Score based procedures for the calculation of forensic likelihood ratios – Scores should take account of both similarity and typicality

Abstract

Score based procedures for the calculation of forensic likelihood ratios are popular across different branches of forensic science. They have two stages, first a function or model which takes measured features from known-source and questioned-source pairs as input and calculates scores as output, then a subsequent model which converts scores to likelihood ratios. We demonstrate that scores which are purely measures of similarity are not appropriate for calculating forensically interpretable likelihood ratios. In addition to taking account of similarity between the questioned-origin specimen and the known-origin sample, scores must also take account of the typicality of the questioned-origin specimen with respect to a sample of the relevant population specified by the defence hypothesis. We use Monte Carlo simulations to compare the output of three score based procedures with reference likelihood ratio values calculated directly from the fully specified Monte Carlo distributions. The three types of scores compared are: 1. non-anchored similarity-only scores; 2. non-anchored similarity and typicality scores; and 3. known-source anchored same-origin scores and questioned-source anchored different-origin scores. We also make a comparison with the performance of a procedure using a dichotomous "match"/"non-match" similarity score, and compare the performance of 1 and 2 on real data.

Using the likelihood ratio in bloodstain pattern analysis

“…Applications of LRs in the context of forensic expertise [20] are not only limited to evaluate the relation of the evidence with either its source, but also to the associated activities, or the offense itself. These three types of propositions are what Cook et al define as the hierarchy of propositions [35], explaining that propositions of lower hierarchy (about source) are easier to weigh than higher ones (about activity or offense).…”

System Performance as a Function of Calibration Methods, Sample Size and Sampling Variability in Likelihood Ratio-Based Forensic Voice Comparison

“…At the feature-to-score stage, acoustic data are extracted from recordings and used to generate speaker models. Pairs of speakers models (both SS and DS) are compared to generate LR-like scores which capture the similarity and typicality [2] between samples. The scores from the training data are then used at the score-to-LR stage (i.e.…”

Avoiding overstating the strength of forensic evidence: Shrunk likelihood ratios/Bayes factors

“…8 Scores should be calculated in a manner which captures information about both similarity and typicality, i.e., the procedure for calculating a score should be an attempt to estimate a log likelihood ratio value. Scores based only on similarity lack information about the typicality of the questioned-origin data with respect to the relevant population, and that information cannot be adequately incorporated as part of the score to likelihood ratio conversion stage [26]. 9 [27] describes some models which are in principle non-monotonic, but which in practice may be monotonic (but not linear) within the operating range of the system.…”