Stats, Glass and Crime: Let's Make the Right Decision

• Activity Number: 504 - The Future of Statistical Consulting and Collaboration
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Felix Jimenez* and Amanda Koepke and Ruthie Corzo and Eric Steel
• Companies: University of Colorado, NIST and National Institute of Standards and Technology and National Institute of Standards and Technology and National Institute of Standards and Technology
• Keywords: forensic science; permutation test; statistical collaboration; trace evidence; sensitivity analysis

Abstract

In the forensic analysis of glass fragments, a questioned (Q) sample is compared to a known (K) sample to determine whether they originated from the same glass source. Forensic practitioners currently use two approaches to compare the K and Q: a qualitative inspection of the u-XRF spectra and match criterion using elemental ratios. The former is subjective, which may result in differing conclusions amongst practitioners, and neither addresses the effect of the fragment’s shape and size on the fluorescence intensity.

To address these shortcomings, a statistical framework is developed to analyze the u-XRF spectra. One procedure is built around a permutation test, and another takes a model-based approach. The sensitivity of the results to differences in the morphology of the fragments is studied. In addition, special attention is paid to how collaboration between the domain experts and the statisticians helped create this procedure. We discovered that adjusting the number of K samples changes the amount of diversity in the glass fragments present. This can in turn influence the rate of false positives/negatives, which affects the usefulness of this procedure in the field.