The belief revision model: Asymmetrical effects of noncontingency on human covariation learning

Abstract

A noncontingent experience affects the subsequent detection of positive and negative contingencies between the same events. Experiments 1 and 2 showed that such preexposure can produce both an impairment in the detection of subsequent positive contingency and a facilitation of a negative one, independent of the level of contingency during the contingent phase. Experiment 3 raised difficulties for a model that assumes that associations to the context can explain this asymmetrical effect. Experiment 4 suggested that the different weights usually assigned to the different types of trials when computing the contingency between events can change as a result of a noncontingent experience with the same events. This change supports an account of the asymmetrical effect by a belief revision model based on a mechanism that updates the weights of the different trial types as a function of previous experience. More generally, the belief revision model is a statistical (i.e., nonassociative) model of learning that is capable of accounting for trial-order effects, which have long posed problems for statistical models.