Elementary, my dear Bayesian

A mathematical technique first developed in Britain in the early 1760s is now a hot favourite in helping forensic scientists tackle the shortcomings in their work.

Called "Bayesian inference" after its inventor, the Reverend Thomas Bayes, the mathematical technique uses information about known events to make deductions about unknown ones. It has slowly gained support since the academic statistician Dennis Lindley, now retired, and Ian Evett, head of the interpretation research group at the Home Office Forensic Science Service in Birmingham, began championing it in the late 1970s.

Further support came last week in a major study of modern methods of interpreting forensic scientific evidence. The book's New Zealand-based authors Bernard Robertson, a senior lecturer in business law at Massey University, and George Vignaux of the Institute of Statistics and Operation at Victoria University of Wellington, say that there is often a failure of communication between lawyers and scientists.

The authors argue for a complete rethink of the scientific basis that underpins forensic science. "It is our contention that logic, probability and inference provide the only language in which the two groups should communicate with each other." Each group needs to have a deep understanding of Bayesian inference in order to think about its own work carefully, they say.

John Buckleton, a scientist at the Home Office Forensic Science Service in Birmingham, explains that Bayesian inference has strong parallels to the way the human mind makes deductions.

Applied to the analysis of criminal evidence, the method can remove a lot of the confusion that litters conventional statistical methods used by forensic scientists, he says.

Dr Buckleton cites a case where the technique could have had a powerful impact - nine rapes of young women in Manchester between May 1987 and October 1988. Andrew Deen was tried and convicted of three of these rapes, mainly based on DNA evidence. In December 1993, an appeal by Deen was heard largely on the basis that the DNA evidence had been misrepresented and may have been misunderstood.

Dr Buckleton says: "This situation was contributed to largely because not all the participants in the trial, especially the lawyers, understood the nature of forensic inference." At the retrial in 1995 Deen pleaded guilty to one offence; the other two were dropped.

Dr Buckleton recently acted as assistant to Bruce Weir, the prosecution's primary statistics expert in the O. J. Simpson trial, a hugely expensive saga that he regards as yet another example of a complete lack of communication between lawyers, scientists and expert witnesses. Mr Simpson is accused of the double murder of his ex-wife Nicole and her friend Ronald Goldman. Both were brutally stabbed and it seems likely that the assailant bled at the scene.

Much of the debating, confusion and time wasting in the case arises from the examination of evidence of mixtures of blood in a car and on a glove. Mr Buckleton says that this is because the participants in the trial do not agree on how evidence should be interpreted.

Despite worrying examples in recent years of the shortcomings of the British criminal justice system, Dr Buckleton says that the O. J. Simpson trial is an indication that the criminal trial process in Britain is still superior to that in the United States. He says: "We just would not tolerate something like the O. J. Simpson trial over here. The prosecution costs alone would not be sustained. The worrying thing is that the trial might set a precedent in the US. It serves to highlight the importance of forensic scientists, lawyers and expert witnesses beginning to use the new methods of analysis as part of a coherent and uniform framework that all parties would be obliged to operate within."

Interpreting Evidence: Evaluating Forensic Science in the Courtroom by Bernard Robertson and G A Vignaux. Published by Wiley, Pounds 24.95.