Prior uncertainty and amount of diagnostic evidence as variables in a probabilistic inference task

Abstract

Subjects' probabilistic inference capabilities were evaluated in a simulated threat-diagnosis task. Subjects revised probabilities on the basis of equivocal, contradictory, and unreliable evidence. Revisions of subjective probability were compared with theoretical revisions calculated using a modification of Bayes' theorem. Subjects' revisions and the theoretical revisions showed a significantly increasing disparity as the amount of evidence to be processed was increased. The overall disparity between subjects' and theoretical revisions obtained when a uniform prior probability distribution was assumed did not differ significantly from the disparity obtained under an assumed nonuniform prior probability distribution. A general paradigm for complex inference task situations is discussed.