Friday, August 10, 2012

The myth of fingerprints

I've been reading Ross Anderson's epic tome, Security Engineering: A Guide to Building Dependable Distributed Systems (2nd edition, 2008, Wiley), and have just gotten into the chapter on biometrics (ch. 15).  Section 15.5.2, on Crime Scene Forensics, points out three major criminal cases where fingerprint matches have been in error, including the Brandon Mayfield case which I wrote about at this blog back in 2007.  Anderson points out that law enforcement agencies have claimed to juries "that forensic results are error-free when FBI proficiency exams have long had an error rate of about one percent, and misleading contextual information can push this up to ten percent or more" (pp. 470-471).  It's probability at work:
Even if the probability of a false match on sixteen points [the UK standard, the U.S. has no minimum] were one in ten billion (10-10) as claimed by police optimists, once many prints are compared against each other, probability theory starts to bite. A system that worked fine in the old days as a crime scene print would be compared manually with the records of a hundred and fifty-seven known local burglars, breaks down once thousands of prints are compared every year with an online database of millions. (p. 471)
One of the other two cases Anderson discusses is that of Scottish policewoman Shirley McKie, who was prosecuted on the basis of a 16-point fingerprint match found at a murder scene and could not find any fingerprint examiner in Britain to defend her.  She found two Americans who testified on her behalf that it was not a match (Anderson shows the crime scene print and her inked print on p. 469; the crime scene print is heavily smudged).  McKie's own fellow officers tried to convince her to give false testimony about her presence at the crime scene, which she refused to do.  She was acquitted, but lost her job and was unable to get reinstated.

The third case Anderson mentions is Stephan Cowans, who was convicted of shooting a police officer after a robbery in 1997.  He was convicted, but argued it was not his fingerprint.  After Cowans was able to get crime scene evidence tested for DNA which was found not to match, a re-examination of the fingerprint also found that there was no match.  So six years after his conviction, he was acquitted on appeal.

Further evidence of the errors which can arise from fingerprint examination comes from two studies by psychologist Itiel Dror which Anderson describes.  In one study, five fingerprint examiners were each shown a pair of prints, allegedly the falsely matched prints from the Mayfield case, and asked to point out the errors.  Three examiners gave explanations for the non-matches, one said that they did, in fact, match, and one was uncertain.  In fact, the pairs of prints were each purported matches by the corresponding examiner from a recent criminal case, so only one of the five was still certain that a match testified to in court was in fact a match upon re-examination with a skeptical mindset.  In a second study, Dror gave each of six experts eight prints that they had matched in previous cases, and four of the six gave inconsistent results.

Anderson points out that belief in the infallibility of fingerprint evidence has the effect of promoting carelessness by examiners, not giving proper critical scrutiny to the method or its assumptions in changing conditions (e.g., the increase in the number of fingerprints to match against in the age of the computer), and increasing the negative consequences of cases of failure.  In the McKie case, Anderson points out, "there appears to have arisen a hierarchical risk-averse culture in which no one wanted to rock the boat, so examiners were predisposed to confirm identifications made by colleagues (especially senior colleagues).  This risk aversion backfired when four of them were tried for perjury." (p. 472)

Itiel Dror's two papers (references from Anderson, p. 923):

IE Dror, D Charlton, AE PĂ©ron, "Contextual information renders experts vulnerable to making erroneous identifications," in Forensic Science International 156 (2006) 74-78

IE Dror, D Charlton, "Why Experts Make Errors," in Journal of Forensic Identification v 56 no 4 (2006) pp 600-616; at http://users.ecs.soton.ac.uk/id/biometrics.html

(Previously, which includes reference to Simon Cole's book on fingerprint evidence which shares the title of this post.)