CLPEX Chat Board

I was planning on posting this a few weeks ago but then I got busy and I don't think I posted it. I looked and didn't see it anywhere so if I posted this before then I apologize.


http://www.henrytempleman.com/

I did see this website, however, I don't think I quite understood his T-model approach. However, quite amazing to have a latent print examiner state a 23.6% error rate and requiring a 12 point minimum. What was your observation about his statistical model?

Thanks for the link, Michele

Interesting read.

PS, I have only briefly scanned through the website and so this is personal opinion totally uninfluenced by Mr Templeman...

Surely we would all like to have / claim a 0 error rate but I feel this is unrealistic.

Maybe we should consider an "internal error rate" and an "external error rate". Trainees, junior LPE's AND senior LPE's make mistakes for many reasons, not all bad...

The priority and place for 0 error rate, as I see it, is to ensure that errors do not go out in report form.

On "Relevant Populations"

The decision on statistical sufficiency to individualize is, aside from the results of studying the statistical nature of fingerprint characteristics in humans, depends on the relevant population. This relevant population is discussed in the article seems to be determined in terms of residential distance from the scene. (Although another example of a cruise ship population is more reasonable.) It presents examples of suspects apprehended at particular distances and seems to use that to define the relevant fingerprint population.

"It is significant to emphasize that many crimes, especially for crimes against persons in which a suspect is apprehended in a timely manner, there is a relatively small relevant fingerprint population which subsequently requires less amounts of corresponding fingerprint ridge formations in order to establish valid basis for sufficiency to individualize. As a result, numbers of criminal identifications made in law enforcement will increase since fingerprint examiners would require lower minimum thresholds of corresponding ridge formations in two impressions in order to establish positive identification."

Is it really sufficient to know only how many people could have physically been at the incident to determine a relevant population to use in the match probability calculation? I suppose so, but I wonder how often that can be determined. The cruise ship example is one thing. So would be a video record that put the suspect in a location at a known time. But how often can we determine how many people had the opportunity and where we would establish the relevant population in, for instance, a robbery in which we apprehended the suspect 2,000 miles away. Would we be bound to view it in terms of time - how far could anyone who committed the crime travel in the time between the crime and the arrest? After a week, how would we limit the population? Would we have to know how many people boarded international flights and their destinations? Would we assume the actor would not travel far to commit a common burglary? How far is too far? How far is far enough? Do we assume the world population? Is it really valid to leave determination of the relevant population to the happenstance of where the suspect is found at an arbitrary time?

I guess I buy the argument, but I don't see that it will often provide much in the way of limitation. Still, it's not very reasonable to imagine that someone would travel from China to commit a burglary in a small town in the U.S. But it also means that a trier of fact would be bound to consider the likelihood that a defendant with a high probability match would have been at the scene at the time of the crime. And I wonder how that would play out in terms of burden of proof. Shall we imagine for a moment that Mr. Mayfield's was actually a high probability match to the Madrid print, but the Government could not show that he had traveled to Europe. Would he have to overcome a theory that he had traveled covertly, or would he be presumed to have been unlikely to have been there? Any calculation of likelihood of guilt must combine the probability of match with probabilities of ability to commit the crime. (See the analysis of the Sally Clarke case. Neither she nor the Crown can establish the probability that a mother would kill a child and then have another and do it again, nor was the probability of genetic defect known, so neither would be combined with the opined one in 73 million odds of two SIDS deaths in sequence.)

It will be very interesting one day, if these inquiries are matured and applied to limited population cases.