On the uncertainty associated with using a signal detection theory model to analyze data from forensic black-box...

A parametric signal detection theory model has been used in recent literature to model data collected from black-box studies of forensic examiner accuracy, and to predict how error rates might change if examiners were to be either more or less demanding in their requirements for making IDENTIFICATION or EXCLUSION calls. Such models depend on latent (i.e., unobservable) scores, summaries of which are only partially estimable from the data collected on categorical conclusion scales. As a result, inferences based on signal detection theory models are sensitive to the probability model used, and imprecise due to the limited information available about the latent scores. This paper explains some of these uncertainties, with the aim of demonstrating that they can be substantial in typical forensics applications.