South Dakota State University
Approximate Solutions to the Forensic Identification of Source Problems
Traditionally, forensic identification of source problems are an application of methods for non-nested model selection. The goal is to determine which model for the generation of evidence is supported by the evidence where one model corresponds to a prosecution hypothesis and the other to a defense hypothesis. The typical approach for solving these problems is to use a fully subjective Bayesian method to construct a summary statistic that allows a decision-maker to update his/her belief about the relative merit of each of the competing forensic hypotheses. Currently, the fully rigorous Bayesian approaches are too computationally intensive to be feasible in practice. Therefore, several approximate solutions to the forensic source identification problem will be needed. In this presentation, we will discuss current research initiatives to develop appropriate approximations to the Bayesian value of forensic evidence. These approximations include a modified Neyman-Pearson likelihood ratio and the recently developed Bernstein von-Mises approximation. This presentation will show that the newly developed approximations are asymptotically equivalent to the Bayes Factor. Finally, an example applying the approximations to a dataset for the trace elemental analysis of copper wire is provided.
Refreshments at 3:45pm in Snedecor 2101.