Home | Departments | Researchers | Overview
Development
of new statistical methods for, and statistical analysis of, the reliability of evidence
identification.
(Carriquiry)
Forensic investigation typically involves the analysis of datasets that can often be massive. A solid understanding of statistical theory and methods is a valuable complement to graduate-level education in the forensic sciences. Statistics plays a role not only in the analysis of data that may be collected in the course of an investigation, but also in the construction and updating of reference databases, and in the determination of the probative value of evidence obtained while investigating a crime. During the last several years, the use of DNA evidence in court has led to increased statistical sophistication of judges and juries, who now often expect precise probabilistic statements accompanying other types of evidence what it is presented in court. Accompanying a witness' statement that a piece of evidence found at a crime scene matches some other evidence found in the possession of a suspect, juries now expect to receive as well an assessment of the probability that the match has occurred by chance alone. In other words, courts expect an estimate of the likelihood ratio that weighs the relative probabilities of observing the evidence given two competing hypotheses: the suspect was at the crime scene and deposited the evidence, or the suspect was not at the crime scene. Thus, establishing the probative value of evidence today requires that the probability of a coincidental match be estimated along side the probability of a "real" match.
A quantitative assessment of evidence of the type discussed above requires the construction of scientifically defensible reference databases. Such databases must be built by combining subject-matter knowledge with solid experimental design principles, and with the appropriate statistical methods for the analysis of a wide range of quantitative information. For evidence consisting of a vector of specimen attributes (such as the concentration of trace elements in bullet lead, or the size of the peaks at various chemical elements resulting from a chromatographic analysis) there is a wide range of statistical tools that can be employed for analysis. Yet for other types of evidence, such as fingerprints, less standard techniques, perhaps involving non-parametric methods or even image processing methods might be more effective. Still in other situations, computationally intensive approaches such as those used in data mining, might provide the most information.