An investigation of the utility of goodness-of-fit measures used in current facial recognition research

Sketch artist and Identi-kit images of faces have been assessed in various ways. The current project examined the utility, construct validity, and relationships between four goodness-of-fit measures used to quantify the match between a photograph and constructed image. Sketches and Identi-kit composites of 60 target faces were evaluated using four goodness-of-fit measures: six-point similarity ratings, seven-point likeness ratings, and rectangular and quasi-normal distribution rankings. All four goodness-of-fit measures showed consistent relationships with each other, indicating that each dependent variable quantified the same dimension. Examination of relationships between rates of identification in an identification accuracy task and mean goodness-of-fit values for sketch and composite targets provided information regarding the utility and construct validity of individual measures. Two measures, likeness ratings and rectangular ranks, exhibited the most utility as objective and valid means for assessing goodness-of-fit for both sketches and composites. Likeness ratings were recommended over the remaining measures as the best dependent variable for assessing goodness-of-fit between an image and a photograph.