Positive and negative correlations between confidence and accuracy for the same events in recognition of categorized lists.

Different researchers have reported positive, null, and negative relationships between confidence and accuracy in reports from memory. One possible reason for this paradox is the variety of materials used across experiments, but the two experiments reported in this article show that positive and negative confidence-accuracy correlations can be observed using a single procedure and the same materials. Subjects studied words from semantic categories and then took a recognition test while making confidence ratings. For previously studied items, positive correlations between confidence and accuracy were obtained using three different measures. Yet when confidence-accuracy correlations were assessed for unstudied items from studied categories, the correlations were zero or negative. The critical factors in determining when negative correlations will be found are the similarity of lures to presented items and the type of analysis used. These results indicate that one should be cautious about relying on confidence of recognition when rememberers must decide among highly similar events.