Abstract:
|
Many pattern-comparison disciplines of forensic science are developing quantitative metrics (or scores) to automate the comparison process. A score from a given case is often interpreted by computing a score-based likelihood ratio (SLR), which represents a likelihood of obtaining the observed score when comparing two patterns from a common source divided by a likelihood of obtaining the observed score when comparing two patterns from different sources. SLRs are generally based on consideration of a single score obtained by comparing a single test impression (or control sample) with a given crime scene impression. This talk will argue in favor of viewing SLRs as likelihood approximations and present a corresponding analysis framework that accommodates multiple scores per comparison, replicate control samples from a source of interest, and/or control samples from multiple sources. We illustrate how this framework often leads to more powerful analyses and more demonstrable interpretations.
|