Learning (not) to yield: An experimental study of evolving ultimatum game behavior

Whether behavior converges toward rational play or fair play in repeated ultimatum games depends on which player yields first. If responders concede first by accepting low offers, proposers would not need to learn to offer more, and play would converge toward unequal sharing. By the same token, if proposers learn fast that low offers are doomed to be rejected and adjust their offers accordingly, pressure would be lifted from responders to learn to accept such offers. Play would converge toward equal sharing. Here we tested the hypothesis that it is regret - both material and strategic - which determines how players modify their behavior. We conducted a repeated ultimatum game experiment with random strangers, in which one treatment does and another does not provide population feedback in addition to informing players about their own outcome. Our results show that regret is a good predictor of the dynamics of play. Specifically, we will turn to the dynamics that unfold when players make repeated decisions in the ultimatum game with randomly changing opponents, and when they learn not only about their own outcome in the previous round but also find out how the population on average has adapted to previous results (path dependence).