Error rates for high confidence eyewitness identifications

Eyewitness identification of strangers is vulnerable to error, even if the eyewitness reports high confidence at the initial police identification procedure. In support of this claim, we report a new meta-analysis of data from actual criminal investigations. This analysis shows that when eyewitnesses were tested in the field by a blind lineup administrator, 1/8 of the high confidence identifications were known errors, i.e., mistaken identifications of lineup fillers. We argue that these field data are more informative than the available wrongful conviction data because in the latter eyewitness confidence at the initial identification procedure was almost never recorded. Our claim is also supported by lab data, which show that error rates for high-confidence identifications of the suspect can range from 0 to 40%, depending on the level of bias against the suspect. We highlight three types of suspect bias: appearance-based suspicion, social media contamination, and misplaced prior familiarity.