A woman's purse was snatched on a Los Angeles street by a fleeing man who entered a getaway car driven by a woman. Witnesses could not make a positive in-court identification but described the culprits as an interracial couple: a Black man with distinctive facial hair and a white woman—often described as having a blond ponytail—driving a yellow car. Police eventually apprehended an interracial married couple who shared several of these features. At trial, rather than relying solely on eyewitness identification, the prosecution called a mathematics instructor who assigned numerical probabilities to the occurrence of each observed trait (e.g., a yellow car, a man with certain facial hair, a woman with a blond ponytail, an interracial couple, etc.). Assuming independence among the traits, he multiplied the figures to produce a purported probability—on the order of one in millions—that any randomly selected couple would share all these characteristics. The prosecutor then argued that this vanishingly small probability effectively proved that the defendants were the culprits. The jury convicted on the strength of this demonstration, despite the absence of definitive identification or other conclusive corroboration.
Did the trial court err in admitting and allowing the prosecution to rely on statistical probability testimony, based on unsupported assumptions and independence, to prove the defendants' identity beyond a reasonable doubt?
Expert probability or statistical testimony is inadmissible unless it rests on an adequate evidentiary foundation demonstrating that (1) the underlying frequency data are reliable and derived from the relevant population; (2) the mathematical method used is appropriate to the data; and (3) necessary assumptions—such as independence among variables—are established, not merely asserted. Courts must exclude such evidence when its speculative nature and potential to mislead or unduly prejudice the jury substantially outweigh any probative value. Further, the probability that a randomly selected person (or couple) would match certain traits is not the same as the probability that a particular defendant is guilty given a match; conflating these is error.
Yes. The conviction was reversed because the admission and prosecutorial use of speculative statistical probability evidence, based on unfounded assumptions and an improper independence assumption, constituted prejudicial error.
The court identified multiple, compounding defects in the prosecution's use of mathematics. First, the numerical values assigned to each characteristic (e.g., frequency of yellow cars or men with certain facial hair) were not derived from credible data tied to the relevant community and time period; they were conjectural. Expert testimony grounded in speculation lacks the required foundation for admissibility. Second, even if each marginal probability had been established, the method used—multiplying the probabilities—assumed independence among the characteristics. The prosecution offered no evidence that the traits were independent; indeed, several were plausibly correlated (e.g., facial-hair traits, demographic pairings, and car color preferences). Without proving independence, the product rule cannot validly yield a joint probability. Third, the calculation ignored the definition of the relevant population and the base rate problem. A one-in-millions figure can be highly misleading if applied to a very large population or without specifying the appropriate reference class; even a very small probability of a coincidental match may imply that multiple such matches exist in a large metropolitan area. Fourth, the court condemned the prosecutor's rhetorical move of equating the probability of a random match with the probability of the defendants' innocence—an instance of the transposition-of-the-conditional fallacy. The fact that few random couples would match the described traits does not logically establish that defendants who match must be the perpetrators. Finally, the court emphasized the substantial danger of unfair prejudice: a dramatic, pseudo-scientific number is likely to overwhelm jurors' common sense and substitute for the constitutionally required standard of proof. Because these errors struck at the heart of the State's theory of identity, the court found the error prejudicial and ordered a new trial.
People v. Collins serves as a canonical warning against "trial by mathematics." It frames core evidentiary safeguards for quantitative or probabilistic proof: empirical grounding, methodological transparency, relevance to the correct population, and avoidance of logical fallacies. The case is regularly cited in modern forensic contexts—DNA, fingerprint statistics, toolmark and bite-mark testimony, and likelihood-ratio presentations—to ensure that numbers presented to juries do not exceed their empirical support or distort the burden of proof. For law students, Collins crystallizes how evidentiary reliability (foundation) and Rule 403-type concerns (prejudice and confusion) converge when courts confront expert statistics.
People v. Collins stands as a seminal rejection of convictions built on dazzling but ungrounded mathematics. The court insisted that expert statistics must rest on real data, sound methods, and valid assumptions—and must be presented in a way that does not invite jurors to conflate numerical coincidence with proof beyond a reasonable doubt. In short, numbers cannot bypass the constitutional demand for reliable, properly contextualized evidence of guilt.