Nondiagnosticity and Typicality Effects in Prediction

Abstract

Previous work (Nisbett, Zukier, & Lemey, 1981; Zukier, 1982) on prediction shows that nondiagnostic information may dilute the relation between a predictor and a criterion: Intuitive predictions based on a mix of diagnostic information and other valueless or nondiagnostic information are considerably less extreme than are predictions based on the diagnostic information alone. The present studies explored the prediction strategies that underlie this dilution effect. In one study, subjects predicted the guilt of a defendant in a murder trial. In a second study, subjects predicted the academic success of college students in terms of high or low grade-point averages (GPAs). In both studies, some subjects received only information about the targets that was diagnostic of an extreme outcome (e.g., information predictive of guilt or of an above-average GPA). Other subjects received the same diagnostic information, as well as other nondiagnostic information describing the target as either typical (average) or atypical (extreme) on dimensions unrelated to the outcome. There was a strong “typicality” effect, such that “typical” nondiagnostic information produced highly regressive judgments while “atypical” nondiagnostic information produced extreme judgments. The results suggest that subjects use a “pattern-matching” strategy in predictions, and that dilution is due to contrasts in typicality, rather than to nondiagnosticity as such.