Does Correlation Matter in Probability Matching? A Laboratory Investigation

Probability Matching, as a well-documented suboptimal stochastic choice, describes subjects’ tendency to mix or match their choice frequency to the outcome probabilities when repeatedly asked to predict two payoff-relevant outcomes that only differ in their probabilities of occurrence. It is a classical failure in expected utility maximization, and previous literature has struggled to find a consistent explanation. We use the correlation and frame in the payoff structure as tools to decompose three broad classes of theories: Consequentialist Decision Theories, Correlation-Sensitive Preference, and Bounded Rationality. Our experimental design uses a diagnostic approach, using three treatments to decompose the three classes. We find that Correlation-Sensitive Preference dominates: subjects fail to make the utility-maximizing choices because they want to avoid the regret of doing so. Our results also emphasize the importance of correlation in individual decision making: subjects are more likely to make suboptimal stochastic choices and the learning process is slow and partial when it is difficult to perceive the correlation. Our findings suggest that some correlation neglect biases could be, in fact, due to the complexity of correlation perception rather than intentionally ignoring the correlation structure