Erroneous eyewitness identification evidence is likely the leading cause of wrongful convictions. To minimize this error, scientists recommend collecting confidence. Research shows that confidence can predict the likely accuracy of an identification from a properly administered lineup. However, confidence is far less predictive of accuracy when an eyewitness rejects the lineup instead. This study aimed to strengthen the confidence-accuracy relationship for lineup rejections. I compared the standard, simultaneous procedure used worldwide to the novel “reveal” procedure designed by scientists to boost rejection accuracy. I also compared verbal and numeric confidence using machine-learning techniques. Results showed the reveal procedure produced a stronger confidence-accuracy relationship for rejections than did the standard procedure. Moreover, verbal confidence captured unique diagnostic information about the likely accuracy of a lineup rejection separate from the diagnostic information captured by numeric confidence ratings. These results inform recognition memory models and may increase the diagnostic value of lineup rejections.

We tested whether a lineup constructor’s knowledge of the suspect’s appearance undermines their ability to use the match-to-description method of filler selection. Lineup constructors created lineups using the match-to-description method; they were either blinded or unblinded to the suspect’s appearance. A third group used the match-to-suspect method. The similarity relationships between lineup members were tested in two experiments. In Experiment 1, participants (N = 105) rated suspect-filler similarity. In Experiment 2, participants (N = 100) rated the extent to which each lineup member fit the description. Suspect-filler similarity was higher for match-to-suspect lineups than for blinded match-to-description lineups; the opposite was true for description fit. Contrary to predictions, the unblinded match-to-description method produced lineups that were more similar to the blinded match-to-description lineups than to the match-to-suspect lineups. This suggests that lineup constructors were able to successfully use the match-to-description method, even if they were aware of the suspect’s appearance.

Current DOJ guidelines direct police to construct lineups with fillers similar enough to prevent the suspect from standing out but not so similar that they impair accurate eyewitness identification (DOJ, 2017). However, the optimal suspect-filler similarity remains debated. This study explores competing theories using Betaface, a new facial similarity metric. Some experts suggest that fillers should simply match witness descriptions, while others advocate that fillers should match descriptions but differ in unique features. Some propose fillers closely resemble suspects to protect the innocent, whereas a recent meta-analysis suggests moderate suspect-filler similarity. To refine filler selection strategies, this study manipulates guilty-innocent suspect similarity, lineup construction strategies, target presence, exposure time, and target race. We hypothesize that eyewitness performance will be highest with low (cf. moderate or high) suspect-filler similarity, particularly when guilty-innocent suspect similarity is also low (cf. high). We will discuss our results regarding discriminability, confidence-accuracy relationship, and practical relevance.

When a crime is caught on camera, investigators can use face matching to identify the perpetrator in the footage. We tested this in (a) 1-to-1 comparisons between the footage and an image of a suspect, or (b) face matching lineups, which included the suspect's image alongside images of five non-suspects. For both procedures, participants provided a global confidence rating in their identification decision. In the lineup condition, participants also rated their confidence that the nonidentified lineup members could be ruled out. For suspect identifications, accuracy was positively associated with confidence in both 1-to-1 and lineup conditions. For nonidentifications, the global confidence rating in the identification decision was only related to accuracy for 1-to-1 comparisons and not for lineups. However, using the rule-out ratings, we found that accuracy was positively associated with confidence for nonidentifications of the suspect. Therefore, without rule-out ratings, confidence in lineup nonidentifications will be of limited value.