Comparing the strength of the confidence-accuracy versus response time-accuracy relationship for eyewitness identification

Abstract

Research indicates that eyewitness identification (ID) accuracy increases with faster IDs and those supported with immediate high confidence, but it is not clear which measure, confidence or response time, is the better reflector of accuracy. It is also important to know how well these patterns hold up across important factors affecting eyewitness ID accuracy such as memory strength for the perpetrator's face. We conducted four pre-registered experiments to investigate these issues across different levels of target memory strength (via encoding time or image quality) and ID procedure (showups vs. lineups of different filler quality). Correct IDs were faster than false IDs regardless of memory strength, and this difference was greater for lineups than showups. There was a consistently strong positive CA relationship for those who made an ID, but the RTA relationship was significantly weaker. Both relationships were weaker for those who made a rejection decision, but the CA relationship remained stronger than the RTA relationship. We conclude that immediate confidence may be a more important reflector of accuracy than response time, regardless of the quality of the memory for the perpetrator's face.

Keywords: Confidence; Eyewitness identification; Lineups; Memory strength; Response time; Showups.

Fig. 1

ROC curves depicting greater empirical discriminability with better target memory in the form of either longer encoding time (Pilot, upper-left; E1, upper-right; E2, lower-left) or clearer target image (E3, lower-right). “Long” indicates an encoding time of 5 s; “Short” indicates an encoding time of 1 s.

Fig. 2

CAC (left panels) and RAC curves (right panels), with standard error bars, from the pilot experiment. In the top row, proportion correct is based on identifications (correct/[correct + false IDs]), and in the bottom row it is based on rejections (correct rejections/[correct rejections + incorrect rejections]). “Long” indicates an encoding time of 5 s; “Short” indicates an encoding time of 1 s.

Fig. 3

Example target (top), fair lineup (left), and biased lineup (right) from Experiments 1 and 2.

Fig. 4

Correct ID rate (top left), false ID rate (top right), and discriminability (bottom), with 95% confidence interval bars, from Experiment 1.

Fig. 5

Mean log suspect ID time (with 95% confidence interval bars) from Experiment 1 (top-left), Experiment 2 (top-right), and Experiment 3 (bottom).

Fig. 6

CAC (left panels) and RAC curves (right panels), with standard error bars, from Experiment 1. In the top row, proportion correct is based on identifications (correct/[correct + false IDs]), and in the bottom row it is based on rejections (correct rejections/[correct rejections + incorrect rejections]). “Long” indicates an encoding time of 5 s; “Short” indicates an encoding time of 1 s.

Fig. 7

The value of confidence versus response time as reflectors of discriminability, from E1 (top left), E2 (top right), and E3 (bottom). Bars are 95% confidence intervals.

Fig. 8

Correct ID rate (top left), false ID rate (top right), and discriminability (bottom), with 95% confidence interval bars, from Experiment 2.

Fig. 9

CAC (left panels) and RAC curves (right panels), with standard error bars, from Experiment 2. In the top row, proportion correct is based on identifications (correct/[correct + false IDs]), and in the bottom row it is based on rejections (correct rejections/[correct rejections + incorrect rejections]). “Long” indicates an encoding time of 5 s; “Short” indicates an encoding time of 1 s.

Fig. 10

Example target images from Experiment 3: original (left), blurry (middle), and grainy (right).

Fig. 11

Correct ID rate (top left), false ID rate (top right), and discriminability (bottom), with 95% confidence interval bars, from Experiment 3.

Fig. 12

CAC (left panels) and RAC curves (right panels), with standard error bars, from Experiment 3. In the top row, proportion correct is based on identifications (correct/[correct + false IDs]), and in the bottom row it is based on rejections (correct rejections/[correct rejections + incorrect rejections]).

Fig. 13

Informational value about accuracy derived from confidence versus response time, and identifications versus rejections. These are unweighted averages of the z-values in Table 2 across all experiments. IDs = identifications, Rejs = rejections.