How Can Learning More Information Lead to a Worse Outcome? A Probabilistic Formalization of Contextual Bias in Forensic Analysis

• Activity Number: 395 - Addressing Individual Variation to Improve the Analysis of Forensic Evidence
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: Advisory Committee on Forensic Science
• Author(s): Maria Cuellar* and Amanda Luby and Jacqueline Mauro
• Companies: University of Pennsylvania and Swarthmore University and Google
• Keywords: contextual bias; psychology; bayes rule; DAG; formalization; law

Abstract

Forensic analysts and triers of fact need to be aware of (and avoid) errors due to considering extraneous information in forensic analyses, sometimes called errors of contextual bias. They also need to provide quantitative measures of their certainty about source propositions. Although researchers have proposed that Bayes' rule be used to describe the updating of information in forensic analyses, it is unclear how extraneous information should be included. We propose a probabilistic formalization, as well as directed acyclic graphs (DAGs) for clarity, of contextual bias in forensic analysis to describe why bias leads to the improper assessment of guilt, and the proper way of updating information. We hope this formalization can help avoid contextual biases, and that it can serve as a tool for transparency: analysts can use it to explicitly state what information was included at what point in their analysis, and a trier of fact can incorporate each analyst's conclusion appropriately to reach a conclusion about probability of guilt.