Statistical analysis in DNA-fingerprint matching is a case in point of the need for more science in forensics (Nature 464, 325; 2010)

In 'confirmatory cases', suspects' DNA is found to match that from the crime scene. A serious problem for crime laboratories, however, is how to present the evidentiary value of DNA-profile matches when those matches arise from trawls of the DNA database, sometimes referred to as 'cold hits'. The issue stems from differences in 'frequentist' and Bayesian statistics, and is beyond the ability of most courts to adjudicate.

Statisticians of the frequentist school argue that a trawl involves many independent trials for matching, so that a match from a cold hit within a database of N individuals, each with a match probability P, provides a hit with probability NP. Bayesian statisticians, on the other hand, argue that a match between suspect and crime scene provides a likelihood ratio that is independent of whether the match came from a trawl or not — in which case the evidentiary value of a hit is equal to P in both cold and confirmatory cases.

The differences can be profound. In one case in California (The People v John Puckett), now on appeal, the Bayesian value of 1 in 1 million was allowed, whereas entry of the frequentist value of 1 in 3 was not permitted.

Some panels of experts have recommended the frequentist NP value (including the US National Research Council's Committee on DNA Forensic Science and the US Department of Justice's DNA Advisory Board). Others recommend the Bayesian value of P.

Crime laboratories are frequently unsure of which value to present, or whether to report both and leave it to the attorneys and judges. The proposed US National Institute of Forensic Science could help in solving this kind of problem.