More than One Kind of Probability Matching: Evidence from a Dual-Task Paradigm

Probability-matching is a well-documented suboptimal behavior that arises in simple prediction tasks. We identify two distinct, local choice strategies that both give rise to probability-matching behavior on a global level. Using a dual-task paradigm, we evaluate the hypothesis that these qualitatively different strategies exhibit different demands on individuals' central executive resources. We find that participants placed under a concurrent working memory are driven away from the one-trial-back strategy utilized by participants without a working memory load and towards a strategy that integrates a longer window of past outcomes into the current prediction. In other words, the demands of the concurrent task appeared to shift the prediction strategies used by decision-makers in our study.