Quantified Assessment of AFIS Contextual Information on Accuracy and Reliability of Subsequent Examiner Conclusions

• NCJ Number: 235288
• Author(s): Itiel Dror; Kasey Wertheim
• Date Published: July 2011
• Length: 61 pages
• Grant Number(s): 2009-DN-BX-K224
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)
• Award: Quantified Assessment of AFIS Contextual Information on Accuracy and Reliability of Subsequent Examiner Conclusions

Annotation

This study examined the impact of using Automated Fingerprint Identification Systems (AFIS) to facilitate human decisionmakers, with attention to potentially biasing effects on human experts of AFIS contextual information.

Abstract

The findings indicate that AFIS affects human examiners, but in some ways more than others. No guidelines or stipulations exist regarding how fingerprint examiners may need to change their decisionmaking threshold when using AFIS. AFIS changes the way that comparisons are presented to an examiner. In non-AFIS searches, an examiner is often presented with a limited set of prints for comparison. In an AFIS setting, on the other hand, an examiner is presented with a ranked set of prints, beginning with the most probable match. Most AFIS "hits" are provided as the top candidate on the list. Although this ranking may be useful information for the examiner, it may also create a bias. This research examined the potential biasing effects of the ranking of prints in an AFIS list. In the current study, thousands of AFIS lists were presented to 23 latent fingerprint examiners as part of their normal casework. The matching print was included in some of the lists, and the position of the matching candidate image was manipulated in the AFIS list, placing it either at the very top, near the top, near the bottom, or at the very bottom. The data indicate that latent fingerprint examiners were affected by the position of the matching print, especially for lower comparison time, but they were unaffected by the scores of the matching prints. Erroneous decisions were more likely chosen from the top of the list. Such errors occurred even when the correct match was present further down the list. 17 tables, 47 references, and a listing of the venues where research findings have been disseminated