Dynamics of base-rate neglect : Disregarding prior probabilities while accumulating evidence

Prior information is invaluable to decision makers facing noisy evidence, allowing them to increase their chances of choosing the best option. Yet human choice behavior often exhibits base-rate neglect. In order to better understand the mechanisms underlying such sub-optimal behaviors, we psychophysically measured when and how a prior enters the choice process, by experimentally controlling stimulus duration in a perceptual choice task. We found that subjects' choices neglected the prior entirely when they were unconstrained, even when it could have improved their choice accuracy. Instead, subjects only took prior information into account when the amount of evidence they could accumulate was constrained and confidence was too low to be conclusive. Yet in this case their reliance on the prior was consistent with Bayes-optimal decision-making, decreasing systematically with increasingly reliable information. This suggests, further corroborated by subjects' decision confidence, that the prior is maintained by a cognitive mechanism that is distinct from evidence accumulation, which subjects resort to only when accumulated evidence is inconclusive, but not in deciding how much evidence to accumulate. Our psychometric method is more broadly applicable and yields behavioral data that identify directly how the otherwise unobservable decision variable evolves with time. Our data reveal that accumulated evidence is perfectly linear in log-time, confirming the qualitative prediction of drift-diffusion models while providing additional empirical restrictions for their refinement