UPDATES ON CLPEX.com

Updated the Detail Archives.

Updated the Smiley Files with Smiley Bonanza 2007!  We had 7 new smiley submissions this week!  Thanks to Roxanne McLean, Riverside Sheriff's Department for one, and Cynthia Rennie from Toronto Police Department for the other six submissions!!  As always, thanks to the Smiley Czar, Bill Wolz for his efforts, and if you want to send in your smileys, Bill can be reached at: ridgeranger21@yahoo.com

_________________________________________
 

Cynthia Rennie brought us her lecture notes on the appellate process from an ABFDE seminar presentation by Lisa Steele.

Steve Horn explores the use of statistics to support or refute erroneous identifications in combination with crime scenes.  He uses hypothetical scenarios and the Marion Ross murder case as examples, but I encourage you to read this article in its entirety, apply these principles to your work environment and draw out any truths that could apply to a latent print examiner you know.

_________________________________________
Statistics and Misidentifications - Town Fingerprint Project
by Steve Horn

This imaginary exercise is to illustrate a point I have been making for some time about the Shirley McKie perjury allegation.  A fingerprint investigation is normally started because a crime has taken place.  In the McKie case the disputed match between her and latent print Y7 came first, then a new previously unimagined offence was suggested to explain the match.  Statistical theory seems to indicate that the pre-existing crime is an essential part of the safety of fingerprinting. Using a few simple calculations I can show that this is true and demonstrate that there is no certainty at all that McKie deposited Y7.

Town Fingerprint Project

Imagine that a town decided to have all its citizens fingerprinted and the prints held in a database. The town also decided to send fingerprint teams to every house to uncover latent prints.  All the latents would be checked against the database and everyone identified would have to give an explanation for why they had been at the location.

The results would show that the overwhelming majority of identifications would not be disputed because the person identified would be happy to accept that they had deposited the latent print. Only in a tiny proportion of all the identifications would the person say that they have never been at the location.  There are two explanations for a disputed identification.  Either the person was up to no good and is lying about it, or it was a misidentification and an innocent person is being accused of depositing the latent print.

Let's say the town has about 25,000 people and 10,000 households.  In every house 400 latent prints were found. So throughout the town 4 million identifications were made (10,000 x 400). Let's assume that misidentifications run at the rate of 1 per million identifications so there will be 4 misidentifications during the project spread over the 10,000 houses.  Let's also assume that crimes occurred recently in 40 houses and in 20 of these, the criminal left a latent print.

We are now going to look at the chances of finding a misidentification or someone lying at 3 different houses, using different methods to choose the house.

Case 1. A house is chosen at random.

If we pick a house by sticking a pin in a map then we are extremely unlikely to hit on one that contains a latent print where the identification is disputed. After all, most houses don’t get burgled or contain a murder, and misidentifications are very rare. The chances of finding a latent left during a crime are 0.2% (20 in 10000). The chances of stumbling across a misidentified latent are 0.04% (4 in 10000). 

Case 2. A house is chosen because it contains a disputed identification.

If we choose a house because it contains a latent print where the identification was disputed then how can we weigh up the relative strengths of the two explanations (lying or misidentification)?  If the person identified is to be prosecuted that is what the jury will have to do.

A good starting point is the ratio of lies to misidentifications for the town as a whole. 20 latent prints were left by criminals and there are 4 misidentifications, a ratio of 5:1.  If this was put to a jury I don't think they would consider an 83% chance of guilt and 17% chance of innocence to be 'beyond reasonable doubt'.

But using the town's overall ratio is only the starting point. We need to ask if anybody saw the identified person in the house and we need to look for evidence that a crime or offence has actually taken place.  If there is no evidence of a crime and nobody saw the identified person in the house then it is reasonable to say that the balance of probabilities shifts to be in favour of the 'misidentification' explanation.

Case 3. A house is chosen because it contains a crime

We choose one of the 40 houses where a recent crime happened.  We now know for sure that someone has been in the house who will lie and if we are lucky he or she will have left a latent print.  We have said that during 20 of the 40 crimes the criminal left a print so we have a 50% chance of finding it.  The chance of finding a misidentification is the same as a house chosen at random, 0.04%.  So if an identification is disputed in this house it is 1250 (50 / 0.04) times more likely to be a criminal lying than a misidentification.

Conclusion

Even with an error rate of only 1 in a million identifications, it is not safe to propose that a previously unimagined offence has occurred to explain a disputed fingerprint identification.  It would only be safe to do this if the error rate could be proved to be zero. If a crime scene is chosen because a crime has occurred then fingerprinting is reasonably safe with an error rate of 1 per million identifications.

The Marion Ross murder inquiry is an example of case 3 above but the Shirley McKie perjury allegation is an example of case 2.  With no other prior evidence than the fingerprint there is no certainty that Shirley McKie is lying and she should not have been prosecuted for perjury. This conclusion can be reached without looking at the quality of the fingerprint work.  Instead of leading to an allegation, a disputed identification in these circumstances should be seen as a time to question whether best practice has been followed. 

I believe that this exercise shows that the SCRO fingerprint experts at Shirley McKie's perjury trial unknowingly misled the jury because they did not point out that McKie's identification carries no certainty.  If they misled the jury it was because they were not taught the statistical theory that underpins their profession.

----------------------------------------------

Supplementary Points

The assumptions and calculations

The main purpose of this exercise is to show that the method of choosing a crime scene has a large effect on the certainty of an identification. It might be interesting to compare it with real fingerprint work.

10,000 crime scenes represents the whole of fingerprinting activity in an area. I would imagine that in Scotland or the UK we will get up to that number in a few years.  In this exercise the whole town was treated as suspects. In real fingerprint work eliminations account for the vast majority of all the identifications.  In an investigation with only one site a misidentification during elimination will result in a latent being erroneously eliminated but it is not likely to result in a false accusation. In investigations with multiple sites a person on the elimination list would be thought to be innocent at one location but could become a suspect if he or she was identified (or misidentified) at another location.

400 latent prints in each house was chosen because that is the number of latents in the Marion Ross murder house

Of course in an area such as Scotland over a period of years there will be a lot more than 40 crimes.  However, for the purposes of calculating the safety of making accusations of lying when a disputed latent is found during an investigation, and the person identified is not accused of the original crime, it is not the original crimes that we need to estimate, it is the number of other hidden events that would result in the identified person lying that occur in the 10,000 crime scenes.

In this exercise each of the 4 million latents was compared against all 25,000 people in the database so 100 billion (100 thousand million) comparisons were carried out. To achieve only 4 misidentifications our imaginary fingerprint experts have to achieve a random match error of 1 in 25 billion. This is 4 times the population of the world.

Can we calculate the probability that Shirley McKie is lying?

If someone is accused of lying in similar circumstances to Shirley McKie, I think that the probability that they are lying might be estimated as follows:

For any given area and time period divide:

The number of people who have their fingerprints checked during investigations who visit the crime scene and will lie about it and will leave a useable latent but leave no other evidence of their presence and are unseen by anybody but will not be charged with the original crime. This will be a small number and will be impossible to estimate.

by:

The number of misidentifications from all fingerprint activity in the area and time period except for misidentifications during elimination where the error would not result in an accusation.  This will also be a small number and will be impossible to estimate.

This is at best a "balance of probabilities" situation.  Nobody should be accused of lying on the ratio of two small numbers that are impossible to estimate.  In a normal crime-led accusation (case 3) we are on much more solid ground because we would be dividing a big number (because we KNOW that the criminal visited the crime scene the probability of finding his or her latent is high), by a very small number (because we have minimised the chance of misidentification by limiting the investigation to only a few hundred latents).

An on-line version of this paper is available with some more supplementary points:

http://www.stevehornsc.pwp.blueyonder.co.uk/tfp.htm

Stephen Horn
Computer programmer working in the field of statistics for industry
West Lothian
10 February 2007
sz@hornsc.clara.co.uk