Do sequential lineups impair underlying discriminability?

Abstract

Debate regarding the best way to test and measure eyewitness memory has dominated the eyewitness literature for more than 30 years. We argue that resolution of this debate requires the development and application of appropriate measurement models. In this study we developed models of simultaneous and sequential lineup presentations and used these to compare these procedures in terms of underlying discriminability and response bias, thereby testing a key prediction of diagnostic feature detection theory, that underlying discriminability should be greater for simultaneous than for stopping-rule sequential lineups. We fit the models to the corpus of studies originally described by Palmer and Brewer (2012, Law and Human Behavior, 36(3), 247-255), to data from a new experiment and to eight recent studies comparing simultaneous and sequential lineups. We found that although responses tended to be more conservative for sequential lineups there was little or no difference in underlying discriminability between the two procedures. We discuss the implications of these results for the diagnostic feature detection theory and other kinds of sequential lineups used in current jurisdictions.

Active exploration of faces in police lineups increases discrimination accuracy

“…Sometimes the results of an ROC analysis based on an atheoretical measure like pAUC do not agree with the results based on a theoretical measure like d' obtained by fitting a theoretical model to the same data (e.g., Kaesler et al, 2020;Rotello & Chen, 2016;Wilson et al, 2019). Therefore, we fit a signal-detection model to our data, which confirmed our pAUC findings.…”

“…We note that some sequential lineup studies use a stopping rule in which the first lineup member that is identified terminates the lineup procedure. The use of a stopping rule in sequential lineups can impair empirical discriminability as measured by ROC even if d' (the degree to which underlying memory signals generated by innocent and guilty suspects overlap) is not reduced (Wilson et al, 2019; see also Kaesler et al, 2020). The simultaneous superiority effect that we observed in Experiment 2 cannot be explained by the artificial constraint on empirical discriminability imposed by the use of a stopping rule, since we did not use a stopping rule in our sequential procedure.…”

“…The diagnostic-feature-detection theory suggests that discrimination accuracy is enhanced by procedures that enable witnesses to more easily detect and discount features that are non-diagnostic (i.e., are shared in common by all lineup members, . The theory was developed to account for the findings that simultaneous lineups, in which all of the faces are displayed at the same time, can yield a higher ROC curve and higher d' than sequential lineups in which the faces are displayed one at a time (see Seale-Carlisle et al, 2019 for a mini meta-analysis, but also Kaesler et al, 2020). One benefit of an interactive lineup is that the faces can be presented simultaneously, harnessing the potential benefit of simultaneous comparison of faces in a lineup, while still allowing an opportunity for retrieval support (e.g., active exploration, pose-reinstatement, and motion cues).…”

A Model of Position Effects in the Sequential Lineup

“…In the analysis of identification performance, it is important to distinguish between empirical discriminability and underlying discriminability or memory strength . This is because the shape of an ROC curve, and hence empirical discriminability as measured by AUC, may be affected by characteristics of the identification task that are unrelated to memory strength (Rotello & Chen, 2016;Kaesler, Dunn, Ransom & Semmler, 2020). This effect was shown by Wilson et al (2019) who simulated ROC curves for targets and suspects appearing at different lineup positions based on the first-above criterion sequential lineup model described in detail in Kaesler et al (2020).…”

“…This is because the shape of an ROC curve, and hence empirical discriminability as measured by AUC, may be affected by characteristics of the identification task that are unrelated to memory strength (Rotello & Chen, 2016;Kaesler, Dunn, Ransom & Semmler, 2020). This effect was shown by Wilson et al (2019) who simulated ROC curves for targets and suspects appearing at different lineup positions based on the first-above criterion sequential lineup model described in detail in Kaesler et al (2020). Although underlying discriminability was held constant across all target positions, the resulting ROC curves varied substantially and empirical discriminability as measured by AUC was found to decline substantially across target position (see Figure 6A of Wilson et al, 2019).…”