Applications of Multi-Variate Statistics to Forensic Science

Upon being presented with a questioned and known sample of evidence, the duty of the forensic scientist is to evaluate if the two samples may share a common origin.  This can be accomplished by examining class characteristics, or features that can distinguish groups of samples from one another but cannot individualize an item.  Hopefully, the probability of two unrelated samples having indistinguishable class characteristics is low.  However, the burden is on the forensic scientist to evaluate this risk by demonstrating that the sample type in question is inherently diverse.  Furthermore, reliable laboratory techniques must exist that can discern this diversity in both pristine and evidentiary samples.

Chemometric techniques have been increasingly used to study data and sample types that are relevant to forensic science.  This talk will discuss the use of multi-variate statistical methods such as agglomerative hierarchical cluster analysis, principal components analysis and discriminant analysis on various types of data such as chromatograms, mass spectra, UV-visible absorbance spectra and FTIR spectra.  Applications of chemometrics to forensic samples in our research has included the analysis of human hair, cotton fibers and automotive clearcoats by UV-visible microspectrophotometry, pigmented inks by pyrolysis gas chromatography-mass spectrometry (GC-MS), black electrical tape by FTIR and ignitable liquids by GC-MS.  By combining instrumental and statistical techniques, issues such as the extent to which evidence can be truly differentiated, which analytical techniques are more discriminating, and quantitative associations of questioned and known samples can be addressed.