W11 AM Environmental Chemical Forensics
Wednesday, 16 November 2005: 8:00 AM - 11:40 AM in 343-344

(GIN-1118-239671) Statistical tools for the analysis of PCB and dioxin profile data.

Ginevan, Michael¹, Shields, Walt ², Benton, Laurie ², ¹ Exponent, Washington, DC, USA² Exponent, Bellevue, WA, USA

ABSTRACT- When addressing the question of source attribution PCB and dioxin contamination it is common to look at the relative composition of congeners across a set of samples, commonly referred to as "evaluation congener profiles." While one can often often get a good idea of the relative similarities across samples using procedures like principal components analysis (PCA) or by simple visual examination of graphs of profiles, it is more difficult to say which samples are statistically most similar to one another or whether two samples are statistically different. Our presentation shows that log-transformed ratios of scaled concentrations will generally be expected to follow a normal distribution and uses this fact to construct difference metrics that have defined statistical properties. We then use this result to develop hypothesis tests of whether the profiles of two samples or two groups of samples are statistically different.

Key words: Dioxins and PCB's, Profile Analysis, Ratio statistics