A comparison of correlation, calibration, and diagnosticity as measures of the confidence–accuracy relationship in witness identification.

The relationship between witness confidence and accuracy (CA) has traditionally been measured by the point-biserial correlation (rpb). Recently, 2 alternative indices for measuring the CA relation have been proposed, namely calibration and diagnosticity analyses (e.g., P. Juslin, N. Olsson, & A. Winman, 1996). In this study, the 3 measures were compared quantitatively using 52 independent data sets. The measures rpb and calibration were weakly correlated, whether computed across earwitness data sets, eyewitness data sets, or all data. Thus, when applied to the same data, these 2 measures sometimes yield different conclusions. A modest relation was observed between the rpb and the diagnosticity of confidence. Finally, calibration and degree of over- and underconfidence covaried with task difficulty, consistent with K. A. Deffenbachers' (1980) optimality hypothesis.