ABSTRACT

Basic Information

Abstract Number: 1320 - 6
Author Name: John W McIlroy - Michigan State University
Session Title: New and Emerging Analytical Technologies in Forensic Science
Event Type: Symposia
Event Title: Applications of Multivariate Statistics in Forensic Science

Presider Name:Ruth SmithCo-Author:Ruth Smith, Victoria L McGuffin
Affiliation:Michigan State UniversityAffiliation:Michigan State University

Date: Tuesday, March 19, 2013
Start Time: 04:25 PM (Slot #6)
Location: 118C

Abstract Content

A National Academy of Sciences (NAS) report published in 2009 highlighted limitations in objective comparisons of forensic evidence. These limitations can be addressed using statistical procedures; however, due to the complexity of the data routinely generated in forensic laboratories (e.g. chromatograms, spectra, etc.), the application of basic statistics is limited. Therefore, multivariate statistical procedures are necessary for objective comparison of these types of data.

This presentation will demonstrate the application of multivariate statistical procedures to a variety of forensically-relevant data. Examples include chromatograms of fire debris samples, infrared spectra of controlled substances, and elemental profiles of document paper. Prior to statistical analysis, data pretreatment procedures must be considered to minimize non-chemical sources of variance within the data set. The application and evaluation of pretreatment options for each data type will be discussed and demonstrated.
Multivariate statistical procedures can be used for exploratory analysis or classification purposes, both of which are useful for objective forensic comparisons. Exploratory procedures are used to investigate relationships among samples, while classification procedures can be used to assign questioned samples to known reference standards. Examples of both exploratory and classification procedures will be demonstrated for the different data types. In addition, the utility of these procedures will be discussed and recommendations will be given based on the type of data being considered.

Application of multivariate statistical procedures in this manner begins to address limitations highlighted in the NAS report by providing more objective methods for comparison that can be applied to a variety of evidence types.