REALISM IN CONFIDENCE JUDGMENTS AS A FUNCTION OF WORKING IN DYADS OR ALONE

This study compares the realism of confidence judgments made by indivi duals and pairs of their answers to general knowledge questions, using a within-subjects design. In the initial Control condition, subjects first answered 30 questions and then confidence rated the chosen answe r. For the next 30 questions (Condition 2), they wrote down an argumen t for the chosen answer and then confidence-rated their answer. In the third condition, the 40 subjects were divided into 20 pairs who then answered and confidence-rated the same 30 questions as in Condition 2. Pair members were asked to collaborate on all parts of the task. The results showed that overconfidence decreased in the Pair condition com pared with the Single conditions. Analysis of the interaction in the P air condition showed a higher overconfidence in those instances where one pair member dominated the interaction totally. Other analyses of t he interaction in the Pair condition also supported the importance of argumentation between the pair members for realistic confidence judgme nts. A comparison with previous research suggests that the fact that t he subjects first attempted the questions individually may have increa sed the proportion correct answers and limited the increase in confide nce in the Pair condition. A control study, checking for the effect of subjects answering the same questions twice, found no effect of repet ition for any of the three calibration measures used. (C) 1996 Academi c Press, Inc.