To reach a conclusion of “match” or “non-match,” a forensic analyst necessarily applies some threshold or burden of proof. To reach a conclusion of “guilty” or “not guilty” in a criminal trial, a jury uses the forensic conclusion along with other evidence, but also applies a burden of proof — the familiar beyond a reasonable doubt standard. What is the relationship between these two burdens of proof?1 Should forensic “matches” be made according to the beyond a reasonable doubt standard? If not, then what standard should forensic examiners use?2 Although seemingly technical at first, this question implicates some of the criminal justice system’s deepest values. Applying the wrong forensic threshold could conceivably water down the reasonable doubt standard. Worse yet, by manipulating the threshold, forensic examiners could effectively usurp some of the jury’s role in deciding guilt or innocence.
The recent Organization of Scientific Area Committees (OSAC) Letter3 that is the subject of this Harvard Law Review Forum Commentary Series represents a laudable first step in addressing these questions. The Letter reaches two important conclusions: First, it states that the reasonable doubt standard does not preclude the use of different statistical procedures.4 Second, it argues that “report of a match without more information about the probability of a match . . . would not fulfill the expert’s role of impartially and adequately educating the trier of fact.”5 Both of these statements are facially correct, but ultimately the Letter fails to connect the dots completely. Indeed, as this Commentary will show, the two conclusions are fundamentally linked: if the jury receives proper contextual information — specifically, the likelihood ratio associated with the “match” or “non-match” — then the burden of proof does not require a specific statistical procedure.
The surprising answer to the original question about what is the “right” threshold is that there is actually no “right” threshold, nor do we need one. As long as the jury receives the likelihood ratio, it does not matter what threshold the forensic examiner uses.6 The likelihood ratio, which measures the evidentiary worth or probative value of the forensic conclusion, incorporates the stringency or laxity of the threshold used by the examiner. Armed with that additional information, the jury can weigh the forensic conclusion along with everything else presented at trial using the proper burden of proof. The key lesson is that this contextual information matters, and it matters a lot. If forensic examiners present mere conclusions, then we do have to worry about the threshold. Without additional contextual information, the jury must weigh the conclusion in some generic way, and the forensic examiner’s threshold can hijack the proof process. But given a measure of probative value like the likelihood ratio, the jury has enough information to maintain firm control.
The remainder of this Commentary develops these ideas with greater sophistication. Part I explores the fundamental tradeoff between false positives and false negatives in decisionmaking, and observes that every statistical test represents a tradeoff between these two types of error. Part II introduces likelihood ratios, and describes how they account for the false-positive–false-negative tradeoff in a forensic test, freeing the forensic examiner to use whatever threshold he likes. Part III briefly concludes by noting some practical challenges to this solution to the threshold problem.
* Professor of Law, Vanderbilt Law School. Thanks to David Kaye, Paul Edelman, and Alicia Solow-Niederman for helpful conversations and comments, and to Rachel Johnston for research assistance.