practice of weighting other-regarding terms with free parameters increases model flexibility and introduces the risk of estimating an overfitted model. Overfitted models are problematic because, although they may fit well known data patterns, they fail to predict novel patterns. We use data from members of the general population to evaluate if overfitting is a problem of inequity-aversion models in their core domain of ultimatum bargaining. We find across all inequity-aversion models high in-sample fits to the data from the two-person ultimatum game, but low out-of-sample predictive power for the data from the three-person ultimatum game. We provide a solution to this overfitting problem by developing a heuristic model of other-regarding preferences. The model is based on two heuristic processes that empirically have been found to be present in human behavior: (i) satisficing of an aspiration level instead of maximizing, and (ii) lexicographic processing of money, equality, and efficiency concerns instead of weighting and adding them. We show analytically that the heuristic is able to explain diverse patterns, ranging from money maximization to a variety of other-regarding bargaining behaviors. We show empirically that the heuristic fits and also predicts ultimatum bargaining behavior well. Interestingly, a simple version of the heuristic performs competitively with the best-fitting version of the heuristic. In sum, this work demonstrates a way of modeling other-regarding preferences that leads to high predictive power.
