I have prepared a new document which illustrates this point without statistical jargon using a simple numerical example here
I am a computer programmer working in the field of statistics for in industry. I have noticed some unusual aspects of the Shirley McKie case which I think might be significant.
· The apparent match between McKie and latent print Y7 was found during the ‘elimination’ phase of fingerprint work. This is where large numbers of crime scene fingerprint marks are eliminated from inquiries because they match with people who have a legitimate reason to be at the location.
· Only AFTER the match was reported, a new previously unimagined offence was suggested – that Shirley McKie entered the house where Marion Ross was murdered without permission. Usually the crime comes first then the fingerprint work gets under way. Apart from the location of the latent mark, the new offence is not related to the original murder.
My understanding of statistical theory is that these factors greatly increase the probability that Shirley McKie's perjury case was based on an erroneous fingerprint match (misidentification) compared with a normal case.
Normally the first thing we know is that a crime has been committed, then fingerprint work is used to help solve it. In the Shirley McKie case there is no evidence that anybody entered the murder house without permission so we do not even know that an offence was committed. The only reason anyone suspects that Shirley McKie committed perjury is because fingerprint misidentifications are very rare.
But rare events DO happen. The Prosecutor’s Fallacy is to state that the probability of the accused being innocent is the same as the probability of the rare event happening to one person. This is a fallacy because the accused has not been chosen at random.
I imagine that probability in fingerprinting works like this - for every fingerprint comparison (checking one crime scene latent mark against one person’s inked fingerprint) there will be a very small chance that the person will be wrongly identified. This may be due to poor work by fingerprint experts or perhaps even with good fingerprint work the wrong person can be identified by chance. The aim of the fingerprint examiner is to make this Random Match Probability as low as possible (see comment below about 'individualisation').
The probability of error due to random match in any one criminal case is the probability of error in one fingerprint comparison multiplied by the number of comparisons in the inquiry. Increasing the number of comparisons increases the opportunities for examiners to be presented with the fingerprint of an innocent person which is so similar to the latent that an error becomes possible.
Other factors can cause errors such as the judgement of the examiner being altered by bias if he or she is allowed to know anything about the case or the person being checked. So every fingerprint identification carries a potential for error. It will not be possible to calculate a probability value for it and it should be very very small. However small the risk of error is, unless it is zero then as we increase the number of comparisons and the number of identifications in an inquiry, we will increase the chance that the inquiry will contain an erroneous identification.
The probability of error in the McKie case looks to me to be very much higher than a normal case because of two reasons:
Instead of the probabilistic approach, fingerprint experts claim that their work achieves 'individualisation'. This is the belief that a competent fingerprint match identifies a single individual in the whole world with certainty. In other words they say they can always distinguish between the real donor of a latent mark and the person in the world who has the fingerprint which is most similar to it - a Random Match Probability of zero. 'Individualisation' has never been proved or disproved, nor, to my knowledge, has any attempt been made to do so.
To be sure that Shirley McKie is lying, therefore, we have to ask just one question. Were the methods used to identify Shirley McKie from Y7 good enough to be confident of true 'individualisation'? It may be that the SCRO and the independent experts who confirmed the identification used methods which were adequate for most circumstances but were not safe in the particular circumstances of the McKie case.
But can we be sure that any fingerprint method provides true individualisation? Those experts who say that the McKie case is based on good work, and those who say that it was a misidentification both claim that their methods have been empirically verified over many years by a low number of proven misidentifications.
In science all claims and differences of opinion must be resolved with reference to evidence obtained through research. There is practically no research base in fingerprinting nor is there much appetite to get it. Science has discovered the insidious nature of bias and results will not be trusted unless steps have been taken to make bias impossible.
Steve Horn BSc (Electronic Engineering)
Computer Programmer working in the field of statistics for industry
West Lothian
First posted June 2006.
Last update 22 January 2007
I would be happy to hear comments about these observations. sz@hornsc.clara.co.uk
Facts about the Shirley McKie case can be found here or here.
I first noticed that the Shirley McKie case could be an example of the “Texas Sharpshooter Fallacy”. An explanation is here. More about probability here
Link to the section on DNA and "The Prosecutor's Fallacy” in the CPS guidelines on the use of Scientific Evidence:
http://www.cps.gov.uk/legal/section13/chapter_f.html#_Toc7839912
Dr Itiel Dror's research on bias
http://www.ecs.soton.ac.uk/~id/Abs_FSI%20contextual%20influences.html
http://www.ecs.soton.ac.uk/~id/Abs_ACP%20emotions%20&%20fingerprint%20ident.html
Paper by Simon A. Cole Ph.D. on fingerprint error rate
http://www.nlada.org/Defender/forensics/for_lib/Documents/1128572507.37/FinalProof.pdf