CLPEX Chat Board

Pat A. Wertheim wrote:But based on my "training and experience," the odds against any given person in the population having that exact pattern (not just any old generic whorl, but an exact overlay of the one from the lamp) and the several points that are visible would be phenomenally great. Steve Horn may want to comment here because we are not talking about an AFIS hit, but about a suspect whose name came up in the investigation independently of the fingerprint.

Pat is correct that the method of selection of a non-individualised fingerprint match (database search or police suspect) has a big effect on the significance of the match.

This also happens with other types of evidence. Say, for example, the criminal was seen wearing a particular article of clothing. If the type/brand was not common and an example of it was found in the possession of a pre-selected suspect this would be highly significant. On the other hand, if the police locate the shops that sell the brand and track down purchasers of the item, for each person on the purchasing list ownership of the item would not have the same significance of guilt as it would have for the pre-selected suspect. If you think that the significance is the same, you are falling foul of the Prosecutor’s Fallacy. It would be OK for the police to use the list of purchasers as leads to select a suspect, but they must use other, independent evidence for that.

The Prosecutor's Fallacy should be fully understood in police/legal circles by now. I would call it the Texas Sharpshooter Fallacy - was the suspect selected (the target painted) BEFORE the article of clothing was found in his possession or did he become a suspect AS A RESULT of owning the article of clothing (painting the target after firing the shot).

http://www.stevehornsc.pwp.blueyonder.c ... hooter.htm

As long as any change in the use of fingerprint evidence is properly researched (practical simulation-based research using real-world latents, and jury lab simulations, to make sure significance is understood by jurors) I think that probability based evidence could have a number of benefits.

I suspect that we will never reach the level of complexity in a computer algorithm that an LPE uses in performing ACE. Statistical probability is fine as supporting weight to an ident, but the idents that cause all the problems are those that are farthest removed from neat, clean point counting. Who cares if a latent print with 18 level II details has a 1x10 to the 23rd power of accuracy?

And some questions for the attorneys:

Is it not the function of the judge, as gatekeeper, to determine the admissibility of a witness as an expert, subject to defense challenge?

If we then consider virtually all other expert witnesses accepted by the court, why is it that only latent print examiners are not "allowed" to express professional opinions other than absolute idents or non-idents (this excludes medical examiners, toxicology chemists, field investigators, detectives, firearms examiners, footwear examiners, toolmark examiners, hair and fiber techs, blood spatter experts, and essentially every paid defense witness)?

Are LPEs bound by historical convention or by ethics or by law to limit findings to only one of three absolutely positive conclusions?

We are, no doubt, linked to the "absolute certainty" of a positive ident, but what of those who have been forced toward that black or white mentality and pushed the gray line the wrong way? Have those few LPEs caused their own demise by refusing to admit uncertainty?

If we put limited data into a "canned" picklist in a computer and come up with a probability, what does that tell the court? What probability percentage is certainty in latent print comparison? Are we not replacing one issue with yet another, and creating increased complexity beyond the ability of a judge and jury to evaluate?

mdavis wrote:the idents that cause all the problems are those that are farthest removed from neat, clean point counting.

I think that this issue can be divided into two different areas for consideration.

The first is the pure statistical/probability study of different types of fingerprint pattern. The second is how the messy world of crime scene prints connects with this pure world. It is one thing to say that 10 of points of similarity has a certain significance, but if the LPE has misinterpreted the latent and there are, in fact, only 9 then a wrong value will be output from the process. But I see that as less of a problem than the current situation. Saying that the significance is X, but the reality is 60% of X is not as bad as saying that the match provides certainty, but the reality is something less.

Even if you retain “individualisation or nothing” you could learn so much from a proper study of how normal latent print examination connects with the “pure” idea of individualisation. In industry the best companies research processes all the time, looking for ways to improve (“continuous improvement”). For example, one shift might have a greater reject rate than another. Without a proper study you just get opinions and arguments. With a proper study you can find out why. It might be that one shift does a better job and you can find out the reason then get everyone to adopt best practise, or it might be that the quality inspection process is different. In industry it is notoriously difficult to get consistency with visual inspection.

I would want to know, for example, the average number of points of similarity or discrepancy that every examiner finds over all their comparisons for a year. A bit of statistical analysis will determine if differences between people or departments are significant or just the normal “noise” that is to be expected due to different sources. If there is a significant difference, why? Analysis of variation is the bread and butter of research. With this type of study you might be able to detect bad practise before a misidentification occurs. At the very least it would transform discussions on this forum. If, for example, the SCRO finds, on average, more points of similarity and fewer points of discrepancy than other departments that would be a fact and not an opinion.

One thing that has been found in industry is that you must make sure that nobody is ever disciplined as a direct result of research. People must feel safe and see research as something beneficial. Failure to improve might mean a move to a different job.

If, for example, the SCRO finds, on average, more points of similarity and fewer points of discrepancy than other departments that would be a fact and not an opinion.

But does this solve the problem? I think not. Who is the judge of what is the definition of a "point" (your terminology), a "similarity" and a "discrepancy?" Who judges these "points" for the study? Are we not right back at the same old question of who sees what? If we use a committee, how do we totally eliminate bias from those entrusted with evaluating a claimed ident?

Outsider is exactly right when (s)he said one issue is "how the messy world of crime scene prints connects with this pure world" (of statistical/probability study of different types of fingerprint pattern). Once we figure that out, the rest is a piece of cake.......