For the past ten years or so, motions to exclude testimony of “probabilistic genotyping” results have been commonly lodged—and almost always denied. With rare exceptions, appellate courts have held these rulings to be proper (or at least within the trial judge’s discretion). Then came People v. Williams, 35 N.Y.3d 24, 147 N.E.3d 1131, 124 N.Y.S.3d 593 (N.Y. 2020).
In this murder case, New York’s highest court held that the output of one form of probabilistic genotyping software (PGS) was being admitted prematurely, before the scientific community had an adequate chance to evaluate it. But that was two years ago. Two weeks ago, in People v. Wakefield, 2022 N.Y. Slip Op. 02771, 2022 WL 1217463 (N.Y. Apr. 26, 2022), the court returned to the issue of PGS evidence. As in Williams, PGS produced "likelihood ratios" associating the defendant with a murder weapon, but this time the court held that the PGS in question had achieved the general scientific acceptance required to admit scientific evidence in New York.
This posting discusses Williams. A separate posting will consider how Wakefield differs from Williams, and why.
Cadman Williams was accused of a fatal shooting in the Bronx in 2008. The DNA in the case came from a gun hidden in William’s former girlfriend’s apartment. At trial, an expert from the New York City Office of the Chief Medical Examiner (OCME) testified “that it was millions of times more likely that the DNA mixture found on the gun contained contributions from defendant and one unknown, unrelated person, rather than from two unknown, unrelated people.” 35 N.Y.3d at 31. At least, that is how the court understood the testimony. It is not, however, an accurate statement of a likelihood ratio involving the identity of the individual (or individuals) whose DNA is in the sample. Such a likelihood ratio involves only the probability of the DNA data conditional on source hypotheses, not the other war around. (With large enough likelihood ratios, however, the distinction can be academic.) A further issue is the choice of hypotheses. Why not the former girlfriend rather than a random person's DNA?
Nonetheless, the Williams opinion did not need to consider the proper interpretation of the likelihood ratio for a pair of hypotheses or the selection of those hypotheses. The appeal only concerned the general scientific acceptance of the method that the OCME had devised for generating likelihood ratios for minute quantities of DNA. The Court of Appeals held that the trial court erred “in admitting expert testimony with respect to LCN and FST results in the absence of a Frye hearing.” 35 N.Y.3d at 42. LCN stands for “low copy number” and refers to the fact that the OCME laboratory tweaked the usual method for producing data on the personally identifying features of these fragments to obtain some results. FST stands for “Forensic Statistical Tool,” a computer program developed within the OCME to calculate likelihood ratios. Frye refers to Frye v. United States, 293 F. 1013 (D.C. Cir. 1923), a famous case from the District of Columbia that announced the rule that a scientific method had to achieve acceptance within the scientific community before its results can be received as evidence in court.
Although the Court of Appeals referred to one New York trial court opinion finding general acceptance of the OCME method as “questionable,” the Williams court did not hold that the computer output from the FST was necessarily inadmissible. The precise legal error lay in the trial court’s allowing the testimony without first conducting an evidentiary hearing on whether OCME’s methods were generally accepted within the scientific community. The majority opinion, written by Judge Eugene M. Fahey distinguished between the LCN and FST parts of the OCME method and determined that neither could be said to be generally accepted based on the information presented to the trial judge (namely, prior opinions, mostly without evidentiary hearings; scientific publications; internal studies from the OCME laboratory; and a review conducted by a subcommittee of New York’s forensic science commission).
“With respect to the FST issue,” the prosecution had “maintained that such evidence should be admitted without a Frye hearing because ‘numerous articles published in peer-reviewed scientific journals’ supported the point that ‘the analytical software employs well-established principles such as Bayesian statistics and likelihood ratios which are used in many areas of science including forensics, medicine and social sciences.’” 35 N.Y.3d at 35 (note omitted). The prosecution added that “given both the thorough review of the FST by DNA Subcommittee of the New York State Forensic Science Committee [sic] and the exhaustive validation of that tool by OCME, the relevant scientific community had accepted the FST as reliable.” 35 N.Y.3d at 31.
The Court of Appeals was unmoved. It wrote that:
If the analysis was as simple as determining whether FST is comprised of existing mathematical formulas that are individually accepted as generally reliable within the relevant scientific community, then FST evidence probably would be admissible even in the absence of a Frye hearing. [¶] The point remains, however, that FST is a proprietary program exclusively developed and controlled by OCME. The sole developer and the sole user are the same. That is not “an appropriate substitute for the thoughtful exchange of ideas ... envisioned by Frye” (Wesley, 83 N.Y.2d at 441, 611 N.Y.S.2d 97, 633 N.E.2d 451 [Kaye, Ch. J., concurring] ). It is an invitation to bias. .... [That the] tool has ... been vetted and approved by “the distinguished scientists making up the DNA Subcommittee of the New York State Forensic Science Committee” is certainly relevant [but] that insular endorsement is no substitute for the scrutiny of the relevant scientific community. [¶] Indeed, here, defendant was hamstrung in demonstrating the existence of conflicting scientific opinions in order to show the need for Frye review of the FST based on the “black box” nature of that program, but his papers adequately showed that OCME's secretive approach to the FST was inconsistent with quality assurance standards within the relevant scientific community. Those papers also showed that facts adduced in challenges to the FST made in Frye applications in other proceedings suggested that the accuracy calculations of that program may be flawed. .... In short, the FST should be supported by those with no professional interest in its acceptance. Frye demands an objective, unbiased review.
147 N.E.3d at 1141–42.
This language was too much for three of the seven judges. Their concurring opinion written by Chief Judge Janet M. DiFiore balked at the “pejorative view of the ... OCME's ... LCN DNA typing technique and its ... probabilistic genotyping software program ... .” 147 N.E.3d at 1147. To be sure, the concurrence agreed that “the issues ... in this 2014 motion [were] ripe for a Frye hearing”—but only because it appeared that the internal studies did not appear to encompass the small quantity of DNA that was analyzed in this case. According to the concurrence,
The LCN DNA profiles drive the FST analysis, and FST results are only as reliable as the predicate assumptions integrated into the FST software program. The People did not meet their burden of establishing the validity of the empirical data used to fuel the calculations performed by this statistical model, including the manner of accounting for the occurrence of the stochastic effect and allelic dropout in a multiple contributor sample of less than 25 picograms, in a manner sufficient to bypass a Frye hearing. Fundamentally, the combined use of that statistical tool with DNA typing on samples that fell beneath validated thresholds may have impacted the reliability of the results, raising a valid challenge to the admissibility of that evidence in a criminal prosecution.
35 N.Y.3d at 52–53. In other words, the concurring judges seemed to buy the state’s broad argument that there was no need for a Frye hearing to find general acceptance of the OCME system in many cases. But not in this case, where the laboratory pushed beyond what it had validated to its own satisfaction. These judges evinced no concern with the limited outside testing of the FST and expressed no doubts about the conclusions of general acceptance reached in the clear preponderance of trial court rulings on motions to exclude FST likelihood ratios. And the entire court agreed that admission in the case was harmless error because the other evidence against Williams was overwhelming.
In sum, Williams was a case with broad reasoning (by the majority) on a narrow topic—the OCME’s home-grown Forensic Statistics Tool as applied to data from an especially challenging DNA sample (as stressed by the concurrence). What about other FST likelihood ratios with data from larger DNA samples? Other brands of PGS?
One thing was clear. Williams would not be the last word on probabilistic genotyping. Another case, involving an older and more established computer program known as TrueAllele, was on its way to the Court of Appeals. Stay tuned for thoughts on this case.
No comments:
Post a Comment