Chapter 9 Further Reading: Forensic DNA Statistics

Annotated and grouped by the book's three citation tiers (see the style bible, §7). Tier 1 are sources we are confident exist and can stand behind. Tier 2 are real ideas and bodies of work attributed honestly without a pinned-down exact citation. Tier 3 are illustrative or constructed material used for teaching. Pursue the Tier-1 reports first; they are the spine of this chapter's argument.

Tier 1 — Verified canonical

  • National Research Council / National Academy of Sciences, Strengthening Forensic Science in the United States: A Path Forward (the "NAS 2009 report"). The field's foundational self-criticism. Its central finding — that nuclear DNA analysis is the one forensic method rigorously shown to connect evidence to a source — is the reason this chapter places DNA statistics at the strong end of the validity spectrum. Read it for why the quantified statistic is what separates DNA from the pattern disciplines.

  • President's Council of Advisors on Science and Technology, Forensic Science in Criminal Courts: Ensuring Scientific Validity of Feature-Comparison Methods (the "PCAST 2016 report"). Essential for §9.3: PCAST's two-part verdict on probabilistic genotyping — foundationally valid for simple mixtures within validated limits, not yet established for complex ones — is quoted throughout this chapter. The report's framing of foundational validity (a measured ability to do what a method claims, with a known error rate) is the yardstick the whole book uses.

  • People v. Collins, Supreme Court of California (1968). The published decision is the primary source for Case Study 1 and the origin of the prosecutor's fallacy in court. Read the opinion itself for the court's dissection of both the fabricated, non-independent probabilities and the fallacious leap from coincidence to guilt.

  • R v. Sally Clark, Court of Appeal (England and Wales), with the Royal Statistical Society's public statement on the case. The primary record for Case Study 2. The RSS statement is a rare instance of a national statistical body publicly correcting the misuse of probability in a criminal trial; it states the transposed-conditional error in plain terms.

  • The Innocence Project (case record and policy materials). For the broader pattern: DNA's power to exclude and exonerate, and the role overstated forensic testimony has played in wrongful convictions. The organization's documented exonerations are the empirical backdrop to why honest communication of a DNA statistic matters.

  • SWGDAM (Scientific Working Group on DNA Analysis Methods) interpretation guidelines, and NIST/OSAC materials on DNA mixture interpretation and probabilistic genotyping. The standards-body sources behind §9.2–9.3: the push toward likelihood-ratio reporting, validation expectations for software, and the definitions of validated limits. Consult the most current published versions for operative guidance.

  • Landmark admissibility decisions referenced across the book — Daubert v. Merrell Dow Pharmaceuticals, Frye v. United States, Kumho Tire v. Carmichael (treated in full in Chapter 5). Relevant here because the source-code and validation questions around probabilistic genotyping (§9.3) are litigated under these gatekeeping standards.

Tier 2 — Attributed, specifics unverified

  • The statistical and legal literature on the "prosecutor's fallacy" and the "defense fallacy." The paired fallacies were named and analyzed in a body of work by statisticians and legal scholars writing on forensic identification evidence; the framing of the transposed conditional as the core error is standard in that literature. Attributed here without pinning a single canonical paper.

  • Forensic statistics texts on the likelihood-ratio framework and evaluative reporting. A well-developed literature (associated with European forensic-statistics traditions and the "logical approach" to evidence evaluation) lays out the LR as the proper measure of evidential weight and the principle that the scientist reports the LR while the court supplies the prior. §9.2 and §9.5 follow this approach; consult a current forensic-statistics textbook for the formal treatment.

  • Validation studies of probabilistic genotyping software, published both by developers and by independent laboratories, reporting performance within stated limits — and independent comparison studies finding that different validated programs can return materially different LRs on the same complex mixture. Both bodies of work inform §9.3; specific studies are attributed in the aggregate rather than individually here.

  • Appellate litigation over source-code access to probabilistic-genotyping programs. Multiple jurisdictions have addressed defense requests to examine proprietary software, with courts reaching differing results. The existence and substance of this split is well attested; the specific docket citations are left for the reader to pull in their own jurisdiction.

  • The English appellate guidance on statistical testimony arising from cases of the 1990s (commonly cited as R v. Deen and R v. Doheny and Adams), instructing experts to state profile rarity and leave the conclusion to the jury, and the R v. Adams line addressing Bayesian reasoning before a jury. Discussed in §9.4–9.5; attributed by case name with the holdings stated in substance.

Tier 3 — Illustrative / constructed (teaching material)

  • All worked numbers in this chapter — the 1-in-43-million RMP, the LRs of 1,000,000 and 2,000,000, the prior-odds examples, the four-locus multiplications in the exercises — are illustrative teaching figures, chosen for round arithmetic, and should never be quoted as values from real cases.

  • Figure 9.1, "What the software is weighing," is a constructed schematic of a two-person mixture at a single locus, not a reproduction of a real electropherogram.

  • The Mill Creek / Marcus Diallo cold case and its mixture interpretation (the LR strongly supporting Roy Keller as a minor contributor, the detective's prosecutor's-fallacy claim) are fictional teaching constructs internal to this book. The persons are invented; any number the reader assigns to the LR is illustrative.

  • The courtroom exchanges modeled in the chapter and exercises are constructed dialogues written to demonstrate honest versus fallacious testimony, not transcripts.

Where to start: if you read only two things, read the PCAST 2016 report's treatment of DNA and probabilistic genotyping, and the People v. Collins opinion. Between them they contain this chapter's whole argument — the legitimate power of the quantified statistic, and the precise way that power is turned into a fallacy.