Optimizing the selection of fillers in police lineups

Abstract

A typical police lineup contains a photo of one suspect (who is innocent in a target-absent lineup and guilty in a target-present lineup) plus photos of five or more fillers who are known to be innocent. To create a fair lineup in which the suspect does not stand out, two filler selection methods are commonly used. In the first, fillers are selected if they are similar in appearance to the suspect. In the second, fillers are selected if they possess facial features included in the witness’s description of the culprit (e.g., “20-y-old white male”). The police sometimes use a combination of the two methods by selecting description-matched fillers whose appearance is also similar to that of the suspect in the lineup. Decades of research on which approach is better remains unsettled. Here, we tested a counterintuitive prediction made by a formal model based on signal detection theory: From a pool of acceptable description-matched photos, selecting fillers whose appearance is otherwise dissimilar to the suspect should increase the hit rate without affecting the false-alarm rate (increasing discriminability). In Experiment 1, we confirmed this prediction using a standard mock-crime paradigm. In Experiment 2, the effect on discriminability was reversed (as also predicted by the model) when fillers were matched on similarity to the perpetrator in both target-present and target-absent lineups. These findings suggest that signal-detection theory offers a useful theoretical framework for understanding eyewitness identification decisions made from a police lineup.

Could precise and replicable manipulations of suspect-filler similarity optimize eyewitness identification performance?

“…In one of their two experiments they manipulated filler similarity relative to the suspect (i.e., using different fillers in target-present and -absent cases). Consistent with the original propositions of Luus and Wells (1991) and Wells et al (1993; see also Oriet & Fitzgerald, 2018)-that increasing filler similarity beyond matching to description risks affecting guilty but not innocent suspect identifications- Colloff et al (2021) reported a decrease in pAUC as filler similarity increased. They contrasted these results with a second experiment where filler similarity was manipulated relative to the target.…”

“…However, it is important to note that the (six) target replacements in target-absent cases were selected to represent the worst-case scenario for innocent suspects: being high in similarity to the targets and likely in need of high similarity fillers to protect them from being mistakenly confused with the target. Under these circumstances, increases in suspect-filler similarity would presumably be more likely to help protect an innocent suspect than if they were low in similarity to the target (i.e., both because a low similarity innocent suspect is less likely to be confused with the target and because increasing suspect-filler similarity would be unlikely to increase the fillers' similarity to the target)-a notion that was formalized in recent modeling conducted by Colloff et al (2021). Furthermore, Lucas et al (2021) flagged their manipulation of filler similarity as potentially protecting innocent suspects to a greater extent than would occur in real cases, as a result of filler similarity being manipulated relative to the targets in both target-present and -absent cases.…”

powe(R)OC: A power simulation tool for eyewitness lineup ROC analyses

“…To find out, I downloaded open data from OSF for five recent, large-N eyewitness lineup papers comprising 8 separate experiments and 22 unique experimental conditions (Akan et al, 2021;Colloff et al, 2018Colloff, Seale-Carlisle, et al, 2021;Colloff, Wilson, et al, 2021;Morgan et al, 2019) There are general trends in the distributions of ID proportions across confidence levels-a curvilinear relationship for culprit-absent suspect IDs and a positive trend for culpritpresent suspect IDs. Despite these trends (which generally align with our current understanding of confidence-accuracy relationships; Brewer et al, 2021), there is a decent amount of condition-to-condition variability (e.g., in the mid-confidence range for culpritabsent IDs and the high-confidence range for culprit-present IDs).…”

“…Thus, this effect POWE(R)OC 14 size readily translates to changes at the base data level required for ROC data simulation 3 . A graphical example of practical effect size application (using data from Colloff, Wilson et al, 2021) is shown below in Figure 4.…”

“…For instance, powe(R)OC's random sampling ensures that samples and ROC curves maintain important general characteristics while still allowing for variation around characteristics of the base data. The figure below illustrates simulation ROC curve variation across random samples for different sample sizes (again using data from Colloff, Wilson et al, 2021) is shown below in Figure 8:…”