Accuracy of confidence ratings associated with general knowledge and eyewitness memory.

Abstract

Listen

Translate article

The confidence-accuracy (C-A) relation for general knowledge (GK) and eyewitness memory (EM) was compared in both within- and between-subjects analyses. Researchers in the cognitive tradition tend to use within-subjects designs and to find moderately positive C-A relations, whereas those in the forensic tradition tend to use between-subjects designs and to find no relation. Eighty subjects took part in one of two conditions—EM or GK. No difference between conditions was found on the within-subjects measure of the C-A relation, but there was differentiation with a between-subjects measure. There was a strong positive C-A correlation (r = .58, p < .01) for GK but not for EM (r = -.11, ns). One source of this difference may be the differing opportunities for calibration offered by the two kinds of memory. There is a marked degree of agreement in the cognitive literature that there is a moderate yet robust positive relation between subjects' confidence evaluations and their performance. This relation exists over different populations, tasks, and experimental materials (see Nelson, 1988, for a review). However, Wells and Murray (1984), in a review of 28 studies of the relation between eyewitness confidence and eyewitness accuracy, found the average correlation between confidence and accuracy across all the studies to be an unimpressive .07. Of course, there are many differences between laboratorybased work and the more forensically motivated research that might account for such a difference. Eyewitness memory is based on an event witnessed only once, under nonoptimal conditions, with incidental learning, perhaps even with a strong emotional element. Specific recall of details from episodic memory is required. All these aspects differ from the memory requirements of the typical feeling-of-knowing study, which evokes no emotional involvement and which tests memory for relatively familiar material. As well as differences in the kind of encoding that takes place, there are likely to be differences in the opportunities for confidence calibrations afforded by semantic and episodic memory (Wells, Lindsay, & Ferguson, 1979). For semantic memory, one has many opportunities to self-test one's ability at retrieving particular facts. Moreover, there are specified and agreed upon answers to general knowledge questions, so one not only can gauge how appropriate one's confidence in an answer should be but also can determine one's relative performance compared with others in the same situation. However, for eyewitness memory there is no way of knowing that a specific item retrieved is correct because the event cannot be revisited. There is no agreed upon answer to calibrate one's performance against, nor can one know whether one's ability to recall an event is better or worse than anyone else's