Poster #49 - Expert Opinions on the Accuracy and Variability of Forensic Methodologies

• In Event: ASC Poster Session II

Abstract

Though forensics is a mainstay of the criminal justice system and its presentation is a common expectation among jurors, there are no current estimates of the accuracy of forensic methods beyond the designations of “foundationally valid” or “not foundationally valid” from 2016’s PCAST report. A dearth of statistically rigorous black box studies and the reliance on the current Daubert standard in which judges screen expert witnesses for testimony in court results in a system where jurors are asked to assess forensic testimony describing methods that may or may not be well-regarded within the larger scientific community. In this study, we survey forensic expert practitioners and researchers on the accuracy and variability of six forensic methods. While the majority of experts consider themselves more skeptical than both the wider pool of practitioners and scholars, belief in the accuracy of these methods remains high, even for controversial fields associated with high-profile wrongful convictions, and does not differ much from surveys of the broader public. We also demonstrate that ConsensusGPT, a large language model created to summarize the research corpus, approximates expert opinion well even without the current availability of such expert surveys in its underlying data.

Authors

• Maya Moritz, University of Pennsylvania

• Maria Cuellar, University of Pennsylvania

• Shichun Ling, California State University, Los Angeles