The Lab for Social Decision-Making & Risk

One of the main difficulties in the development of descriptive models of learning in repeated choice tasks involves the abstraction of the effect of losses. The present paper explains this difficulty, summarizes its common solutions, and presents an experiment that was designed to compare the descriptive power of the specific quantifications of these solutions proposed in recent research. The experiment utilized a probability learning task. In each of the experiment's 500 trials participants were asked to predict the appearance of one of two colors. The probabilities of appearance of the colors were different but fixed during the entire experiment. The experimental manipulation involved an addition of a constant to the payoffs. The results demonstrate that learning in the loss domain can be faster than learning in the gain domain; adding a constant to the payoff matrix can affect the learning process. These results are consistent with Erev & Roth's (1996) adjustable reference point abstraction of the effect of losses, and violate all other models.