Building a unified statistical framework for the forensic identification of source problems

The interpretation of forensic evidence for source-level identification takes place within the context of two competing propositions for how the evidence has arisen. In this article, we introduce two different frameworks for quantifying the value of evidence in identification of source problems which formally and explicitly state the propositions based on sampling distributions. The first is referred to as the identification of common source problem, and focuses on providing evidentiary support for deciding whether or not two sets of unknown source evidence can be attributed to a common, but unknown, origin in a relevant background population. The second is referred to as the identification of specific source problem, and focuses on providing evidentiary support for deciding whether a single set of unknown source evidence can be attributed to a known, specified source or if it originates from a source in the relevant background population. The motivations for this work come from recent calls for greater transparency and rigour in reporting evidential value, especially from the European Network of Forensic Science Institutes (ENFSI) guidelines, President's Council of Advisors on Science and Technology (PCAST) and the Department of Justice's review of the testimony of forensic examiners. As a working example, we will consider a debate that took place during the recent NIST Colloquium on the Weight of Evidence.