Convergence of Different Computationally Efficient Approximations of the Weight of the Forensic Evidence

• Activity Number: 370
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Christopher Saunders and Danica M. Ommen* and Cedric Neumann
• Companies: South Dakota State University and South Dakota State University and South Dakota State University
• Keywords: forensic ; Bayesian ; asymptotic ; M-estimation

Abstract

In this presentation, we will review recent developments for different statistical approximations to admissible decision-rules associated with the quantification of the weight of forensic evidence. These approximations include (1) a Bayes Factor, where the probability density of the evidence is integrated with respect to a MCMC simulation of the respective posterior distributions; (2) a "Bernstein-Von Mises LR", where the posterior distributions of the suspect and population parameters are approximated by multivariate normal distributions; and (3) a "Neyman-Pearson LR", where the probability density of the evidence is assigned using plug-in estimates of suspect and population parameters. Approximations (2) and (3) are computationally interesting in the sense that they are far more efficient than the commonly used approximation (1). Also, we will sketch the development of these approximations using statistical approximation theory, and illustrate their application to a classical forensic dataset of glass fragments.