Calibration of Confidence among Eyewitnesses and Earwitnesses

Abstract

One of the more alarming and intriguing results in forensic psychology is the weak relationship between confidence and accuracy in experimental studies of eyewitness identification. This relationship has traditionally been measured by the point-biserial correlation coefficient, r _pb . In the present chapter the confidence-accuracy relationship in witness identification is studied with two alternative and, as we argue, more suitable indices, namely calibration and diagnosticity analysis. When calibration analysis is applied to eyewitness identification, the participant (witness) is required to assess on a scale the subjective probability that the identified person is identical to the culprit. The subjective probabilities are compared to the corresponding relative frequencies of correct identifications. With the use of diagnosticity analysis of confidence, which is based on a modified form of Bayes’ theorem, it is possible to determine the informational impact of positive identifications made with different levels of confidence.

We present three empirical studies. In Study I it was concluded from two experiments that eyewitness confidence can be both reasonably well calibrated and diagnostic, despite a low r _pb . Study II showed that in comparison to eyewitness identification in similar circumstances, earwitness accuracy is poorer, with overconfidence and low diagnosticity of confidence, even in easy tasks. In Study III, a meta-analysis showed that the measures r _pb and calibration were weakly correlated. A modest relation was observed between the r _pb and the diagnosticity index. The calibration and over/underconfidence scores co-varied with task difficulty. Overconfidence was again observed for voice identification tasks.