Forensic scientists often make analytical chemical measurements with one of two objectives in mind: to determine the amount of a chemical substance present in a questioned sample, or to classify a questioned sample regarding membership in one or more known groups of objects. The first task is a calibration problem; the second is a pattern recognition problem. Analysis of a single sample can produce as much as 10 MB of data using modern spectroscopic and chromatographic instruments. Despite this wealth of multidimensional information, often just a single measurement is used to make decisions. Statistical data handling in the laboratory enables interpretation of multivariate data in chemically meaningful and statistically valid ways by extracting relationships and dependencies among the variables. With State, District, and even Supreme Court rulings having an impact on the way that scientific evidence is presented in legal proceedings, statistics can also play a significant role in the quality and reliability of scientific evidence.
Examples to be presented include the forensic determination of carboxyhemoglobin in blood samples by UV-Vis spectrometry, the classification of copy toners from infrared reflectance spectra, the identification of automobile paint chips by pyrolysis gas chromatography/mass spectrometry, and the identification of glass types from elemental and physical data. Statistical methods applied to these data sets include principal component analysis, principal component regression, multiple discriminant analysis (canonical variates analysis), cluster analysis, and classification and regression trees (CART).