Recent News - Articles and Discussion

[Gerald Clough]

I'm bowing out of this discussion. IF a

simple validation study of strictly L2 will indeed scientifically validate the theory that friction ridge skin has features that can identify an individual ....

and the courts buy that, then what? It doesn't matter if I don't buy it for reasons given elsewhere. If you send me to court with a probability percentage, so what? What does the court do with the numbers? What do I do with the numbers? What does it mean? Are we to allow the court to draw the line based on a "validation" study, or is that my job? I don't know what the numbers mean as applied to any given comparison and neither does anyone else. The "strength" of probability would depend on the comparison at hand. I restate my example of the plain arch. Generic "validations" will put such a comparison out of reach. If a L2D "validation" so mis-fits such a case, how many other cases does it mis-fit on both ends and the middle of the spectrum? Gerald seems to think we can do a L2D validation, put it in the bank and go back to business as usual with the blessing of statistics. All I'm saying is that if and when we come up with probability percentages for a given number of L2Ds, we are back to point counting using different semantics. I think this will open up a whole new can of worms because we are using numbers based on half truths and inadequate attention to detail because the complexity of our analog craft is too great to digitize. A simple "validation" study of L2D is a trap, not a lifeline.

The quote is accurate, but the implication that it is all that will be required is neither what I said nor what I think. I used the example of a validation strictly using L2 details and relationships as a plain validation of the principle that features of fingerprints can indeed be used to discriminate among individuals to a high probability. That is the fundamental principle of fingerprint identification. Why would a court or anyone else not "buy" that demonstration? But that hypothetical study specifically shows that one can use L2 detail and their relationships to discriminate. That study does not validate the power of using any other features, such as L1 and L3. Nor does it address the use of L1, L2, or L3 in any possible combination. Again as an example, L1 is pretty well validated by the studies of the frequencies with which they occur in the population. L1 alone is shown to be not very powerful, as it discriminates with a high likelihood of many, many individuals being indistinguishable from only L1 pattern class. Nor does validation of observing the features of a single impression speak to the validity of using simultaneous impressions. That requires more validation study, but I think it's quite doable, although the more complex the combination of observed phenomena, the more work will have to go into the study to put the metrics in terms of what examiners observe and characterize. There is no generic validation study. Each is specific to the particular sorts of data being observed. That's why it's going to take a while and why, too, that it may reveal more aspects of fingerprint identification than we would today be prepared to propose as conclusive.

A court that demanded scientific validation would indeed "buy" an strict L2 validation, but would only apply it to a conclusion reached by observing strictly L2. What will the court do with the numbers? Exactly what they do with the numbers in other validated analysis. The finder of fact considers the weight of the results. Very low likelihoods of another individual matching the data is very compelling and will be considered along with other evidence. If, for instance, it is shown that an individual could not possibly have been the source of physical evidence, to the degree that that showing is reliable, it implies that one of those few other individuals was the source. That's not an outlandish example. If the Mayfield case had reached court without the error having been revealed, it would have been shown likely that someone else contributed the latent, since he was not in Europe at any time that would reasonably allow him to be the source. That's just the way it is with all evidence. And that's exactly what this whole to-do is about, giving the finders of fact appropriate information about evidence.

Does validation lead to "point-counting?" Essentially, yes. No one, including the IAI, ever said point-counting wasn't valid. They said there was no scientific basis for concluding from point-counting - it isn't validated. But no one ever doubted it was valid. (Make that distinction. Not Validated does not mean invalid.) They could not show how many points it took or what it meant to have enough, because no acceptable study showed what it meant to observe N points in agreement. No one ever rationally proposed that there was any number N that made identification absolute. The results of validation studies show exactly what the results of an examination mean. You take the examiner's word for how the features of an impression were interpreted. If you accept that interpretation, through validation, there is no argument about what it means, what it says about the likelihood a given individual was the source. At present, you are asked to take the examiner's interpretations of features and the examiner's belief that the likelihood of another individual being the source is very, very low. If I ask that examiner if it's possible that someone else was the source, the examiner must admit that it is possible. If I ask the examiner if it's possible that one in 10,000 people might have been the source, can the examiner deny that and defend that denial with anything but the examiner's best guess?

There's a strong reaction in the latent print community when you talk about probablistic conclusions. But they have always been probabalistic. We just couldn't say what the probability was and didn't want to, because the conclusion was stated as a belief, and the culture expected infallible fact.

[mdavis]

I really don't have much argument with what has been said except for one thing. It is not acceptable, in my opinion as a trained scientist, to claim meaningful "validation" with L2D while ignoring L1D and L3D, that's all. There are too many exceptions that would render such percentages meaningless, and as such "invalid", or perhaps we could say "valid" but meaningless. There are some impressions that could be "made" with eight L2Ds and others I wouldn't touch. So I go to court with two impressions both with eight L2Ds and I call one of them and I render the other as inconclusive. The court will have a field day with that, both based on identical "verified" probabilities.

I'm not arguing with the concept of validation. I just don't think we can do it in a scientifically meaningful manner with complex analog impressions to support the variations that we encounter everyday on the bench. If you'll read back a few posts, you'll find that I did state that lack of validation does not mean invalid. But if we shortcut this thing and try to sell a validation ignoring L1 and L3, we'll pay dearly for it. We cannot claim "validation" for L2D and then pull in modifying data from L1D and L3D unless we choose to ignore all but L2D and just count points. That's a basic scientific "no-no" of uncontrolled parameters.

A court that demanded scientific validation would indeed "buy" an strict L2 validation, but would only apply it to a conclusion reached by observing strictly L2.

Yeah, just what I've been saying. We can't afford the consequences of dumbing down to L2D alone and we can't use L1D/L3D if they are not validated. Heck, we already have some L1D stats. Why not use that and throw in L2D/L3D for kicks?

If I go into the lab, I must use "validated" procedures and I must be "competency tested" in the use of those procedures. If I choose to modify a dye stain formula, it cannot be used on evidence before it is tested and "validated," even if it is only a change of solvent or carrier. Furthermore, I may not use it myself until I have been "competency tested" in it's use. That's exactly what will happen here. "Validate" L2D, get your probabilities, and then take your impression to court and rely on L1D and L3D to make your case? I think not. It would probably fail ASCLD/ISO to begin with, and the courts would eat you alive.

[Gerald Clough]

I really don't have much argument with what has been said except for one thing. It is not acceptable, in my opinion as a trained scientist, to claim meaningful "validation" with L2D while ignoring L1D and L3D, that's all.

I completely agree. It goes beyond L1, L2, and L3 to other modes, such as using all those in simultaneous impressions and very likely into modes that haven't been seriously considered yet. L2 validation study is an obvious first step and a necessary one, since a great deal of examination uses only L2. It also serves as a basic place to begin and learn some things about how to craft validation studies and how to make them apply to examinations. There are some tricky issues. Even with only L2, there's the issue of defining relationships among L2 details. I don't have an answer to that one that feels right. It's one thing to define details in a way that can be used in examination. It seems to me to be quite another to define relationships. We don't have much precise terminology for that in examinations. It may not be quite as bad as I think it will be. I went through the same thing thinking about how to code details, until I realized you only want data defined well enough that it does not require more examiner interpretation than necessary. I'm looking forward to seeing what people come up with. I'm especially interested in the idea of modeling friction ridge skin so that you can have a arbitrarily large set of data that's statistically identical to the human population. And I fear the next generation of examiners will be getting at least a homeopathic dose of statistics, something I spent my undergraduate work on a math degree avoiding as much as possible. If I hadn't, I'd have a better grasp of how to combine results of different studies.

[Neville]

Seems to me you are handing over your court arguments to a statistician to argue who has the correct figures, rather than keeping it within the fingerprint expertise. What is wrong with saying there is a 10,000,000 chance to 1 of this fingerprint being identical to another person if you compare 12 points (if you "are" point counting) saves a hell of a lot of time and effort. Lets face it we have had computers capable of landing men on the moon for 40yrs, so it aint about computer power is it?

A friend of mine once stated that "10,000,000 to 1 is no chance at all in his books," and that was 20 years ago. No one argued the toss with him over that.

Why do you think that to have this ability will holt the attack, it just moves the goal posts a bit.

[Outsider]

A friend of mine once stated that "10,000,000 to 1 is no chance at all in his books," and that was 20 years ago.

Then your friend was very wrong. If there is a 1 in 10 million chance of this fingerprint being identical to another person (the odds of the evidence given that the suspect is innocent) it does not mean that there is a 1 in 10 million chance that someone else deposited that fingerprint (the odds that the suspect is innocent given the evidence).

It all depends on the context of the identification. If there is nothing else, not even proximity, to link the suspect to the crime then the odds of that person being innocent will be nothing like “no chance at all” or 1 in 10 million.

For example, if the suspect was found by running 10 different crime scene prints against a database of 50 million records, there will be up to half a trillion comparisons. There could be 50 hits with odds of 1 in 10 million. If the perpetrator’s prints are in the database then the first hit will have a 49 in 50 chance of being an innocent person. Of course, the real perpetrator will probably have SOME connection with the crime, even if only by having a criminal record and living nearby.

You (or a statistician or a piece of software) may be factually correct to tell a jury that there is a 1 in 10 million chance of this fingerprint being identical to another person but unless the suspect's attourney or the judge does a very good job in explaining how to relate this statistic to the facts of the case, the jury could to make the same mistake your friend made.

I think even more urgent than validation studies is to ensure that all criminal justice professionals understand the logic that underpins the probative value of forensic evidence.

I put my dice analogy on a web page to explain “transposing the conditional”.
http://www.stevehornsc.pwp.blueyonder.c ... posing.htm

[Les Bush]

Hi all,

I'm beginning to see a separation of logic in how we deal with validation. One side is thinking that validation has a mathematical solution by virtue of the fixed arrangement of features within source skin as the actual truth. The other is looking at a real world situation where reproductions of the source skin are determined on the basis of their qualitative and quantitative factors. Those using maths as their validation technique propose a model incorporating a frame of variables to meet a reliable expectation of repeatability even though the frame is looking at certain features and certain arrangements as determined by the test designer. The other group are looking at the whole range of variables as presented in reproductions and make a determination based on the technique of sufficient, sequential, spatial and similarity factors. This second approach appears too broad for a mathematical model so a compromise is needed if traditional validation methods are to be employed. Thinking outside the square, what if fingerprint identification science isnt meant to be validated using mathematical logic? Validation is a test of reliability, we have a science that is deemed to be reliable but must achieve scientific proof of how reliable. Our experience forms a truth that the accurate application of the technique has produced reliable results. The example of a ten print search in a large database is highly accurate and reliable and verified by the experience of an expert who ultimately uses all detail to make a reportable conclusion. The same applies to a good quality latent print. The one thing our practice hasnt been doing is confirming a relationship between these reproductions and the source skin. Our confidence is that we have reliably proved the source skin pattern so bringing the two ends together would complete the picture. If a control group of a reasonable size sample of ten prints and latents were analysed through a large database, eliminating all candidates proposed by the AFIS algorithm and then confirming individualisation to a source skin donor not known to the AFIS would that constitute evidence suitable for a validation test? If we ran that type of test across the globe and pulled the data results together would that be sufficient to meet scientific proof of reliability? This sort of validation would be conclusion based (right/wrong) so it is important that a technology is used to capture the decisions made in the examinations. Who would peer review such a validation test and confirm its reliability? Fingerprint expertise is a qualification gained by study and practice until errors become an unacceptable outcome. Would we as a community be prepared to share our validation test results with another country not involved in the test so that independent peer review can occur? Cheers from oz. Les

[Gerald Clough]

Hi all,
Thinking outside the square, what if fingerprint identification science isn't meant to be validated using mathematical logic? Validation is a test of reliability, we have a science that is deemed to be reliable but must achieve scientific proof of how reliable.

I just quoted that bit because it relates to my take on this. But you point out genuine issues in the whole post.

My take is that validation and reliability of practical identification are two very distinct issues. One of the objections, often the first raised by critics, is that there simply has been no scientific study to prove the fundamental principle. That principle is simple - That friction ridge skin has observable features that distinguish one individual from another. We take that for granted from a lot of experience, but that is not acceptable science. No one doubts it's true, but in the broad scientific view, not being much doubted doesn't count. While this strikes some as a silly concern and a lot of work to confirm what everyone "knows," it is the key first need to place fingerprint identification on solid scientific ground.

One may argue that it's silly because we already know the outcome. We do know the outcome with regard to answering the question yes or no. But we also know that you can't make the simple statement of principle, because it must always be conditioned. You can't say that just any portion of skin is distinct in its practically observable and well-defined characteristics. In fact, you can't say and certainly can't prove that any portion of skin is unique. You can only say, once you've done scientifically rigorous study, that some quantity of observed data produces a particular likelihood of it representing a particular individual. Like anything else you can count, there is some finite number of variations. Therefore, you can never say there is one and only one portion of skin that can be distinguished from all others.

When you know that likelihood is for particular accumulations of observable data, you then have the basis for forensic use of that skin. (We are really approaching this as if we had just discovered fingerprints, except that we have the advantage of avoiding many false starts because we have considerable experience making clinical use of it.) When I say "observable," I do not mean we are testing the products of observation. I mean that the data must, of course, be limited to what may be observed. It just defines what characteristics of skin we are talking about, casting out such characteristics as genetic and other biological markers and other things that cannot be visually observed. Nothing in the validation of this principle really has anything to do with latent print examination, per se. Except that, because we are interested in proving a principle that we wish to apply in examination, we wish to do the validation studies in the same terms used in examination. At this stage, we are proceeding as if we have no visual sensory apparatus and are just seeing if it's worth developing a way to get the data.

We have to agree on what we mean by validation and by reliability. I think it most clearly explains the issues by referring to validation as strictly addressing how distinct friction skin may be and by referring to reliability as how well examiners collect data from actual latent impressions. Validation is strictly mathematical, because it asks questions about absolute reality, the skin itself, unfiltered by how it is observed. There is no issue of reliability in validation. (That isn't exactly true, because one may indeed criticize one or another validation study on the basis of how accurately it's data represents actual skin, but that is a very different thing from examiner reliability.) In reality, while there will be much discussion and refinement of validation studies, they will be accepted as having clear meaning. Various studies reliability will more address best practices and maybe how suspicious one might appropriately be of a given examination.

I like that last part of the portion of your post that I quoted. I take it to mean that examination is today substantially reliable. (But NOT validated, which does not imply Invalid.I agree, but with the caveat that the reliability of examination depends on doing the analysis properly. That seems to me to be consistent with the revealed errors, where mostly the error has been in how the examiner interpreted the impression. I think this is very important for us to keep in mind. There are more things we may do to make observations better and to protect the end-product against the inevitable errors in interpretation, but when examiners are fully capable of making reliable observations. We have a lot more to learn about insuring it's done properly and under the proper conditions, but that's just getting better and better, a right and proper thing in any field.

So, two issues.

(1) How reliably do we interpret impressions to determine the characteristics of the skin that made the impression. This one is Reliability. It does indeed involve all sorts of variables, many of them not measurable. We can't make solid determinations of how consistently any one examiner interprets under given conditions. We can't know much about any one examiner's state of mind and general biases. We can to a considerable degree study the effects of specific exposures to potential biases. So we can only do some very general and far from all-inclusive studies about reliability that tells us something about how examiners as a whole perform. Ultimately, we are saved from the inability to give each examiner an absolute reliability rating, because we can always reexamine any case. In fact, it's only prudent to reexamine every case.

(2) Once we have made that observation of an impression and interpretation of characteristics, what does it mean? When we know what the data means, so long as one accepts that the validation was properly done, there's no reliability issue at all beyond the question of interpretation above. If you take the data collected in examination to accurately reflect the skin the impression represents, you know what it means.

So, latent print examination as a whole, the human task of observing features of a skin impression to determine to a very degree of probability that one individual is the source, is not amenable to overall scientific validation, and no one pretends that is can be. Our theory is NOT that every latent print examiner, every time, invariably makes an absolutely accurate interpretation of an impression. That would be a silly theory. And, if it weren't simply a silly theory, it's been disproved. But we can prove the theory that we will apply to assign meaning to what we observe, because it has nothing to do with how the data is collected and addresses absolute reality. We should not have to argue the validity of the theory that, according to the amount of available information about skin, we can say something conclusive about what it means. We will always, in at least some cases, have to argue that our interpretations made in characterizing the source skin from observing impressions are accurate. We do that now. The OAG argued very effectively that the interpretations in Mayfield were wrong.

Again, to make a distinction that is very important, the interpretation of characteristics from an impression is clinical, a matter of knowledge, ability, and experience, just as physician makes judgments from a set of observations and chooses an interpretation of them in the form of a diagnosis. Another physician may disagree. Someone who needs to choose between the two will make that judgment by judging the knowledge, demonstrated ability, and experience of the physicians. Each physician believes his diagnosis is the most likely correct. Neither is being sloppy or dishonest. You and I can look at an impressed feature and disagree as to what reality it represents. If a third part can't choose which of us to rely upon, the question then becomes what the other, uncontested interpretations mean. Validation has nothing to do with knowledge, ability, of experience. It is reality.

Latent print examination need not and cannot be "validated." Only the applicable theory can be validated, proven. And that proof will disavow absolute identification, as we have disavowed absolute identification, every time we confronted the question, "Is it possible someone else could have been the source?" It is not possible to claim to be a scientist or to be wholly scientific and at the same time say, "I might be wrong, but I believe." We can, though, as all science does, say, "If my observations are correct, and I have made them as accurately as I know how, I know what they mean and can prove it." That is the correct representation of what we do, and it's the one the finders of fact in the criminal justice system deserve.


One more point.

If a control group of a reasonable size sample of ten prints and latents were analysed through a large database, eliminating all candidates proposed by the AFIS algorithm and then confirming individualization to a source skin donor not known to the AFIS would that constitute evidence suitable for a validation test?

You would have proven that the data coded for AFIS is capable of selecting a number of individuals whose data collected from their impressions cannot be distinguished by AFIS. You would also have shown that an examiner can select the correct source from among those individuals. The objections are:

This validates AFIS as being capable of generating a set of individuals among whom it cannot distinguish and that if you place that particular source in the database, that particular source will be among the candidates. It is limited to one particular impression.

The AFIS data is not the same data that an examiner would accumulate and characterize in an examination. That preselection from different data spoils any use as validation that can apply to what examiners do.

It does not define the likelihoods of indistinguishable individuals according to different quantities of different characteristics. If it doesn't do that, it doesn't tell us what particular accumulations of data in examinations means. It just shows it is possible to pick out the source. And we already know it's possible, just as we know it's possible for the doctor to pick the right diagnosis, if we presume his observations to be accurate. We can't know the probability that the doctor is right, even if we presume his observations to be accurate. His testimony on diagnosis will largely be strictly clinical, not provable (except perhaps at autopsy, which we would prefer to avoid). We can know the probability that, assuming the observations and interpretations are right, the identification is right. We get that from coding a very large dataset representing friction ridge skin and measuring how many members of that set are indistinguishable for each quantity of particular data.

[Les Bush]

Thanks again Gerald,

Using the term reliability as the purpose of validation was intended to address the principle of uniqueness in patterns of fingerprint detail. The skin is formed by a combination of conditions resulting in a permanent and unique arrangement. Proof of this principle can be observed by examination of the patterns. Fingerprint science relies on that principle for declarations of source attribution to the exclusion of all others. The example I gave of taking a sample of ten prints and latent prints from known skin sources was meant to provide the conditions by which the results of fingerprint examinations could be assessed as valid or not. By undertaking the test we subject the principle to a situation of proof of reliability since we know the correct answer. If the test fails then the principle is not reliable. What would be obvious is where the particular statistic has shown the principle to be false. To date for all the examinations conducted globally there has not been a single example of the principle being shown to be false. The AFIS operates as an analytical filter that checks a variety of feature types, locations and arrangements so in effect it does a very effective holistic scan of all the details present in a fingerprint pattern. When the AFIS produces a list of candidates it does so on the basis of probability and then possibility using a grading system with the highest confidence ranked first. The AFIS algorithm has been validated to repeat this analysis product. Following AFIS the expert then performs the 'clinical' evaluation leading to a conclusion. For the purpose of scientific acceptance it appears that conclusion is reliable only when the principle has been validated. The Law of Large numbers supports the confidence of experts in maintaining their approach to examinations and conclusions. Over the time period we have been conducting comparisons the focus has primarily been the distal phalanges so in effective we have used a defined area to test the reliability of the principle. The only sticking point appears to be that our collective knowledge is not deemed scientific since we failed to prove the result with a comparison to the source skin.

All sciences are conditional so fingerprint identification is no different to any of the mainstream sciences such as Maths. A simple calculation such as 1+1=2 is a valid answer but it is conditional. A long division that has no meaningful end point beyond a certain number of significant figures is conditional on the determination of what is a valid result. All statisitical data is interpreted on conditions framed for the analysis. Our fingerprint results are similarly issued based on the conditions of the examination. Medical science is valid when the symptoms of one condition are correctly interpreted and this too doesnt require mathematics. Regards. les

[Gerald Clough]

Thanks again Gerald,

Using the term reliability as the purpose of validation was intended to address the principle of uniqueness in patterns of fingerprint detail. The skin is formed by a combination of conditions resulting in a permanent and unique arrangement. Proof of this principle can be observed by examination of the patterns. Fingerprint science relies on that principle for declarations of source attribution to the exclusion of all others.

We might have to differ here. No examination of fingerprints can absolutely exclude all but a single source. That is simply not a tenable conclusion. We believe the likelihood of the evidentiary latent being indistinguishable from someone other than the proposed source is very, very small, but there is no logically valid basis for declaring that it is absolutely unique and therefore can only have been made by a particular individual. My logic is:

Since the area of friction ridge skin found on an individual is finite, and because there is a maximum density of detail beyond which individual details couldn't even be identified (and we have a pretty good idea of the approximate average density, even without rigorous study), there is some unknown but finite upper limit to the number of details that could be found on the greatest possible area of friction ridge skin that could be found on an individual. Therefore, since that number is finite, there is a finite number of ways to arrange all possible details.

The same is true for any portion of skin less than the largest possible. Again, a finite number and a finite and smaller number of ways to arrange those details. It is therefore possible for two portions of skin from two distinct individuals to be indistinguishable on the basis of the observed details and their arrangement. That does not admit to any statement of fact that the impression of the portion of skin described must have been made by one individual, to the exclusion of all others.

Now, that is the science. Here is my problem with an examiner stating a unique source strictly from observing impressions. A statement of unique source can only be a belief, a statement of faith. It is a statement that the examiner is presuming that no other possible source had access to the place or object upon which the impression was found. That information can only come from evidence other than the fingerprint evidence. That statement leaps from what fingerprint examination can say to what a finder of fact in possession of all the evidence can say they believe most likely to be the case. That is grossly improper and absurd.

The classic question is, "Is it possible that someone else could have contributed the impression?" If one answers Yes, it invalidates any statement of absolute uniqueness. If one answers NO, one cannot base the answer on anything but a belief, a guess, and one that is denied by the physical nature of skin.

Now, it is not unreasonable for an examiner to hold a belief that one identified person is the source. The clinical experience of all the examiners who have ever worked suggests that the belief is very likely to be correct. But that is absolutely not a proof and will not stand up to logical and scientific scrutiny. What is changing is the recognition that statements of fact that purport to be scientifically valid must be grounded in proper science, not clinical experience. Clinical experience can only say something is the best guess possible from what can be observed. It may be powerful. It may be persuasive. But it cannot be misrepresented as absolute knowledge of the truth.

The Law of Large numbers supports the confidence of experts in maintaining their approach to examinations and conclusions. Over the time period we have been conducting comparisons the focus has primarily been the distal phalanges so in effective we have used a defined area to test the reliability of the principle. The only sticking point appears to be that our collective knowledge is not deemed scientific since we failed to prove the result with a comparison to the source skin.

I think that's a fair statement. No one doubts the principle, but not doubting is not proving.

All sciences are conditional so fingerprint identification is no different to any of the mainstream sciences such as Maths. A simple calculation such as 1+1=2 is a valid answer but it is conditional. A long division that has no meaningful end point beyond a certain number of significant figures is conditional on the determination of what is a valid result. All statisitical data is interpreted on conditions framed for the analysis. Our fingerprint results are similarly issued based on the conditions of the examination. Medical science is valid when the symptoms of one condition are correctly interpreted and this too doesnt require mathematics.

True. 1+1 may equal 01 or 1, depending on what the codes mean. Dividing some integers produce non-terminating fractions that cannot be expressed as a decimal. And the statement of fact that is the product of an examination is conditioned on the presumed accuracy of the examiner's interpretation of the impression. But it must not be conditioned on an unproven theory, unless one also conditions it by admitting that the theory is unproven and might be false. And any theory that holds that no two portions of skin can produce indistinguishable impressions is an improperly constructed theory, because it is inconsistent with what we know of skin. And a statement that there is some accumulation of details and their arrangement that can absolutely identify the source is logically false. One is therefore limited to statements of probability. That's really where we have always been. Statements of identification that are really statements of belief, with the condition that we admit a small possibility that we're wrong. The crunch is that belief is coming to be not acceptable, both because belief in a claim of valid science is unacceptable and because the popular beliefs about fingerprint identification are powerful and tend to lend the weight of scientific fact to an educated guess.

Medical science is validated. Medical art is not. Determination that a particular genetic anomaly associated with a disease that it always produces is scientific analysis, and many such determinations have been validated by rigorous study. Even then, they cannot make totally absolute statements, because the dataset is finite. But many, many diagnoses are made through the practice of medical art, the application of clinical experience. Finding that the diagnosis was correct does not validate the practice. It merely shows that the guess was right in that one case. One might indeed set out to validate the theory that a particular set of signs and symptoms are diagnostic of a particular disease or condition. It would be done by study of a large number of cases. (Proof of the condition might be difficult and might only be available at autopsy.) That validation would never confirm a theory that those signs and symptoms absolutely always mark a particular condition. Sometimes, the condition will not be present when those signs and symptoms are observed. The validation will produce a probability that the condition is present. The similarity between medical practice and latent print examination is that both depend on how accurately the data can be observed and interpreted. The difference between medical practice and latent print examination is that physicians must act. They must make their best guess. They cannot reserve action for only those cases in which they know the probability of correct diagnosis. They must reach a clinical belief that the probability is high, or at least higher than other probabilities, and treat accordingly. They do the same thing in avoiding harm. They know the probability that the patient will die during a surgical procedure but judge that probability acceptable low compared to the probability of mortality or morbidity if they don't cut. They don't do this because they like not knowing the true probabilities. They do it because they don't know the probabilities.

The demand now is for fingerprint identification to make as much of the practice scientific as possible and to do as well as we can with the clinical aspects. Interpretation will always be clinical. We can't formulate any testable theory that will predict how likely it is that a given examiner in a given case interpreted correctly. We can't formulate any testable theory that will predict which internal biases or tolerances may be present for any examiner. That's clinical, and just as with medical diagnosis, if another doctor doesn't agree from their own clinical experience, they can argue the toss. But that dissenter can't argue that the validated theory that this genetic anomaly produces that condition with given probability. That part is validated scientific fact. In fingerprint identification we can discover highly applicable knowledge about what particular accumulations of data mean in terms of likelihood. If we could not do that, if it were not possible, there would be no effort being mounted to accomplish the undoable. The argument would be that we must forever be limited to description of similarities and maybe clearly state that our belief that the identification is highly likely is purely personal belief with which other may differ. (That is pretty much what other clinical practitioners of unvalidated practice do. Psychologists, for example, who commonly are opposed by psychologists with a different opinion. They are stuck in that status, because they cannot formulate and validate fundamental theories of physical reality to apply. They get to testify from that position because there's NO other possible position for them, and the courts believe their testimony has value.)

I fully understand the tendency to be peeved with attacks on the beliefs that grow out of a vast body of experience. It galls some especially that it is being sought in the face of the virtually universal acceptance, inside and outside of the LPE community, that fingerprint impressions of quantity and quality of detail very often found in real cases are extremely powerful identifiers. But it's not acceptable to say that we know we can speak with greater precision and simply choose not to pursue that capability. But I also understand the dismay and outrage of those who know such precision is possible and who have been rebuffed for years.

Seeing both sides. Can't help it. It's a curse. Probably raised wrong.

[Neville]

May I please correct my own error? I know every fingerprint expert out there has no doubt noticed I left a few zeros out, hopefully they have recovered by now, it was not 10,000,000 it was 10,000,000,000,000. But then even math experts can get it wrong, remember OJs DNA trial. Maybe its our Auckland water!

Anyway let me know when Cummings and Midlow and Goulton have been proved idiots of the scientific world. Didn't Darwin even have a very quick look at this science as well, gee wiz maybe I am not a monkey's son or was that uncle, after all and the world may still be flat, no that's right science has proved its round. Keep up the good work Les I can almost understand your thinking.

Not sure that AFIS is really that great otherwise we wouldn't be getting hits so far down the list would we, they should always be the first one on the list with a big jump in the score to the second. I suppose it would be nice to push a button that would produce mathematical proof that an Ident is a fact on a particular Ident, the prosecutor could do this in the court and the defence would bow to this undeniable fact, and to quote a local beer advert "yeah right"!

Sorry I have been around courts too long, perhaps even too long in the police; I have become a cynic haven't I?

[Outsider]

it was not 10,000,000 it was 10,000,000,000,000.

That is a pretty impressive statistic but entirely theoretical. How many of these zeros do we knock off to account for human frailty and the messy real world? This could make a difference to the likelihood of guilt in a case if the supect is completely unconnected to a crime in the absence of the fingerprint.

In industry visual inspection is notoriously difficult to implement in a consistent way. Inspection criteria can slowly change over time (unless something is done to prevent this).

[raybie]

I am in favour of the Cartesian approach. I observe and look at what the moderates say rather than the extremist views on both sides.

I suppose that if I had the time and my employer had the resources I would spend more time researching. At the moment I am what I am, a practitioner not a research scientist. I attend crime scenes and mortuaries; I preserve latent prints through various means including photography; I analyse and make comparisons of latent prints using my magnifying glass; I search prints on AFIS and attend court to give evidence on my findings, in order to assist the fact finder.

Is a likely hood ratio going to make fingerprint identifications more rigorous? I don’t know. In the end I will strive to make prudent decisions based on the information available, and reach a conclusion and hopefully our verification procedures will pick up any bias or misinterpretations.

The Brandon Mayfield case is a serious error and made by seasoned experts. The fact remains that fingerprint practitioners discovered the error. The same can be said of the Mackie case.

All erroneous identifications show is how unreliable humans even experts are. This is not new!

Humans make mistakes, take for example the deaths caused by human error in medical diagnosis; the ultimate sacrifice when test pilots fly untested rockets and a design failure occurs due to a faulty engineering or statistical calculation; Large corporations that have the best business and statistical minds and resources, that now require government bailouts.

100 years and the principles of fingerprint identification have not been falsified yet. That’s not to say it can’t be, nevertheless the principles have not changed. Thus the science is inherently reliable and observable.

The two biological principles of fingerprint identification (Uniqueness and Persistence) have been researched and published. Thus far there are no competing theories that can contradict this.

The most important quality of experts is that they are trained and devote years to recognising, classifying and comparing information within their chosen field.

Regular training, competency testing and certification (professional and organisational) for quality assurance can be used to mitigate an erroneous identification but not prevent it.

There is no such thing as the gold standard. There is no reason why a visual comparative science such as fingerprint identification requires statistics to make it significant to the fact finder.

I will leave it to the respected jurists to accept or reject my evidence. In the mean time I will do my job and be prudent in my judgements, always concerned that my evidence may lead to the incarceration of an individual, which to me is a very serious matter.

Let me leave you with a quote:
“Of crucial importance in identifications, is the spatial relationship between the points – called the ‘coincident sequence’ in the UK. It is the establishment of the coincident sequence which is one of the hallmarks of the expert: even the most sophisticated of computer systems cannot emulate the expert's skill of matching and discriminating between these two dimensional patterns.

Just as important is the consideration that the marks, which arise in crime cases are not of a copybook standard which a simplistic statistical approach demands. They may be incomplete, contaminated with background materials, smudged, blurred and, because of the elastic nature of the skin surface, will exhibit distortion of one kind or another. A comparison is then a series of decisions where questions such as “is this a significant difference?” can only be answered by skilled judgement. Likewise, deciding on whether a particular configuration of grey/white patches in a small area is a ridge ending, bifurcation or continuos ridge calls for the same skilled judgement.”

[Gerald Clough]

Is a likely hood ratio going to make fingerprint identifications more rigorous? I don’t know. In the end I will strive to make prudent decisions based on the information available, and reach a conclusion and hopefully our verification procedures will pick up any bias or misinterpretations.

Exactly. Having scientifically demonstrated information about the meaning of the analysis that takes place during examination in no way makes the examination more rigorous, or to be sure we're using the same meaning, more reliable. No matter what kind of valid conclusion statement is made, it's only reliable to the extent that the examiner accurately determines the nature of the skin represented by the impression. If I say, "My experience and what I know of the experience of others, combined with what I observe about this latent, leads me to conclude that A is the source.", one may accept my statement and might join in my believe, if there is reason to believe my reporting of the results of analysis is reliable. Ultimately, if someone questions the analysis, as happened in Mayfield, others take another look and report if they find the analysis reliable.

I guess we have to make a distinction here about reliability. One is how factually accurate is an analysis. The other is how reliable all analysis will be accepted to be. That second one is dangerous to apply. Maybe nearly all interpretations of impressions are accurate. But that didn't make the Mayfield analysis correct, not did it predict that it was inaccurate. I think we have to view reliability as something we always want to improve. And I think we have to take the fact that 100 percent accuracy is impossible to mean we should not reject out of hand any questioning of results. Seems pretty simple to me.

The two biological principles of fingerprint identification (Uniqueness and Persistence) have been researched and published. Thus far there are no competing theories that can contradict this.

The most important quality of experts is that they are trained and devote years to recognising, classifying and comparing information within their chosen field.

Regular training, competency testing and certification (professional and organisational) for quality assurance can be used to mitigate an erroneous identification but not prevent it.

Absolutely correct.

There is no such thing as the gold standard. There is no reason why a visual comparative science such as fingerprint identification requires statistics to make it significant to the fact finder.

That is true, but the sticking point is how it is used by the finder of fact. Demonstrating to what extent one impression agrees with another has value - significant value. Psychological evidence has value. A very great deal of evidence, the meaning of which cannot be proven, has value. But it cannot present statements as scientifically valid truths unless it is proven. That's what validation is about - defining how conclusions may be stated.

Let me leave you with a quote:
“Of crucial importance in identifications, is the spatial relationship between the points – called the ‘coincident sequence’ in the UK. It is the establishment of the coincident sequence which is one of the hallmarks of the expert: even the most sophisticated of computer systems cannot emulate the expert's skill of matching and discriminating between these two dimensional patterns.

Just as important is the consideration that the marks, which arise in crime cases are not of a copybook standard which a simplistic statistical approach demands. They may be incomplete, contaminated with background materials, smudged, blurred and, because of the elastic nature of the skin surface, will exhibit distortion of one kind or another. A comparison is then a series of decisions where questions such as “is this a significant difference?” can only be answered by skilled judgement. Likewise, deciding on whether a particular configuration of grey/white patches in a small area is a ridge ending, bifurcation or continuos ridge calls for the same skilled judgement.”

A useful quotation. It clearly demonstrates the mingling of the issue of interpreting the nature of the skin from the impression it made and the issue of when the source can be known to a sufficiently high degree of certainty. It implies that expertise can make reliable interpretations of impressions to describe the skin that made them AND that expertise can determine that the skin described very likely belongs to a particular individual. Without scientific validation of just HOW likely is "likely," all effort to reliably observe and interpret is corrupted by undefined standards and vague terms.

By every observation that can be made of the scientific, forensic, and legal landscape, no matter how one may wish for the old paradigm to continue, It's just not going to happen. It doesn't matter if you or I agree that it should or should not. Even if I didn't agree with the direction this is moving, I understand why it's moving. The issue has been decided. I don't worry about whether it will come to pass. I know the work that's underway in every aspect of the fingerprint identification field. And I've look hard at the evolution of law in other areas. While it's very hard to predict exactly how rapidly or slowly such a ponderous system will evolve, the evolution is inevitable.

[Neville]

Hi Steve

Yes it is a pretty impressive statistic and is considered today to be very conservative, and like most fields of science entirely theoretical.

"How many of these zeros do we knock off to account for human frailty and the messy real world? " I have no idea that's more your field, but seeing that 'statistics can be used to prove anything' why would I believe that any more than I would believe someone with 5 minutes of training in fingerprint comparison work and a wee bit more in practical application.

Yes I total agree that if I have just had an argument with the boss/wife or no sleep the night before because I have been at a homicide I am more likely to make a rash descion but surely that is why we have triple checking.
In 99.9999999999999999999999999999999999999999999999999999999999999999999999999999999999999999% of the time we are correct. On the other 0.0000000000000000000000000000000000000000000000000000000000000000000000000000000000000001% of the time when we are not then one of the checkers will pick it up. Then when we are told its going to court the whole process gets repeated. Well it does here in NZ and I believe in Aussie and England.

Hi Raybie
In my nick of the woods the Cartesian approach would be said to be fence sitting, but that's OK I have often been accused of that. I would not consider myself to be an extremist but it is amazing how we can move from being a Cartesian to an extremist by the changing of individuals within a setting and without any change in ones own thinking. Is there a scientific term for that? Evolution, maybe yes, no I think its devolution.

Please don't get me wrong here, I am one for change in any field but the change has to be for the better not just to satisfy egos. We do have scientific studies, we do have statistical studies to back up what we have stated for over one hundred years, to say we don't is 'to throw out the baby with the bath water', but remember the water must be muddied first and there seems to be a lot of muddying going on in the waters, and I guess the important thing is that I had my bath while it was still clean and warm. I must be getting old!

[Les Bush]

Good morning from down under and probably good afternoon to Neville in NZ.

Firstly let me apologize to Neville on the basis that I havent been articulating my position very clearly. I think the problem stems from learning and studying fingerprints first then finding myself surrounded by the language of statistical modelling and sampling and now validation. The main incentive to learn more about these subjects was the evidence given in Minn v HULL (2008). 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. 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. I'm not comfortable with using 'likely' or 'probable' or 'possible' and neither have been the committee for SWGFAST or the IAI. Although we have seen a move in recent times by the IAI to fence sit in the issue until research is completed. The type of research being conducted to establish whether the science needs to change its approach is based on statistical modelling, sampling and validation testing. My postings have been an attempt to gain insight into how a validation test design would satisfy scientific scrutiny and our normal practices. While not trying to be an echo using the mantra that our experience shows fingerprints are unique I for one would enjoy the challenge of subjecting the principle of fingerprint uniqueness to a rigorous test using a very strict measure for results (right/wrong). Our science is rich in positive studies such as the Peter Griffiths case, while others like Mayfield and McKie have cast doubt over the competency of experts but the science itself has come through okay. There has been a source attribution in Mayfield and McKie has been eliminated from Y7 so the principle of uniqueness works well. The counter argument is source attribution cannot be expressed as an absolute. The counter to the counter argument is that validation of our science using a test to determine the truth of exclusive source attribution is happening daily with no negative evidence falsifying the principle of fingerprint uniqueness. 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?