Ingredients of intuitive regression

Abstract

Conflicting evidence, mainly from multiple-cue probability learning tasks, leans toward the proposition that people predict uncertain events better than they can express the process that generates them. Since data on that question from single-cue probability learning tasks are almost completely lacking, the experiment attempts to supply it. It used scaled variables in four task validities with positive and negative sign and high and moderate cue-event correlations. Cue weightings inferred from subjects' predictions were more accurate than weightings which they made explicit. Brehmer's findings were confirmed that rate of learning is affected by factors different from those affecting final level of achievement. Overtracking was found, but not as an inverse function of cue-validity; previous reports of that finding are challenged. Subjects were more effective in arriving at a predictive scheme which would have worked well if consistently followed than they were in adhering to it. The hypothesis was strengthened that people's effectiveness in predicting uncertain events exceeds their ability to express insight into their prediction process.