The reproducibility of facial approximation accuracy results generated from photo-spread tests

The accuracies of facial approximations have been measured by determining the frequency that examiners recognize correctly matching faces from photo-spreads under blind conditions. However, the reliability of these studies is unknown and warrants investigation since photo-spread results are based on subjective judgements of typically small single groups of examiners (<150 individuals). Moreover, statistical significance tests hold limited value for gauging reliability since these probabilities are only applicable to exactly matched study samples. To redress this issue, this study measured the repeatability of photo-spread results using three previously published facial approximations, the same photo-spread from the original study, and four independent groups of examiners (the original study group (trial 1); and three retest groups: trial 2=c. 40 individuals; trial 3=c. 75 individuals; and trial 4=c. 115 individuals). Across all three facial approximations, differences in recognition rates varied from 0% to 31% between independent samples of examiners. For the nine faces that commonly received high recognition scores, the largest mean difference was 18%. This indicates that when a photo-spread size of 10 faces is used, the results generated from a sample of <115 examiners should be considered approximate, and that any differences in the recognition rates that do not exceed 18% should be considered to be negligible.