CLPEX Chat Board

Thanks Pat,

I agree with what you said Pat.

And if you come back here, there are a few of us that would love to see you!!

Hi Pat
There will always be a place set at our table for you too.

If I understand what Steve is saying though isn't he suggesting point counting, now I know that is a dirty word in your language (a bit like that donkey type word I used ) but if we are returning to doing that to please stats why can't we just use the 10 million million to one stat as I have already suggested, and that Steve agrees is excellent, or am I missing something here?

By the way no-one has commented on my question on DNA and Identical twins, can it now be achieved.

Hi Pat,

I agree that the fingerprint department has no need to know the prior likelihood of a suspect’s guilt. My suggestion is that when the LR is reported to the Police it is always reported as a “prior” likelihood ratio (prior to their investigations). This would be a constant reminder that they must not see this value as set in stone (setting it in stone would lead to the Prosecutor’s fallacy).

My memory of correspondence a while back with g. and Cedric Newman is that the LR is the likelihood that the suspect deposited the mark, relative to the likelihood that someone else deposited it. This will be modified to a posterior value by new information just like the likelihood of a suspect’s guilt is modified. The LR and likelihood of guilt are linked together. If someone agrees that they deposited a mark, it becomes much less likely that someone else deposited it. If police investigation finds that an identified person has a perfect alibi and could not possibly be the source of a mark, the likelihood that the mark was deposited by the suspect, relative to someone else, is no longer the value reported by the fingerprint department.

If the fingerprint department report an LR that seems to be a high value but the police cannot find any independent reason to link the person to the crime, we don’t want them to get “stuck” on the high reported LR value. They could close their minds to other lines of investigation. Finding no independent evidence against someone itself modifies the LR value by a large measure.

Of course, this is anticipating a time when LRs are used by the police.

Here is a thought, is there an LR value so high that there is no need for any other evidence against the subject? For example if a mark on one continent is matched to a print on another, can the police be confident that they are from the same skin if the LR is high enough? Or someone eliminated from an inquiry who strongly denies having been in the location of the mark. People have to have a reason for lying and if there is no sign of a reason would it be OK to assume the person is lying if the reported LR was high enough? Perhaps the police should always allow for some human imprecision at the input stage and be sceptical if no connection to a crime can be established independent of the fingerprint.

Neville,

You wrote:

By the way no-one has commented on my question on DNA and Identical twins, can it now be achieved.

in response to my statement that: Identical twins have different DNA, but it is a limitation of current STR tests that cannot differentiate them. (go back and read my original statement). It is believed that all humans do have different DNA, including identical twins; when the egg separates they may have the same DNA, but once the DNA begins to make copies, given estimated rates of mutations, they will have different DNA--which would be observed if one goes and sequences every base pair in the sequence. But with the current tests, which only look at a fraction of the DNA (a limited amount of quantity and quality of information of the DNA), identical twins cannot be differentiated.

This is not without analogy to us, as Pat's threads nicely related. The unique friction ridge skin is without question to everyone. It is the fragmentary, distorted, small area of friction skin recordings in the marks that are deposited at scenes that we examine. And while experts enjoy claiming that they can tell it came from one source and no other to the exclusion of all others (with 100% certainty and a zero error rate), we are seeing that the statistics arising from L1 and L2 detail are not ALWAYS supporting such claims. (always meaning that sometimes the Quantity and Specificity of ridge features is sufficient to make such claims, to X degree of certainty).

Pat,

You asked a great question:

Can LRs be so high as to make the prior inconsequential? In a sense, overpower or overide an incredibly small prior?

The answer is yes. If we accept the smallest prior to be 1 in 7 Billion (everyone on the planet with an equal chance of depositing the mark), then as we discussed in class, yes, certainly LRs that begin to approach 10^15 or so, would put us into a 99.999% certainty range. WHich as Cedric and I theorized and extrapolated from the current models, would certainly indicate that a whole fingerprint or 10 print card would be well above that and allow for that no matter how low the prior is. Incidentally David Kaye's recent paper (the one discussed in class) suggests a similar viewpoint that may be acceptable to staunchest of critics.

g.

Thanks g

With regard to the DNA and twins I thought that was the case, but sometimes we get so busy we miss the obvious. As I said DNA is still in its infancy, I expect great things from it in the future.

I'm not sure how you understand 7 billion but in my day that meant 7 million million which means the figure of 10 million million I quoted before with only 12 point is more than sufficient, many people in the world population are little children below say 10yrs old so we could disregard them anyway say about 30%, couldn't we?

We are only playing with figures, but surely 10 miliion million to one has got to be sufficient to PROVE an identification or is there a need to increase this figure.

Pat A. Wertheim wrote:Can LRs be so high as to make the prior inconsequential? In a sense, overpower or override an incredibly small prior.”

Sorry Pat, I didn’t realise you had already asked this question. I must have missed it. I wouldn’t have repeated it in my last post had I known.

g. wrote:If we accept the smallest prior to be 1 in 7 Billion (everyone on the planet with an equal chance of depositing the mark), then as we discussed in class, yes, certainly LRs that begin to approach 10^15 or so, would put us into a 99.999% certainty range.

That is very impressive and if reported to the police or a jury - with the full authority of science behind it - that would be the end of the case. The suspect is lying, lets go home.

But any statistical model is only as good as its inputs. Surely it possible for optical devices or fingerprint experts to see points of similarity that don’t actually exist. I remember something from the Scottish Parliament inquiry about a Evett and Williams paper where they say that some experts (particularly in the UK) reach an inner conviction about the correctness of an identification long before the full requriement for reporting an identification has been found. They then move on to 'teasing the points out'. In this phase they use features that are clearly visible on the print and try to "see" them on the imperfect mark (and didn’t Peter Swann tell the inquiry he can now find 24 points of similarity between McKie’s thumb print and mark Y7 !).

Some imprecision about the real LR will not have a decisive impact if there is some independent incriminating evidence against the suspect. But if there is absolutely nothing, maybe a prior of 1 in many millions or so, and there is a very high LR, then that LR value had better be right. The Thompson, Taroni and Aitken paper I have previously referred to comes to this conclusion about DNA evidence:

Particularly in cases in which there is little other evidence against the suspect, ignorance of the true probability of error creates a disturbing element of uncertainty about the value of DNA evidence.

Over confidence in these matters is extremely damaging and can wreck lives. Engineers always mistrust theoretical values so they “over-engineer” for safety.

Hi Steve
"Engineers always mistrust theoretical values so they “over-engineer” for safety". Rolls Royce were known to double there engineers values and still they broke down! That didn't mean they were a bad car.Though some petrol heads would argue that point. Human error still exists even in Statistic, I point you back to the OJ simpson case, it doesn't mean statistics are bad. Remember the term DNA fingerprints, it was used to gain acceptance within the courts. Why? because when DNA was first introduced the scientists recognised how reliable fingerprints were. So my point is if .0000000000000000000000000000000001% of the fingerprint experts get an identent wrong .0000000000000000000000000000000000000000000000000000001% of the time does that mean fingerprint identifications are bad?

I hope you realise that the numer of zeros are not a true indication of the fact but is just to make a point.

Fingerprint evidence is already accepted as reliable within the courts and it is up to us to ensure it stays that way by training not software. If you get some idiot claiming to be an expert giving fingerprint identification evidence because some software package tells him there is a 1 million to 1 chance blah blah blah, we are in big trouble. When we still can get latent idents on AFIS that are around 15th or further down the list, with scores of less that half the top scoring candidate you can say there is a long way to go.

Who or what will decide which identification points will be used, software or humans. Now that is the big question as the Coles of this world will have you believe we can't trust humans (is he from Pluto?) and the software on AFIS has proved to not be reliable in this task.

thnx

Neville wrote:Remember the term DNA fingerprints, it was used to gain acceptance within the courts. Why? because when DNA was first introduced the scientists recognised how reliable fingerprints were.

No one with any actual knowledge of either field ever intended to do anything but imply that DNA had great value in discriminating among individuals. But if you wish to invoke what is largely an uninformed media platitude equating DNA and fingerprints, then perhaps you will consider that DNA produces strictly statistical arguments for how many individuals will be indistinguishable through observation of that particular configuration among the finite possible variations in how those few alleles are expressed. Anything less than an infinite number of variations (which does not exist in nature) produces such a result of analysis. It is true for DNA, and it is true for friction ridge skin. Friction ridge skin is not a mystery, an infinite thing the nature of which must be taken on faith and attested by chanting dogma. It is quite easy to work up a reasonably accurate, certainly a functional, number of possible variations in arrangements of details.

So my point is if .0000000000000000000000000000000001% of the fingerprint experts get an identent wrong .0000000000000000000000000000000000000000000000000000001% of the time does that mean fingerprint identifications are bad?

But you simply cannot demonstrate either number. You absolutely cannot say how frequently errors are made, whether frank error or insupportable guessing that happened to be right and therefore no different from error in any sort of scientific view, even the loosest. It is quite impossible to know the errors, because you can't know the reality. You may disagree, but you have to entirely abandon science to do so and resort to guessing. (And, in this instance, since you cannot know the truth, it's not even very informed guessing.)

I hope you realise that the numer of zeros are not a true indication of the fact but is just to make a point.

Fair enough. In fact either number or anything of that magnitude would be about as powerfully persuasive.

Fingerprint evidence is already accepted as reliable within the courts and it is up to us to ensure it stays that way by training not software. If you get some idiot claiming to be an expert giving fingerprint identification evidence because some software package tells him there is a 1 million to 1 chance blah blah blah, we are in big trouble.

So, the DNA expert is not an idiot when stating statistical results, but the latent print examiner would be an idiot to seek to do likewise. Exactly why again are we not in big trouble for DNA experts giving their probablistic conclusions? If you say it is because the latent print examiner doesn't know the numbers for his analysis, again, fair enough. But it is hardly comforting that the fingerprint evidence, which absolutely cannot be anything but in terms of probability, is delivered with nothing but an educated guess.

To be sure, it is a guess that is almost certainly correct almost every time. It is a carefully considered guess. Among all the things a human being can "figure," those guesses are surely among the most reliable, fantastically more reliable than most medical decisions. Speaking strictly for the U.S., I do NOT believe courts consider fingerprint evidence to be what it should be. Courts are practical beasts, readily prepared to accept all sorts of educated guessing. I believe the fairest thing to say is that courts join in accepting fingerprint evidence as very powerful because they have had little choice and because almost no genuine expertise came before them except from within the adherents of the catachism. But that landscape is forever altered and will be definitively altered.

When we still can get latent idents on AFIS that are around 15th or further down the list, with scores of less that half the top scoring candidate you can say there is a long way to go.

If you believe one has anything to do with the other, I am simply at a loss.

Who or what will decide which identification points will be used, software or humans. Now that is the big question as the Coles of this world will have you believe we can't trust humans (is he from Pluto?) and the software on AFIS has proved to not be reliable in this task.

AFIS is very much reliable to its assigned task, which is to select candidates, with absolutely no claim to make identifications. It is bogus to follow logic that hold that because AFIS, a system not designed, intended, not expected to perform identifications does not actual perform identifications, and that AFIS is an application of computers, then therefore, nothing we can learn about fingerprints using computers may be applied to the HUMAN activity of latent print identification.

If by Cole not trusting humans, you mean he cautioned that absolute conclusions of identification, declarations that one individual, with no other possible, was, to a scientific certainty, the source of a latent impression that you did not see that individual make, were not to be trusted, you understood rightly. And he was dead right. That's really what fries a lot of people's grits in the latent print community - that the public revelation came from outside. The reality is that you don't have to have more knowledge of science and fingerprints than can be acquired by any intelligent person with a modest effort to understand this. What Cole didn't know, but what he understood it is possible to know, was what kind of statements can be made from what one observes of fingerprint impressions.

Hi all,
As a divergence from our discussion on validation and testing I'll respond quickly to the issues raised in Minn v Hull. A couple of months ago I submitted an essay to the editor of IAI setting out my concerns as to how the testimony in that case was presented. As a community we should all be aware of alternative thinking so that our preparations are adequate for court no matter in which country we find ourselves. In my last posting I referred to the logic of the testimony given in Minn v Hull that individualisation was no longer considered achievable even when the principle of uniqueness was accepted. It was the logic of their testimony and not the language so I fail to see how my posting could be construed by Glenn as a blatant misrepresentation. I encourage the community to read the transcripts and not just the postings on this website. If I had to choose some of the language then the view that the latent fingerprint was of a quality standard suited to inked impressions is a classic. This case did not involve a fragmentary, distorted or small area of skin transfer so the position of the examiners is by extension challenging the ability of fingerprint science to determine individualisation and unique/exclusive source in both ten prints and latents. Individualisation and uniqueness both have the same property they mean one only, not divisible. The examiners in Minn V Hull used the tactic of avoiding use of individualisation and referring to identification as it gave them room to introduce the concepts of statistical probability (untested) along with the idea of duplication of skin features (unconfirmed). Their basis for refuting exclusive source was the view that since not all fingerprints have been examined then we have not established a defined population. This is where mathematics and the practicallities of science are in conflict. My view is that we have performed enough testing of the principle to satisfy the confidence that no two areas of friction skin are duplicated within the conditions needed for individualisation. Meaning they pass the combined test of 1- similarity of type, 2 - spatial arrangement, 3 -sequential arrangement and 4 -sufficiency. So Pat I hope you have read the transcripts and understood that by using the term Identification in your posting the examiners in Minn v Hull will assume you are supporting their view that individualisation (one only) is to be avoided when presenting fingerprint conclusions. The effect being that the skin of Shirley McKie did not come into contact with the door frame surface in the area of Y7 but given that Minn v Hull express a view that duplication is possible then her left thumb cannot be absolutely excluded since the probability remains. Enjoy your time in Scotland Pat, the Inquiry is warming up to the main event meanwhile back here in 'normality' we will still be discussing how to design an appropriate validation test. Cheers from oz. Les

What's always true is you have to read the court opinions yourself. I read a bunch of them on all sorts of issues, and sometimes you get the feeling that the court itself either didn't get reliable information or information that really went to the issue, or they didn't quite have a handle on their own logic. It's hard enough applying opinions to predict the future and especially hard when you don't have a transcript of the testimony. Many (if not most) opinions are written by a clerk (an attorney working for the judge or panel) and not the judge(s). They, too, have their own peculiarities and biases and styles. And even a perfect understanding of one court's opinion doesn't tell you what another court will do. It does, however, tell you what attorneys will find when they do their legal searches for cases to read, and that will give you some notion of what ideas they might pursue. And, of course, in many matters, it's not over (in the U.S.) until the fat lady sings, and she sits on the panel at the federal circuit court.

And you have to hunt for some of the rulings that are really important in terms of trends. They happen down at the trial court level where the action is a lot more dynamic than in the more ponderous appeals courts who substantially defer whenever appropriate to the trial court.

Les,

You wrote:

...so the position of the examiners is by extension challenging the ability of fingerprint science to determine individualisation and unique/exclusive source in both ten prints and latents.

Not sure how much clearer I can state it, mate: I have no problem with individualizations to the exclusion of all others with whole fingers and 10 prints. And I would agree, one should go to the testimony to see for yourself. For example, try page 150 where I say:

"I think you CAN individualize and let me give you an example"
(where I proceed to give two situations where "individualization to the exclusion of all others" I think is justifiable: a deductive reasoning scenario (e.g. where you can exclude all donors in a closed set, say 5 people trapped in a lifeboat in the middle of the ocean) OR through inference (where there is such an overwhelming amount of information that the possibility of reproduction is essentially nill). This would be akin to DNA statements of statistically being able to exclude the world population many times over. All I can do is just keep repeating what I believe and what I can defend with research and data.

But no matter, we will, again, have to disagree on this one. Ironically we are essentially saying and doing the same thing at the end of the day, so I guess I can't understand the fervor. I wonder if Locard had these sorts of discussions when he and his contempories proposed the probabilistic nature of fingerprint evidence 100 years ago....

As it is, I am off to spend a wonderful summer in Lausanne, Switzerland, where I am sure I will clear my head only to have it further brainwashed there, no doubt. I apologize for the intrusion on this discussion of validation. I for one am excited to see how validation of inferential conclusions based on the discriminating strengths of L1, L2, and L3 can be done without falling prey to using probabilities and statistics. Good luck,

Peace, out.

g.

Okay now that Minn v Hull is put aside lets get back into how we can validate our practice of determining individualisation. I agree with a recent posting introducing the concept of 'over engineering' as a safe guard against theoretical values. What could be called the 'rule of thumb' whereby the decision about a problem is made on the basis of certainty rather than hypothetical calculations not incorporating all the variables within the problem. I also agree with Pat that we need to define our language as well as what fundamental principles we submit for validation testing. The earliest stage of a test design is to decide what needs to be tested, the second part is to formulate how that test can be performed. While not pretending to represent a group of experts my understanding of our fundamental principles is that there are three scientific principles, namely;
1. During normal biological development the palmar and plantar surfaces form individual patterns of friction ridge detail. These pattern arrangements are permanent and reliable templates of the skin surface design apart from the effects of disease or damage.
2. Templates of friction ridge patterns transfer their arrangements of details on contact with suitable surfaces under certain conditions.
3. Transferred friction ridge patterns are able to reveal or exclude the individual donor by combining observations of sufficient and similar details with their spatial and sequential arrangements in the pattern design.

The issue of individualisation as apart from the term identification also needs to be clarified since it is ultimately the purpose of conducting examinations about friction skin transfers.

The transfer of friction ridge details will result in a 'mark' which the expert will need to evaluate as being a human pattern and not something else in nature. The identification of the mark as human is the first decision.
The comparison of the mark with the 'print' is another evaluation to determine that an area of human skin is the source of both patterns (mark and print). The individualisation of the area of source skin is the second decision.
The final decision is to report the name of the human donor whose skin is legally recorded on the print form and is also attributed as responsible for the mark. The identification of the mark to the name concludes the examination. Other conclusions are the exclusion of the person, or the determination that the patterns are not able to be matched due to variations in reproduction.

Once again perhaps Gerald can help us here to clarify which of the fundamental principles requires validation. Regards. les

Les,

Historically speaking, the critics have asked that the methodology be validated. More recently they have acknowledged that all prints are unique but they question whether or not practitioners have the ability to recognize the uniqueness, and when they have the ability (how much data is needed). I think this is asking for validation of your 3rd principle.

To go even further, shouldn’t all premises used to arrive at the conclusion be validated?

I know that some people question the use of level 3 details. Others question the validity of the one discrepancy rule. Maybe the use of scars and creases should be validated. What about combining details from simultaneous impressions to make an ID? And what about capturing images at 1000 ppi, is that really needed or is it a good recommendation?

I'd imagine that validation studies on all these things would be useful.

Some times there's just good justice served... this time in a Hot Pockets wrapper. mmmm... yummy.
oh yeah... and his cell phone he left at the scene too. Brilliant!

Suspected burglar leaves cell
SunHerald.com (registration) - MS, USA
The suspect left his cell phone under the trampoline and his fingerprints on a Hot Pockets wrapper, said Police Capt. Shannon Broom. ...
http://www.sunherald.com/local/story/1410012.html

Here is an idea from industry that could be of use in a validation test. It is an established quality assurance principle that a stream of measurements is always superior to simple counts of products that are out of specification, you can see much more in the data.

How about a study of marked-up charts of latent marks from known sources. You could determine how frequently points of similarity are recorded which do not really exist. Results could be compiled according to years of experience and whether or not positive or negative contextual information is provided. This would reveal far more than just recording erroneous idents.

I suspect that the current 12 /16 point or non-numerical standard is “over-engineered” for most situations. If there is some independent reason to link the suspect to the crime the occasional mark-up error would be unlikely to result in a misident. But the safety margin might be a bit thin when there is nothing other than a fingerprint to link the subject to a crime or there is no crime, just a denial.

I suppose the test would need to include print / mark pairs that have many points of similarity but are not from the same skin. In some people this could invoke the “inner conviction” of having found the source that Evett and Williams describe. The study might reveal that some individuals are susceptible to contextual information while others are not. The study would be far more sensitive than just looking for misidents. A fingerprint examiner who is not conservative enough in reporting points of similarity could go for years, or possibly a whole career, without having a single misident used in court (just like airline pilots who use unsafe practices can go a whole career without crashing a plane).

The study could also be of use with statistical models.