Adjustment to disconfirming evidence in a covariation judgment task : the role of alternative predictive relationships

This project investigated the impact of sustained disconfirmation on an acquired belief in a covariation judgment task. Both epistemology and the philosophy of science suggest that data which oppose a hypothesis might not dictate the revision of the hypothesis unless an alternative hypothesis can explain the negative evidence and replace the previous hypothesis. As well, the literature on human categorization and reasoning documents a preference for examples and test instances which confirm rather than disconfirm a prior hypothesis. It was therefore predicted that upon the presentation of negative data for an acquired correlational expectation, subjects would abandon their disconfirmed hypothesis with greater ease if the negative evidence was supplemented with alternative hypotheses. A series of four experiments examined this prediction. Using a within-subjects design, subjects first learned that certain predictor variables signalled the presence of certain outcome variables. In a second phase, the outcomes were systematically presented in the absence of the predictors. Adjustment to the negative evidence was measured on the basis of estimates of correlation and the subjects' tendency to predict the presence of the outcomes on trials where the predictors were present. There were three experimental conditions. In the first, an alternative predictor was present on all trials where the outcomes occurred in the absence of the original predictor. In a second, an alternative outcome was present on all trials where the original outcome was absent. In a third, the negative evidence was not framed in terms of either alternative predictors nor alternative outcomes. While all three conditions produced the same reductions in correlation estimates, the condition without alternatives produced perseverance in outcome predictions in the presence of the predictors. This pattern of adjustment was observed in a simulated medical diagnostic task (Experiment 1), and in a nonmedical