When confident eyewitnesses are mistaken: a mixture model account for the confidence-accuracy relationship.

An influential claim suggests that highly confident identifications have a very high probability of being accurate when police use scientifically validated best practices for identification procedures. However, there are exceptions. Using simulation data, we demonstrate that when memory strength becomes sufficiently poor, the high-confidence criterion needed to achieve a high probability of accuracy becomes implausibly large. As a corollary, highly confident identifications may not be associated with a high probability of accuracy (e.g., under poor viewing conditions). We review field studies that provide estimates of lineup-based d' and conclude by introducing a theoretical account (i.e., mixture model) that can account for highly confident but inaccurate eyewitnesses when memory is sufficiently poor. According to the mixture model, high-confidence judgments derive from two processes whose relative impact depends on the quality of witnessing conditions (e.g., good vs. poor). The model predicts that, as memory weakens, high-confidence accuracy declines.