Theory versus Data in Prediction and Correlation Tasks

Abstract

Research on the ability to assess covariation traditionally has employed dichotomous variables as stimuli and has consistently reported a tendency to perceive covariation where none exists-a tendency exacerbated by prior beliefs about the existence of a relationship. A smaller, more recent literature reports relatively accurate performance when the stimuli are continuous. Five experiments are presented that employ continuous variables while exploring the case in which correlations in the data conflict with expectations. Based on recent findings, performance on a prediction task is used as the dependent measure. Experiment 1 shows a lack of reliance on the data, due primarily to a failure to base responses on a valid but nonintuitive predictor. From the remaining experiments this performance can be attributed to attentional and strategic factors during study and a reluctance to rely on predictors that lack a conceptual connection to the criterion.