Suspect identification accuracy from lineups, in the lab and in the field.

Abstract

A 2016 field study conducted in collaboration with the Houston Police Department reported that simultaneous lineups were diagnostically superior to sequential lineups, that confidence was strongly predictive of accuracy, and that high-confidence suspect identifications were highly reliable. The study also estimated that most lineups (65%) contained an innocent suspect. Because the innocence or guilt of a suspect in a real police lineup is unknown, however, these conclusions could not be based on direct computations from target-present and target-absent lineups. Instead, they were parameter estimates from a signal detection model fit to the data. A recently published mock-crime laboratory study mirrored key methodological details of the Houston field study, allowing for similar analyses based on direct computations. Here, we compare the results of the two studies and find that they yield similar conclusions. In addition, new model-based analyses of the Houston field data weigh against recent concerns that unfair lineups and other potential biasing factors may have compromised the original model-based estimates. Finally, the lab and field data agree that when encoding conditions are poor (e.g., long viewing distance), witnesses make far fewer high-confidence identifications, but the few witnesses who do express high confidence maintain a high level of accuracy. These findings are consistent with likelihood ratio theories of recognition memory and reinforce a growing consensus that, as encoding conditions become degraded, high-confidence identifications become increasingly rare but are still highly diagnostic. Whether this conclusion holds when conditions are degraded in the extreme is unresolved.