More to DNA than meets the eye: Interpreting DNA Evidence: Statistical Genetics for Forensic Scientists by I.W. Evett and B.S. Weir

Abstract

Sinauer, 1998. £29.95 pbk (xv + 278 pages)ISBN 0 87893 155 4Forensic scientists testing fingerprints assert that matching prints must be from the same person. Alec Jeffreys' development of ‘DNA fingerprints’ in 1985 suggested that DNA typing of body fluids would allow the identification of their donor with the same degree of certainty. However, in court, ‘DNA fingerprinting’ has been replaced by single-locus profiling with mini- and microsatellites. No longer does a suspect's match with a scene-of-crime DNA sample necessarily indicate that he was its source. The evidence supplied by DNA profiles is probabilistic, and can be quantified as a likelihood ratio—expressing the relative likelihoods of the match between a suspect's DNA and that from a crime scene under the competing hypotheses of his either being or not being its source. Typically, the likelihood ratio is the reciprocal of the probability that a randomly sampled person would match. This probability is easy to calculate, assuming loci are independent, in Hardy–Weinberg frequencies, and that the population frequencies of alleles match those in sample databases. These assumptions are, indeed, often made, but have been increasingly challenged. In their very clear and accurate new text, Evett and Weir present guidelines on how DNA evidence should be interpreted. Their target audience is forensic scientists with biological but not statistical training, but many geneticists will also enjoy this insight into the issues behind the DNA controversies.Unsatisfactorily, in the adversarial system of the Anglo-Saxon countries, statistical expert witnesses, disagreeing about the likelihood ratio, are often offered by both defence and prosecution. This seems likely to favour the defence. Seeing the mysterious ability of experts to produce contrasting conclusions from the same data, the jury may conclude that they cannot understand the evidence, and, if so, an acquittal should follow. Unfortunately, Evett and Weir's illuminating exposition of the principles of DNA interpretation suggests that it will never be possible to produce an unequivocal statement of the likelihood ratio that no legitimate expert could doubt. One reason is that the DNA controversy has coincided with a larger debate within the statistical sciences themselves, between frequentist and Bayesian schools of thought, a debate that colours even the simplest aspects of data interpretation. While frequentists base the concept of probability on the expected frequency of an event in a series of trials, Bayesians typically start with a prior probability distribution, to be subsequently updated in the light of data. This difference impacts even on apparently straightforward issues in DNA interpretation, such as estimation of the population frequency of an allele from a sample. A frequentist would use the maximum likelihood estimate—the sample frequency. A Bayesian would, more typically, assume a flat prior probability distribution for the frequency, such that, for a rare allele, the most probable population frequency remains slightly larger than the sample frequency.The major source of debate in court, however, has been in the issue of whether population substructure prevents the multiplication of match probabilities from different loci. The authors here offer formulae to allow for population substructure, quantified by the coancestry, θ. The need to use such an adjustment, and the appropriate value of θ to use, are again grounds for legitimate disagreement between experts.The situation is thus fraught with problems, but the book, as a practical guide, has an optimistic tone. Its specific instructions to those appearing in court stress avoiding ‘reversing the conditional’ (implying that the likelihood ratio expresses the probability of guilt). The paradox at the heart of the use of complex statistical tools in court, which is the strong doubt whether juries understand them and respond to them appropriately, is also considered, but not resolved.