This paper discusses research underway on forensic error rates performed with the support of a National Institute of Justice Research and Development in Forensic Science for Criminal Justice Purposes Award.
This investigation aims to develop error rate interpretation tools using regression on decision scores and using receiver operating characteristic (ROC) regression models. This report discusses the determination of error rates in forensic evidence and emphasizes the importance of measuring accuracy and performance, and transforming subjective feature-comparison methods into objective methods. In this research program, the authors plan to work within the ROC regression framework for error rate quantification by allowing covariates specific to source subjects and examiners. The authors will study statistical techniques by 1) fitting regression models in order to relate covariates to decision scores, and 2) by fitting ROC regression in order to relate covariates to error rates quantified by the ROC curve. The resulting covariate-specific ROC curves in face recognition, handwriting, and latent print databases will model the relationship between covariates and decision scores, give the error rates for specific values of covariates. The resulting covariate-adjusted ROC curve will provide error rates by accounting for covariates. These ROC curves will be compared with the pooled ROC curves studied in the forensic literature. The authors will then relate these ROC methods to LR in terms of trace and pattern evidence interpretation by accounting for covariates.