Evaluating reporting practices in fingerprint comparisons using information theory: five response categories are better than three

Abstract Forensic fingerprint examiners communicate the results of their comparisons using a categorical decision scale that traditionally has only two or three categories. Newer scales with five categories have been proposed and, in some cases, adopted. In the present work, we model existing fingerprint comparison data using an approach that characterizes the amount of information provided by examiner decisions when they use a given scale. We demonstrate that the five-response-category scale produces an overall higher information gain, maintains high information gain even for risk-averse examiners, is more tolerant of non-optimal decision thresholds, and tends to encourage examiner decision thresholds that are very close to optimal, especially the critical decision threshold that determines an identification response. We further show that all of these advantages of the five-response-category scale are robust against variations in the base rates of a mated pair. In addition, there appear to be relatively few negative aspects of this scale, and larger scales are unlikely to produce marked improvements over five categories. The current work is an endorsement of efforts to shift to five-response-category decision scales in the friction ridge discipline.