IAI resolution 2010-18 (The Detail #462, 26/07/10)

[JohnT]

Re the quote below from 16_Points...

I'm sorry but this just is NOT true. The only reason I have counted points in the past was to find whether or not the identification met a court identification requirement of 12 points, not whether or not I had an identification. Hence the standard of 'suitable for court' and 'unsuitable for court'. Thankfully we no longer require this 'post' filter on our identifications since early 00's and all identifications are suitable for court. I have a real issue with the sugestion that I now somehow lack credibility because I don't (and say I don't) count points. Fingerprint print identifications are based on an agreement of minutae, but they are also based on and agreement of ridge flow, ridge path and 3rd level detail. To ignore this information in the latent print is inviting disaster.

Studies have shown that statistical models have all fallen short in recognising all the variables with a print (Stoney etc.). They therefore potentially understate the real individuality of any given print. My understanding is that Glenn & Co readily acknowledge this in their current model.
--------------------------------------------------------------------------------------------------------------------------------------------------------

Examiners have always counted points. Any examiner that says otherwise lacks credibility. What we have moved away from is basing our conclusions upon point counting. Furthermore, I submit that most latent print identifications are based on minutae . . . not exclusively . . . but you may say that the minutae do all the heavy lifting when it comes to evidentiary weight. And in many cases the identification can be made based on the minutae alone. Therefore, I do not see a minutae driven statistical model as being at odds with the current practice of latent print identification.

[16_Points]

Re the quote below from 16_Points...

I'm sorry but this just is NOT true. The only reason I have counted points in the past was to find whether or not the identification met a court identification requirement of 12 points, not whether or not I had an identification. Hence the standard of 'suitable for court' and 'unsuitable for court'. Thankfully we no longer require this 'post' filter on our identifications since early 00's and all identifications are suitable for court. I have a real issue with the sugestion that I now somehow lack credibility because I don't (and say I don't) count points. Fingerprint print identifications are based on an agreement of minutae, but they are also based on and agreement of ridge flow, ridge path and 3rd level detail. To ignore this information in the latent print is inviting disaster.

Studies have shown that statistical models have all fallen short in recognising all the variables with a print (Stoney etc.). They therefore potentially understate the real individuality of any given print. My understanding is that Glenn & Co readily acknowledge this in their current model.
--------------------------------------------------------------------------------------------------------------------------------------------------------

Examiners have always counted points. Any examiner that says otherwise lacks credibility. What we have moved away from is basing our conclusions upon point counting. Furthermore, I submit that most latent print identifications are based on minutae . . . not exclusively . . . but you may say that the minutae do all the heavy lifting when it comes to evidentiary weight. And in many cases the identification can be made based on the minutae alone. Therefore, I do not see a minutae driven statistical model as being at odds with the current practice of latent print identification.

I didn't think this was controversial. But I stand by my statement. You are counting minutiae. Even if you follow Ashbaugh, you are counting minutiae! It's Quantity and Quality of ridge detail. Minutiae are a subset of level two detail. I do not mean to suggest that IDs are based on minutiae exclusively (see above quote). I thought this was clear. Ridge flow (level one) is important sure . . . but it is insufficient to identify a crime scene mark. So you MUST have some level two detail (i.e. ridge path). It just so happens that minutiae are robust level two features. Hence we look at minutiae in almost every case. Level three is nice. But how often do you find level three detail suitable for comparison in BOTH the crime scene mark AND the exemplar? I rarely see it. (You may also use open ridge fields as part of your level two detail, but they aren't very useful for identification unless they're accompanied by minutiae).

So, yes, you are counting minutiae even though you may not need or want a SPECIFIC number of minutiae. And 2QA aside, many latent print identifications are based in large part (though not EXCLUSIVELY) on minutiae. Hence, a minutiae based statistical model DOES provide an important quantitative measurement of the feature set YOU selected.

[Gerald Clough]

Gerald as a point of interest can you give an example of how you would state your findings in a 'brief of evidence' using a likelihood ratio or percentage figure, which ever you prefer. I am hoping for something that the average jury will hear and understand, obviously something that will not introduce a string of questions from defence. This maybe a big issue as the average fingerprint expert will need to feel comfortable with the statement.

That's kind of the point. I can't state both. On one hand, I can state a conclusion, which is saying I have formed a belief that if the source of the latent could be known to a certainty, it would be found to be this individual. On the other, assuming a known valid likelihood that a randomly selected member of the population possesses a portion of skin that can be impressed in a way that cannot be discriminated from the latent impression, I can state that number. What I cannot do is state in any satisfactory way that the number has anything to do with my belief. I think they remain two independent statements. In making one, I am saying that my expert knowledge and experience and what I can observe leads me to the conclusion that this will not be an erroneous identification on account of two people making two impressions that cannot be discriminated. Again assuming that I have a valid likelihood number, my belief conclusion is not much at hazard so long as that likelihood is extraordinarily small in terms of population. However, if that number suggests a reasonably likelihood of someone else walking around able to make that latent impression, my belief will probably not be taken as so reliable.

A very low likelihood is no problem, even though it's not zero. We've always said we can't absolutely rule out the existence of an alternative source. But the stating of a belief becomes merely tradition. A serologist could just as well state a belief in the source or DNA evidence. When you have a likelihood, does anyone care what you personally believe? But nothing changes all that much. The work of an examination is not formation of the belief. It's in determining the extent to which the latent and the ideally represented skin of an individual agree. It's in that area that all errors not attributable to fraud or paperwork error are made.

Does anything change in how an identification can be challenged. If all you have are traditional conclusions, it's easy enough for one expert to disagree with another about whether the agreement is sufficient for belief. And if you then have numerical likelihood, it's easy enough to argue that the number doesn't justify such high belief. The difference is that, once you have likelihood, that number is all that's worth talking about. And with most identifications worth contesting, the most productive challenge is likely to be to argue interpretations of features in the latent, challenging the identifying expert's necessary presumption about the skin that made the features he relies upon. And that challenge is exactly the same with likelihoods.

If you have valid likelihoods, it seems to me that your presentation differs little from that for DNA or any other such comparison, with the exception that most other analysis that produces statistical results doesn't involve the kind of interpretive judgments made in LPE. For that matter, there's arguably some imprecise likelihood that any given feature in a latent has been correctly interpreted, and you can get into some real convoluted (to the point of nightmare) combined probabilities or different likelihoods when fewer but more certainly interpreted features are used or more but less certain features are used.

It is exactly the subjective nature of interpreting features that leads me to doubt that any method of assigning likelihoods will ever be so consistent that it can be reliably validated to the point of displacing traditional expert conclusions. While it seems certainly possible to assign likelihoods when absolute knowledge of the nature of each feature being considered is known, the numbers probably become wildly variable for most any case where the validity of the identification might be argued, because of the very many variables represented by varying opinions of individual interpretations. In other words, you can build the model, but you can't cleanly apply it to many real cases when the results are likely to fall either side of conventional expert opinion.

[Neville]

mmmmmm

So where does that leave us Gerald?

Seems to me to be between a rock and a hard place. (a Rock like the me's of this world and a hard place as in the statistical model)

Hi 16 points, I believe JohnT when he made this statement, I know John very well he is as straight as an arrow, he does not count points, he often goes to court with no knowledge of how many points there are.

I have to admit John I would still like to know, but that is just me and I certainly understand your point of view, sometimes your integrity can get in the way of an easier life.

I have seen a fingerprint ident done on just about all 3rd level detail, ridge flow was a non event as it was the top of a finger and Minutiae very few. I think 3 or 4 so they do happen. That was about 5 years ago though.

[raybie]

I don't think I count when I go:

Upward pointing ridge ending- two ridge characteristics from the right of the left hand delta of the whorl, three intervening ridges to the right and up there is a ridge ending, one intervening ridge to the right curving upward bifurcation pointing clockwise...follow ridge clockwise there is a ridge ending....ridge ending two intervening ridges from the right of core...formations in sequence.

IMO

[16_Points]

Look folks. Even if you don't go through your comparison saying/thinking "One minutiae . . . two minutiae . . . three minutiae . . . etc." doesn't mean you're not counting points. Perhaps it would be better to say you're "keeping track" of the minutiae as you go . . . is this less controversial? But I submit that all examiners are "counting" even if they are only viscerally aware of the quantity of minutiae that are present (as in a few minutiae, some, or many).

There is nothing wrong with counting minutiae. Having said that, in the beginning, I didn't "count" minutiae either. I looked at ridges with or without minutiae. When I was finished comparing, I either had a lot of ridges in sequence or I didn't. But I was still counting (quantifying?) even if I didn't put a number to it.

And as Neville has pointed out, yes, there are some stunning identifications made using third level detail. But they appear about as often as Haley's Comet (i.e. rare). I would politely suggest that those examiners who do not "count" minutiae should think about reviewing a sample of their old cases. Go back to old IDs and see how many minutiae you used. I think the results might surprise you.

[Gerald Clough]

mmmmmm

So where does that leave us Gerald?

Seems to me to be between a rock and a hard place. (a Rock like the me's of this world and a hard place as in the statistical model)

I don't know that there is a hard place. IF there was a valid way, a method accepted by general agreement, to determine likelihood of a particular impression having been made by a random member of the population, it's not hard, it's just an advancement that changes the way latent prints are used. No harder than a serologist giving up enzymes for DNA.

But I just don't think it's anything we have to worry about. For one thing, I do not believe we even have a beginning on how to code the features in a latent image in a way that matches that coding to every physical skin portion that, considering every possible distortion or other effect, could have made that impression. Let me clarify.

For any latent impression, we might consider that it was plainly impressed, recording the skin as faithfully in all dimensions as a plain record impression or scan. And we also consider that it might have been subject to various degrees of distortion, from mild distortion to distortion so severe that you might pass over the record as not a candidate unless you made a real effort. And, of course, there are other effects that mimic plain impression. When we're working conventionally, we become convinced that we have found the source, and we account for these distortion and other effects. We may be accounting for blank areas, counting ridges flowing in and out of that area and presuming the blank area would be consistent with the record, if only it had been impressed. We may also be dealing with vaguely impressed areas beyond the well impressed area that was sufficient to form our belief. Perhaps we can't confidently interpret that vague area, but we judge that it does not contradict our presumed source. We are examining to determine if our threshold for belief is exceeded, and once it's exceeded, we demand pretty convincing additional evidence before we abandon our belief. Really, we believe that we're working to a lavish threshold and therefore don't expect to ever see sufficient agreement for belief contradicted by further analysis.

But consider how those distortion and other effects that we fairly easily accommodate now operate in using a probability model. The question essentially is, "How many individuals in the population could have made this impression?" It is NOT a question of whether two portions of skin can be differentiated by inspection. It now becomes a mapping problem of the topological sort where teacups and doughnuts are identically coded but easily differentiated visually. In a population of nothing but teacups, hundreds of different kinds, and doughnuts, hundreds of different kinds, topological mapping says you can't tell which member of the population was coded, even though each is unique. To differentiate them, you need more elaborate coding, coding that takes into account the "distortions." In latent print comparison, it is no longer a question of how many individuals have skin that cannot be differentiated. It becomes a question of how many of their impressions cannot be differentiated by coding, mapping, where the coding or mapping ignores all distortions and other effects.

Further (and I think this is a big one), whenever a feature does not present in the latent impression with crystal clarity, I am always making a non-specific statement of likelihood as to the nature of the skin feature I belief it represents. Don't you always form an internal belief about any feature you observe, that it's nature is that of a bifurcation or some other form? And is that belief exactly like your ultimate conclusion. a belief in what reality would be found to be? What likelihood would we assign to the prospect that I am correct? If I'm after precision, would I have to have been subject to a study testing my likelihood of correct interpretation? In feeding a statistical model, how many alternative codings should be assigned to that latent feature, and what likelihood of being correct should be assigned to each coding? Even if we assume that we could agree on the alternate codings and their probabilities, imagine the vast matrix of complexly related likelihoods generated by even a modest number of features that aren't perfectly impressed. How many potential sources, then, each with it's own likelihood, will be presented by the model? And how many different arrays of likelihoods are generated by two or more experts assigning different likelihoods to different possible interpretations of various features? Crazy, huh?

The point is that any flat statement of likelihood in a given case requires so many interpretations and assumptions, including unknowable correct likelihoods of correct interpretations, that any such statement (except in trivially simple cases of wonderfully clear impressions not much subject to contest anyway) is so subject to an array of alternative presentations that it becomes a matter of the likelihood that the likelihood is correct (insert additional receding levels), and it again comes down to subjective expert belief. In other words, the problem with precision is that it only takes a credible challenge to one element to invalidate the result or introduce so many variations that the whole thing gets discounted. And I think we will find that, in actual use as evidence, modeled likelihoods in latent print comparison will be so easily muddled that fact finders will be right back to choosing an expert whose belief they will adopt. And those experts will be forming their beliefs in exactly the same way they do today.

They may, of course, make use of the results of statistical studies. But that will only be for the purpose of bolstering their confidence, but I am not at all sure they won't be on dangerous ground. Working out statistical studies of perfectly skin impressions, essentially studies of the skin itself, is one thing. Applying that to latent analysis is quite another, considering that when you claim precision, you are bound to account for every possible interpretation. Otherwise, you're right back to the expert guessing mode that has served reliably throughout forensic fingerprint history.

So, carry on.

[antonroland]

Thanks for a great thread folks...

It has been a while since I visited for many reasons but what a read to come back to...it has been a bit of a long day but I promise to read the whole thread in the morning and offer my attempt at a sensible response.

[mart]

[quote="16_Points"]Look folks. Even if you don't go through your comparison saying/thinking "One minutiae . . . two minutiae . . . three minutiae . . . etc." doesn't mean you're not counting points. Perhaps it would be better to say you're "keeping track" of the minutiae as you go . . . is this less controversial? But I submit that all examiners are "counting" even if they are only viscerally aware of the quantity of minutiae that are present (as in a few minutiae, some, or many).

No matter how you try to word it ..... I DO NOT COUNT POINTS

[RL Tavernaro]

Mart,
Just curious... What metric do you use to quantify the matching data elements as you do a fingerprint comparison? And do you differentiate between types and/or weights of different data element categories?
Regards, RLT

[Neville]

thanks Gerald, good responce, there are a couple of issues but very minor and as it is go home time and they are nothing to do with this thread, I am not responding.


But I am wondering if the issue of counting boils down to those who take charts to court and those who don't?

There is no point in counting if you are not preparing charts, and as we do not have a set number and there are no charts then why count? But if you are still in the world of charts you have to count which then makes the argument of counting redundant. If you are charting you have to be counting no argument. So if you are counting there must be a minimum number you will use, right?

[antonroland]

Gentlemen

Please help me to understand this... How on earth do you give evidence in court with or without AV aids and without ultimately handing in some form of graphic comparison, be it paper or electronic?

Further to this, how much attention are we giving to the fact that, though the IAI might be represented in many countries, we ultimately serve a very wide variety of very different judiciaries?

Yes, down here in South Africa we might not be involved in all the latest and greatest policy decisions but we are bound by an Adversarial Anglo / Roman-Dutch Criminal Justice system which draws heavily on case law (law of presedent) and in this point of fingerprint identification draws on legal authors of the 1950's who decided that 7 points were it.

I am surely not the only one who might think this is a bit backward but this is our reality and even though we as South African do not rely on minutiae exclusively (as discussed above) they do count for a lot...oh, bad pun...

Be that as it may, our legal system understands numbered points and this will not change soon.

Back to the point...are any of these models available for comment or perusal? How are they being put together and by whom?

I feel we should do everything within our power to take the discipline forward into the next decade and century but I personally feel that models such as these need to be developed, tested, triple-checked, pulled apart and rebuilt from the ground up, found to be valid and good and THEN AND ONLY THEN should a body such as the IAI promote and enforce them.

Answering questions by starting with "The current practise in..." is nothing short of flirting with disaster, creating confusion and dividing the front. Introducing yet another Pandora's box to our discipline is really not necessary...now THAT would be junk science!

[Gerald Clough]

But I am wondering if the issue of counting boils down to those who take charts to court and those who don't?

I really don't think so. Certainly, one can't avoid at least having some sense of about how many details are being actively addressed in an analysis. I suspect that for a lot of examinations, it's really more being aware that there are "many" points, rather than few, without knowing just how many. Of course, that's talking just about being aware of some number, making the distinction that some may have to know how many, in order to satisfy a legal standard, without necessarily implying their conclusion is strictly on account of that number of points. Surely if one could form a belief but could not make it official due to too few points, one could have enough points to satisfy a legal standard while not forming a belief in an identification. I think that's the fallacy of any point standard, since you can construct any number of situations with N points in agreement but with so much poorly impressed area that you couldn't have faith in an identification.

And while making a traditional chart pretty much insures you will know the number of points, the sort of charts being used more and more today, those in which the chart demonstrates ridge flow, rather than congruent details, doesn't necessarily lead to knowledge of the number of details. But no one should be so defensive as to think admitting they incidentally or intentionally count up the details noted implies they concluded ID from that number.

[David Johnson]

Anton

In the US there are many different court systems, each with their own preferences and their own hot button issues. In many courts it not required that a chart be prepared and it is just enough to testify from your notes and report that I identify Joe Blow to the point of entry.
This comes down to the prosecutor and the courts preference on how evidence is presented.
Some examiners like to use PowerPoint or printed charts but this is not done for every case and definitely not done for every latent.

I had a chance to teach down in South Africa and some of the students were of the mindset that it was the seven points that made an ID and there were others who understood that it was not all about the points but would still not feel comfortable calling ID in class that had less than 7 points.
Since your system relies on case law I would argue that if you worked with your prosecutors you could make a case for presenting ID without the seven point requirement. All it takes is another landmark case in a higher court to trump case law. (At least in the states)

The FSS has a model that they are currently marketing and Christophe Champod is also working on a model. There old resolution that was in place would have banned any examiner from using these models and if they did they could have lost their certification. I think the main intent of the resolution was not to endorse any particular model or its use but instead to lift the ban on such technologies as they are being developed. Just as point counting vs holistic took time to catch on and divided the community this to will have its road bumps but do not be so quick to dismiss a stat model as junk science.
The stat models have an advantage over the traditional methods in that they do not need to make the absolute claims that we have been making for so many years and thus they are a little bit easier to validate and test.

One of the arguments for the stat models solo use of points is that even without all of the other information the model can produce very high likely hood ratios. And if this is enough would there be any need for a stat model to look at everything. There is a lot more identifying information in DNA but they only look at selected loci. This makes the task possible and for the most part produced a number that is high enough to establish identity. Why bother looking at 24 loci vs 12 loci if your numbers are all ready astronomical at the 12 loci level. The same would go for a latent model. If a subset of the level 2 detail is enough then why would you need or want level three. 16 points is correct when saying that the points can do most of the heavy lifting in an ID even if you dont count how many of they there are.

[Neville]

ANTON SEE YOUR PRIVATE MESSAGES