Fingerprint error and a
McKie-type accusation
If
the police consider a fingerprint identification to be infallible or nearly
infallible, if a fingerprint misidentification does occur one of three things
will happen:
A)
The
misidentified person cannot give an explanation for his or her fingerprint
being found and denies having deposited it. The police assume they are lying
and accuse them of the crime they are investigating.
B)
As
above but it is not feasible to accuse them of the crime under investigation
(it might be a police officer in the investigation team or someone with a cast
iron alibi). The police accuse them of
something else.
C)
The
misidentified person can give an innocent explanation for his or
her fingerprint being found (a
householder might be misidentified from a print in their own home). The police
assume they are telling the truth and wrongly eliminate a crime scene print
from the investigation.
In
the absence of any data, let us assume that these three things will occur at
roughly the same frequency.
If the error rate is not zero then in any geographical area we will eventually get 3 misidentifications (one each of A, B and C we are assuming). If the error rate is very low then many crimes will have been solved because of good fingerprint identifications during this period. Case A where the misidentified person is accused of the crime under investigation will be one among this large number and will be indistinguishable from them. Every miscarriage of justice is a tragedy but if the error rate of identifications is very low, for every innocent person sent to jail we should get many criminals correctly prosecuted.
For
Case B, where the misidentified person can not be accused of the crime under
investigation, we need to ask how many correct accusations and solved crimes
which have no connection with the original crimes will the one
misidentification be hidden among?
Because this number is much lower than the number for case A, the error
rate for Case B accusations will be much higher.
For
case B it is not enough to hypothesise that one incident of wrongdoing might
happen in the time period, that would only lead to a 50:50 chance of an
accusation being right. To safely make an accusation of lying we would have to
be certain that a large number of uninvestigated cases of wrongdoing lie hidden
in crime scenes waiting to be uncovered by fingerprints. In fact, to reach the
same safety level as an accusation of committing the crime under investigation,
there would need to be on average one extra hidden wrongdoing in every crime
scene. A disputed fingerprint identification that requires that we must assume
that an instance of wrongdoing has occurred, and yet there is no evidence to
suggest it, will carry a risk of error orders of magnitude higher than a
disputed identification from a place where we know that a crime
occurred, and the criminal was present. This is the “prior probability” concept
of Bayes Theorem (see below).
One
theoretical explanation for a misidentification is a random match (it is assumed
that this never happens for competent work). If a fingerprint match is what
originates the suspicion that some wrongdoing has occurred then it could come
from any fingerprint identification anywhere in the world at any time. There is
an unlimited opportunity for a crime scene print somewhere to be compared with
an innocent person’s fingerprint which just happens to look extraordinarily
like it (a random match). Because the identified person will deny having
deposited the print a new prosecution case will be originated. On the other hand when an investigation
starts in the normal way after a crime, the crime itself limits the number of
prints in the investigation, so the risk of a random match in any one case will
be orders of magnitude lower. Random matching could not be an explanation for
two misidentifications in the same inquiry (McKie and Asbury) but it might be
an explanation for the McKie misidentification or a partial explanation along
with another factor such as failure to individualise or bias (police officers
are expected to be found in crime scenes).
What might be a random erroneous match for one fingerprint team may not
appear to be a match to another team because they make their judgements
differently.
I
first spotted that if you make the assumption that every fingerprint
identification carries a very small risk of error, then the Shirley McKie case
was an example of the Texas Sharpshooter Fallacy. This is a common
misunderstanding of probability illustrated by a gunslinger taking a quick shot
at the side of a barn. He then paints target rings round the bullet hole and
claims to be a sharpshooter. In a normal case the crime comes first then the
police paint the target by drawing a cordon round the crime scene. They do this
before firing the shot (the fingerprint analysis). It is extremely unlikely
that a misidentification will occur within the few hundred fingerprints that
have a connection with one crime. If we
make an accusation which is not related to the crime under investigation when a
disputed fingerprint identification turns up, we are painting the target after
firing the shot.
Steve
Horn
Computer
Programmer working in the field of statistics for industry
West
Lothian
Last
update 25 October 2007
Fingerprinting Innocentville illustrates the same
point in a short document.
The
Prosecutor’s Fallacy is another route to the same
conclusion.
Bayes Theorem and the Shirley McKie case explains the
theorem as it applies to this case.
More
about the Texas Sharpshooter Fallacy
Some
thoughts about verification.