Technical note: Proposed method to objectively evaluate gunshot residue comparisons does not generalize to...

Previously, we proposed a likelihood ratio system for pairwise source comparison of gunshot residue (GSR) samples based on elemental composition. Only pairs of GSR samples from the same location type (e.g., hand-hand) were considered, as the data originates from casework and ground truth is not available for samples from different location types (e.g., hand-cartridge case). This lack of sample pairs taken from different locations is a limitation, as casework will usually require such sample pairs to be evaluated. In this study, we test the impact of this sampling location limitation by evaluating the model with an experimental dataset for which ground truth was available for same-location as well as different-location pairs. We find a sharp decline in the performance of the model, with C_llr deteriorating from 0.35 to 0.98. Additional exploration of various system extensions does not lead to significant improvements. We discuss the potential causes for this decline and conclude the system is not currently ready for application in forensic casework.