A Little Learning...: Confidence in Multicue Judgment Tasks390664

Abstract : A variety of discrimination tasks using complex, multifaceted stimuli were presented to subjects either with or without the opportunity to study a number of labeled examples. These tasks included deciding whether handwriting samples were produced by an American or a European, whether an ulcer was benign or malignant and which of three horses was a winner of a race at Aqueduct in 1969. Complex stimuli were chosen so that there would be a high probability that in the labeled study examples, diligent subjects could find some cue(s) highly correlated with the labels. Such capitalization on chance correlations has often been cited as the source of scientists' unwarranted confidence in their theories. As anticipated, subjects who studied examples were consistently overconfident. However, subjects who studied unlabeled examples or no examples at all were equally overconfident. Some reasons for the independence of confidence from immediate experience are discussed.