U.S. flag

An official website of the United States government, Department of Justice.

NCJRS Virtual Library

The Virtual Library houses over 235,000 criminal justice resources, including all known OJP works.
Click here to search the NCJRS Virtual Library

Two-Level Model for Evidence Evaluation

NCJ Number
217660
Journal
Journal of Forensic Sciences Volume: 52 Issue: 2 Dated: March 2007 Pages: 412-419
Author(s)
Colin G.G. Aitken Ph.D.; Grzegorz Zadora Ph.D.; David Lucy Ph.D.
Date Published
March 2007
Length
8 pages
Annotation
This paper describes a procedure for the evaluation of evidence with multiple characteristics (multivariate) at two levels.
Abstract
The paper presents a mathematical formula for the "likelihood ratio" (LR) that the source of multivariate evidence collected at a crime scene is the same source as a multivariate item found on the clothing or person of a suspect. The LR formula was successfully tested with a database that consisted of measurements of 8 major elements from each of 4 fragments that came from each of 200 glass objects. It achieved a 15.2-percent false-positive rate and a 5.5-percent false-negative rate. The modeling was then applied to two examples of casework in which glass found at a crime scene was compared with that found linked to a suspect. The procedure models the two levels of variation inherent in many data structures considered in forensics. These levels assess the variability between different items and the variability within items. The method can be adapted to situations in which more levels of variability may be a necessary feature of the data. Multivariate data of this nature have been considered difficult to interpret. The error rates obtained by the described procedure have much to offer forensic scientists interested in an objective evaluation of their evidence. One advantage over existing methods is that it does not assume independence among variables. A second advantage is that the loss of information is restricted to that unaccounted for by the graphical model. It also models distributions of between-group variability that are not normal, which provides greater flexibility. Finally it can model data with many more variables without making unrealistic assumptions. 6 tables, 1 figure, and 13 references