Quantifying the Effects of Database Size and Sample Quality on Measures of Individualization Validity and Accuracy in Forensics

• NCJ Number: 248670
• Author(s): Donald Gantz; Christopher Sanders
• Date Published: March 2014
• Length: 216 pages
• Grant Number(s): 2009-DN-BX-K234
• Agencies: NIJ-Sponsored
• Publication Type: Research (Applied/Empirical); Report (Study/Research); Report (Grant Sponsored); Program/Project Description
• Award: Quantifying the Effects of Database Size and Sample Quality on Measures of Individualization Validity and Accuracy in Forensics

Annotation

This project developed methods for quantifying statistically the random match probability (RMP) that quantifies uncertainty in measures that validate a forensic discipline's basic premises (such as a uniqueness claim) and the accuracy of likelihood ratio methods used in making classification/individualization conclusions.

Abstract

Phase I of this three-phase project focused on the RMP as a measure of the validity of a forensic individualization system. This phase developed theoretically sound upper confidence bounds on the RMP, which are estimated using automated pairwise comparisons. The RMP is related to whether or not forensic analysts should use a given biometric modality in general. In Phases II and III, the project focused on quantifying the accuracy of given forensic modalities in individual applications. The use of likelihood ratio methods in DNA analysis is well-established in addressing this problem; however, research on its use in other forensic areas is not as well-developed. The researchers investigated the use of Bayes Factors and likelihood ratios in other forensic fields, such as handwriting and glass fragments. Phase II focused on popular approximate procedures, and Phase III investigated statistically rigorous formal techniques, extending the Bayesian Likelihood Ratio to situations in which the background population has not been accurately characterized, focusing on statistically rigorous formal techniques. Most of the methodologies developed in this grant will apply to any forensics field, since RMPs and likelihood ratios are defined similarly in many of them. The researchers in thi project are in the process of extending this work to databases of automated comparisons of fingerprints. 5 tables, 3 figures, and 17 references