A generalized signal detection model to predict rational variation in base rate use

The predominant models of rational behavior currently used to analyze a large class of experiments imply that subjects neglect or place insufficient weight on base rates when making probabilistic judgments. We argue that the evidence is inadequate for this conclusion because the models make needlessly restrictive assumptions about how base rates should be used. The restrictive assumptions stem from a misuse of Bayes' rule that ignores specific aspects of how the proportions arose. We develop a model of rational behavior that generalizes signal detection theory to reflect the environment subjects routinely face and we reexamine the relevant experimental literature. Variation observed in subjects' responses to base rate information is explained by the present rational model more fully than by extant models.