CLPEX Chat Board

Les,

We've emailed each other many times on the Hull case and we simply disagree on some fundamental issues. But to be fair to the readers following this thread, I must share my view on your statement:

The recent postings here are very similar to the logic used in [the MN v. Hull] case which basically states that fingerprints are not unique since no human can be said to have been individualised by fingerprint pattern recognition. The position of stating fingerprints are 'likely' to be unique is an agenda in support of presenting evidence of fingerprint examinations expressed in terms of a probability of source attribution.

This statement is a blatant misrepresentation of our position in the Hull case, and I assume was simply a hasty post. But in no way have Bergeron, Neumann, and I ever said, ever testified, ever believed that fingerprints are not unique. We strongly believe fingerprints are unique. We believe there is sound theory to support the formation of unique friction skin. We believe that every human has a unique arrangement of friction ridge skin in the same way that every human (yes, including identical twins) has unique DNA. The only question is: can the test discriminate against all others as a source (with current STR tests, that only look at a fraction of the DNA, no, the test cannot discriminate, while fingerprint comparisons can). And to us, the only way to discriminate against and exclude all others is to examine all relevant sources (impossible for the world population of course) OR estimate and infer the likelihood of someone else sharing a particular set of characterisitics as exhibited in the latent print.

The issue is NOT uniqueness for us...it is "are the characteristics in the latent print sufficiently discriminating to determine the source of the impression".

Your position appears to be, if the expert believes the correspondences in the latent and examplar are sufficient, then by God, it is sufficient and it MUST be that person and no other, because all skin is unique. Absolute certainty and no other possibilities. Our position is expressed differently and we support the probabilistic view that "the more corresponding discriminating features you have, the less likely someone is to share those features....and at some point, a competent expert will make a reasonable decision such that he/she is satisfied that X is the source of the latent print (given the considered relevant population)."

It is important in these changing times, especially in the U.S. where challenges to fingerprint evidence are ever present and we are in a different environment than you, that we keep our facts and positions quite clear and accurate.

Thanks for you consideration.

g.

Thanks Gerald I agree with you.

Hi Neville, how are things in Auckland? Wellington is superb!

I am in favor of the direction Britain is taking regarding fingerprint evidence.

There is nothing new about fingerprints, only in explaining our conclusions more clearly than before.

It is unfortunate that in the US the rules of evidence regarding new sciences has encompassed an established science like fingerprints and other disciplines.

The propaganda as I understand it is that DNA is the gold standard as espoused by Simon Cole. I have seen the man here as a speaker for the innocence project and he chose erroneous identifications such as Madrid and Mckie as examples for deficiencies in the science of fingerprints. It was difficult to take him seriously considering he claims to be some sort of expert in the field of fingerprints. Interesting that he never emphasized that the erroneous identifications were discovered by qualified practitioners. The University statistics department were the sponsors of his lecture.

Les Bush wrote: Validation should be conducted under 'normal' conditions of examination and not with test design imposed 'limits'. A latent print area belongs to a ten print area and both belong to one person so how do we test that?

If I get your meaning correctly, that's an interesting point. I think it has to be seen on two levels. One is the simple validation of the fundamental principle that fiction ridge skin can discriminate and how powerfully it discriminates according to what kind and how much detail can be observed. That presumes that friction ridge skin is pretty consistent in the character of how the details being observed are distributed. I suspect that can be established within reasonable limits. In other words, you can approach that aspect of validation as treating any human friction ridge skin characteristics. I think that's "normal," since it's a very common situation, and it applies even when other levels of detail are available. Now, that's just validation of that principle. That's the first and arguably the most important issue.

But there are other than L2 details commonly used in examination. Ultimately, work will continue toward what I would then not exactly describe as validation, although it really is. It's just that simple L2 validation produces likelihoods that can only apply to the observation of L2 details. Without specific information about how combinations with other features affects likelihoods, statements based on observing those combinations cannot be made and shown to be valid. We really want to know what those combinations mean. I think there's a real likelihood that they may be shown to be powerful and may allow very powerful statements that are not presently being made. They are not being made, because such statements are not part of the historical experience. We would all shrink from declaring that we observed, for instance (and purely for example, as I have no idea if it would be validated), that we observed a particular L1 characteristic and three L2 details in agreement and therefor believe it identifies the source. I think something like that may one day be validated.

In my view, there are two reasons latent print examination has shied away from "probable" conclusions. One is that we wanted conclusions to be plain absolute statements of fact, even if we were forced to back down and admit when asked that it was "possible" someone else was the source. The other is that we had long represented conclusions as absolute truth, to the point where the popular myth accepted that, and we feared being shown to have associated the wrong source by saying it was "likely" or "consistent" (that despicable forensic term). How many times have you felt pretty sure that you were looking at a known suspect's latent and could not quite get over the undefined identification hump? You always knew that latent meant something. You knew it had value. But you couldn't fit using it into the traditional paradigm. The reason for the conflict is simply that the science was inadequate. The question of how the information is used is NOT one for the examiner. But the examiner has long been deciding what should be considered and what should not. That's not tenable, not as science and not as law. It's just preempting judicial gatekeeping and protecting the territory of absolute conclusions.

Now, if by "normal" conditions, you mean we should validate from exactly the sort of typical latent impressions commonly found in real cases, I have to disagree. That unnecessarily inserts all sorts of reliability issues into validation. That aspect of "normal" examinations is not relevant to validation. It is highly relevant to reliability, but the variables are uncontrollable, and we need not admit that uncertainty into the validation study. We don't ignore the fact that real world latents are not ideal. But we look at them in the appropriate way.

"Normal" examinations includes such things a simultaneous impressions and impressions in which continuous impression is not available. What is the meaning of that blank area? You cannot tell me it wouldn't have contained the clear contradiction that would today force you to abandon the identification. For that matter, you can tell me that ANY latent wouldn't contain that contradiction, if only another 3mm of skin had been impressed. The holding that a clear contradiction alone vividly demonstrates that examiners do not know what it takes to imply identification. The condition of no contradiction has no place whatsoever in a validated scientific fingerprint identification scheme. Following proper validation, finding a clear contradiction simply tells us that the particular impression is one that had a known probability of existing.

Gerald Clough wrote:

Girl = Defense Attorney
Horse = Defense Expert (some would say, the horse's a**)
Ravine = Evidence against the defendant
Other Side = Not Guilty verdict

You're good, Gerald -- and from this morning's newspaper, to boot!

I have also shown restraint by not calling on Brother N.O. to help argue the validation issues.

hi Les

Thank you, I can agree with almost all you have said.

The statement " The recent postings here are very similar to the logic used in that case which basically states that fingerprints are not unique since no human can be said to have been individualised by fingerprint pattern recognition" is totally correct and any person with 10 minutes of training will agree, only a CSI watcher would not understand, I understand where you are coming from.

(When I now use the word you I am not meaning you personally) My main issue is that with statistics you have to start with the belief that two people may have matching ridged skin, that's like saying one day the sun may rise in the west. Obviously if by some magical trick that happened human existence would have ceased so who cares or knows which way was east. You are saying though that there is a chance that two people may, then you start arguing before a court the figures which become meaningless, because 1 in 1 zillion chance may happen in the next ident you make, not in the 1 zillionth ident you make, or in the next 1 in 100,000th, it makes no difference what figure it is, it proves nothing except you have agree that it may have happened, contrary to everything you know.

If my understanding of this is correct where would that leave us?

Hi Ra
10c degrees this morning much better, shame I have to work today being Saturday.

Hi g

'We believe that every human has a unique arrangement of friction ridge skin in the same way that every human (yes, including identical twins) has unique DNA.'

So can you now distinguish between identical twins DNA, which rock was I under when that happened. I have to say I believe DNA is still at pre school there is alot more to come from this discipline.

Love the cartoons, who is our heavy weight??

If we claim that fingerprints are unique and it is observable (and yes we can't delve into the past and future and the statistician argues we create a circular argument) the reality is that we can assert this truth. May be fraudulent in terms of statistical argument but the truth is truth! The sun will rise tomorrow and I don't need statistical support to prove this.

Science was based on empirical observations, without it medicine, biology and physics would not be where it is.

The concern is that if I make an identification I require statistical support. If fingerprints were not unique then yes we will require statistical support.

Why on earth should we go backwards and fix what ain't broke? POLITICS! and sciences are not immune to it.

yo Ra

bingo

I think the argument is that math is a pure science, and if that were so I would not be able support an opposing argument using statistics derived from the same raw numbers. Then there are those fabulous number games that mean nothing. All smoke and mirrors or as someone dear to me often says stop blowing smoke up my arrrrrrrr, is that allowable on this site, I will just say #$$@.

hey I didn't type that I wrote #$$@ 'a long eared quadruped of horse family'! strange!

I forgot to mention we can delve in to the past though, as I recall there are a number of very large data bases of fingerprint files going back about 100 years, they were totally backcaptured, the NZ one goes back around 40yrs because it was partially backcaptured. To many sciences thats a long way back to be sampling, and there are those fingerprints from thousands of years ago found on various items from long ago.

Hi Neville,

Good luck to those who want to obtain information from AFIS providers who spent billions creating their systems employing programmers, engineers and mathematicians to create the best biometric algorithms in the world.

The data base and system completely supports our ability to make identifications and that even their best system still can't make identifications just a candidate list that requires human confirmation.

What is obvious is millions of comparisons have been made world wide and no two different people have the same fingerprints. There may be similarities but that is how far it goes. It requires the observational skills of practitioners to determine that a latent and an inked print are identical (came from the same source).

Directly photographing my skin surface detail then taking an inked print and developing a latent by that same finger will result in the same print.

How useful is it to say the chances of duplication is astronomical. My personal opinion is that what was relatively straight forward is now convoluted with language that is attributable to statistics not fingerprints.

I spent five years studying (before qualifying) observing and classifying ridge flow and comparing the ridge flow to determine identity non-identity or insufficiency.

Someone then comes along and says the science has not been validated or tested based on other sciences and then latches on to errors! I am sure that our error rate is far less than the most well established practices such as medicine.

In this discussion, the word “fingerprint” is being used to mean three different things without differentiating between the three meanings. In some discussions, that would not present a problem. In this discussion, it leads to fatal misunderstandings.

Fingerprint = friction ridge skin. Let’s call it “skin.”
Fingerprint = a latent fingerprint. Let’s call it a “mark.”
Fingerprint = an inked fingerprint. Let’s call it a “print.”

Skin, marks, and prints.

Skin, even in the very smallest area, is unique. No statistical model can tell us how unique. It is, quite simply, unique – never reproduced in nature.

Marks and prints are also unique. A smeared mark with no visible features is still unique in its microscopic detail and could only have been left by one area of skin in the world. A smudged mark with only a few visible features is unique in its detail and could only have been left by one area of skin in the world. A clear print with a large number of features is unique in its detail and could only have been left by one area of skin in the world. The question about marks should not be whether they are unique, but whether we have the ability to distinguish that uniqueness. That is where the quality and quantity of the details in a mark become crucial.

The statistical models being considered do not look at skin. To do so would be pointless. No values or likelihood rations or probabilities or error rates could ever be calculated, and why should they? We don’t compare skin to skin.

But when comparing marks to prints, errors can be made. The statistical models being studied are designed to approximate likelihood ratios in comparisons between marks and prints. This has nothing to do with the uniqueness of skin, it has to do with our ability to determine reliably that a mark and a print came from the same area of skin.

Likewise, an error rate has nothing to do with the uniqueness of skin, but with mistakes in identifying marks and prints as coming from the same areas of skin.

Statistical models will never define our science nor make identifications. But statistical models may someday be used to support identifications. A good statistical model is nothing to be afraid of. It will help us prove not only that fingerprint identification is reliable, but just how reliable a specific identification is. A good statistical model will not lessen the value of what we do, nor will it make any identification less reliable. It will simply provide additional scientific validation for what we do.

I agree that errors can be made and there are a number of reasons for that but I believe that training/certification, competency tests and a set of good operating procedures can mitigate the errors. However, errors still have a way of arising due to lack of the above procedures or incompetence or criminal act.

Why is there such an emphasis on providing a likelihood ratio and what will it prove if the identification is indeed erroneous.

Hi Pat, long time no see. Since I attended your course here in good old New Zealand.

Any plans on coming down this way in the future? KInd regards

Pat A. Wertheim wrote: Fingerprint = friction ridge skin. Let’s call it “skin.”
Fingerprint = a latent fingerprint. Let’s call it a “mark.”
Fingerprint = an inked fingerprint. Let’s call it a “print.”

Scientific discussion always benefits from using terms precisely so I think this is a great idea.

Here is another suggestion, every time probabilities or likelihood ratios are discussed with regard to fingerprint patterns you add the word “prior”. This will make it clear that this is not the end of the story. It is the likelihood ratio before the other facts of the case are considered.

I also think that most people will need some guidance on how much these probabilities or likelihood ratios can change depending on the circumstances of the case. You could explain this using the lottery. The prior likelihood of someone winning the lottery by chance is several million to one, but in the context of the award ceremony it is almost certain that the person being handed the cheque DID win by chance (cheating, in collusion with a lottery official, is possible but unlikely if there is no reason to believe it has happened). The winner is at the ceremony, and the prior likelihood of having won by chance was modified, as a result of a selection process.

In a criminal court the accused is in the dock, and a particular ID is led as evidence, as a result of a selection process. If my understanding is right then if there is nothing to link this person to a crime other than that he or she cannot explain why their print was identified from a mark, then after this is taken into account the likelihood ratio that someone else’s skin was the source of the mark, rather than the accused's skin, could be many orders of magnitude greater than the prior likelihood ratio.

One question about likelihood ratios is whose job will it be to account for the possibilty of human error? For example the analyst might think he or she can see 12 points of similarity, but in fact there are only 10. You could calculate every LRs with some sort of "safety margin" built in or this could be left to the user. Again, some sort of guidance would be appropriate.

This may sound complicated but I think it would be better than the current system where a fingerprint ID is assumed to be infallible, or nearly infallible, without any consideration of the circumstances.

Steve -- while the priors may have a bearing on the correctness of the identification from a statistician's point of view, it is crucial that the latent print examiner NOT know of priors that might add bias before the examination. For example, if the latent print came from a counter in the bank where the teller was robbed at gunpoint, if you have the video security footage of the robber, if the teller has already picked him out of a photo lineup, and if he was arrested a block away carrying a bag of cash, and then if you tell the fingerprint examiner all of that before he does the examination, you have introduced the very type of bias we are trying to avoid. It is crucial for all of those priors to be introduced into court where the jury can evaluate all of the factors to determine guilt or innocence, but the fingerprint examination itself should be pristine and free of those considerations.

Raybie -- If New Zealand were only a tank of gas away and I could visit for a day trip or a weekend, I would be down there all the time. Unfortunately, a plane ticket costs just a wee bit more than a tank of gas and the flight time alone takes longer than most weekend trips I've ever been on in my life. But I love New Zealand and hope someday to come on holiday for several weeks. If I were to ever decide to leave the US, New Zealand is right at the #1 position on my list of great places to live.

As for the emphasis on likelihood ratios, probabilities, error rates, etc. -- I agree with you completely that all of those things are unnecessary in the way we perform out jobs and will make not a bit of difference in the way we do comparisons and make identifications. But the old saying is true that your enemies make you stronger. If we consider that our critics are our enemies (a point with which I do not necessarily agree), then if we answer their objections, it makes our science stronger. None of this makes fingerprint identification any more accurate or any more reliable than it has been for a century, nor does it change the fundamental way we do our comparisons. But having the extra validation of statistical studies to back up our claims of accuracy and reliability can only make our evidence stronger in court. Do we need these things to do our work in the lab and reach reliable conclusions? Of course not! But will these things help us present our evidence and support our science? Yes, they will. That is why I believe we need to proceed with this type of research.