Proceedings of the Annual Meeting of the Cognitive Science Society

Extensive research on probability learning has reported on the ubiquity of the probability matching strategy—choosing options in proportion to their probability of being correct. The current paper explores why the optimal strategy in this task (always choosing the higher probability option) is not intuitive for participants, by examining their decisions in relation to their working memory capacities. We hypothesize that probability matching is a by-product of an automatic recency-based strategy produced by limits in working memory storage and that deliberate strategizing mediated by working memory processing can override recency in favor of optimal responding. A variant of the Expectancy-Valence Learning Model is fit to participant data from a two-choice probability learning task using hierarchical Bayesian modelling. Point estimates of the best-fitting parameter values are then correlated with working memory measures. Results indicate close relations between them, providing support for our hypothesis.