CLPEX Chat Board

Kevin, I apologize, you're right, and certainly we did not mean to "highjack" the thread! So I thought I'd start a new thread to say something about the phrase "infinite variation" since it so frequently invoked.

We had a good discussion going on this in another thread a few weeks back-good stuff. I would point out that DNA is working with base pairs (that only go together one way G-T and A-C) in specific sequences on a specific location to make a match. They extrapolate their numbers based on the size of their database and interpret that to the jury....with FP analysis you are dealing with infinite variability between just minutia not including level 3 detail. You can't extrapolate that into a probability to any degree when dealing with latents in my opinion...AFIS does with their search algorithims but it can't hit the print by itself. just sayin

1. The analogy to DNA is not quite right. Technically STR DNA methods compare the number of repeating sequences at a site...meaning a "12, 13" is someone with 12 repeating sequences of a certain description (i.e. AGT,AGT,...AGT x12 and AGTx13 on the complementary chromosome. Thus technically, there is no reason someone can't be a "120, 180"...it just doesn't happen. For that site, maybe the rarest someone sees is 30 repeating sequences...but technically 120 could happen...it's just statistically improbable. Thus DNA in this test DOES have infinite variation, but they are not limited by this and can still make statements regarding source attribution.

2. The same is true for fingerprints. Putting aside L3D (which by the way, HAS BEEN and IS CURRENTLY being modeled, therefore...means it can be captured and is not "infinite"), but just focusing on L1D and L2D, the variation is NOT infinite. It might be vast and it might be untestably varied, but it's not infinite.

IF it were infinite, I would occassionally see a fingeprint shaped like Nicole Kidman or maybe see the outline of Johnny Depp, maybe even a little Ricky Gervais profile in another. Instead we predictably see common patterns (loops whorls arches) and the ridges predictably flow through these patterns in typical ways. In certain areas of the prints, the ridges behave very characteristically: deltas have lots of short ridges and ridge endings, type lines lots of ridge endings, cores have stacked/multiple enclosures, sometimes mirrored on either side of the rod. In fact, every single clue you've ever learned to "search smart, not hard" means that there is a pattern. And where there is a pattern, there is a loss of randomness.

No doubt these features are highly variable, exhibit extreme characteristics of ability to discriminate various sources...but I think we may need to move away from invoking this term of "infinite variation".

The very fact that we can model and capture L1D L2D and soon/recently L3D, means that this variation can be expressed at some level. And we will only get better at it.

I have often asked colleagues, when the technology is finally here and validated and can express L1D, L2D, and L3D in terms of probabilities, what then will be the excuse for not using this technology. Since that's what we do, use all 3, what reasons are left?

What do you think, Kevin?

g.,

I couple of comments on your logic:

First, with your point on DNA. I understand your point about STR and repeating sequences, BUT... DNA is a linear molecule with an end. It cannot have infinite variation just for the simple fact that it eventually ends. Also, I think kevin's point was more that there are limited (4) choices at any point along a DNA strand. Applying statistics fits this model because there is a simple random probability even to longer repititions of the same STR sequence.

Secondly, I would argue that even though there is obviously not infinite variation in what a fingerprint COULD look like. There is an infinite random variation as to what each fingerprint DOES look like. The randomness in the formation of fingerprints give them infinite random variety. In other words, I think that our understanding of how fingerprints form give us the ability to say that there is no limit to the number of different, unique forms that fingerprints will take.

I don't think your argument about Nicole Kidman and Johnny Depp holds. Something can be infinitely variable and still fit within the constraints of a specific form. A random number between 1 and 10 is no less random because of the upper and lower limits. If you remove the constraint of having to be a whole number, the random number is infinitely variable. Similarly, a ridge near the tip of the finger will flow parallel to the nail (usually), but there is still infinite variation in the angle of the flow and events along the line. Limits do not remove infinite variation.

-ER

The Nicole Kidman and Johnny Depp fingerprints don't require infinite variability. Either is as likely to occur as any other variation. Well, not really. There are probably discoverable likelihoods for various arrangements of details that would make them less likely, if for no other reason than the overall patterns that strongly tend to form through the known biological processes involved aren't very likely to produce those images. But obviously they could happen. I just wouldn't be waiting around expecting one to crop up.

But, back to the question. Angles and events exhibited by a ridge at any point cannot be said to be infinite for our purposes. While it's true that there are, for instance, an infinite number of angles that can be formed by two vectors, we cannot differentiate between two angles whose difference is below our ability to resolve. If I determine, for example, that I can observe a difference between two angles no less than of 0.1 degrees, I can recognize only a finite number of angles, in this case 3,650 different angles. Because we are limited to our ability to observe differences, every difference we can define produces a finite number of observable differences. There is also an upper limit on how many details we can possibly observe in any portion of friction ridge skin. To put it simply, you can only pack details to some maximum density in any area of skin. And there is an upper limit of the possible area of friction ridge skin.

A finite number of possibly observable details in a finite maximum area of contiguous friction ridge skin and a finite number of observable ways in which a detail can vary. Put whatever believable number you wish on any of those, and you get a calculable finite number of variations in that maximum possible area of skin. Only if the area of skin were infinite could it vary infinitely.

Is anything infinite? Somewhere, as you look closer and closer to see finer and finer variations, you eventually get to a subatomic level and encounter the probabalistics of quantum physics and lose the concept of observability. But we don't have to go there. We look at skin that occurs only up to some area we could at least say can't possibly be larger than some size we could suggest, and we can observe variations in that skin only to a limited degree of differences. So we're stuck with a finite maximum number of observable variations.

Gerald

My thoughts exactly. I loved ER's analolgy of bound 1 to 10, but there's an infinite number between each number as you go deeper and deeper. That's a great analogy. But like you say, and really Ashbaugh's 3rd principle of his Philosophy of Individualization ("having "detectable" (or sufficient) uniqueness), at some point, you lose the ability to differentiate between these things as you go smaller and smaller.

We all know pi in theory may be infinite...but at some point with my handy desk ruler, I can't tell the difference between 3.14, 3.141, or 3.1415, and so on. It might go on to infinity, but I can't tell the difference. Throw in a tiny distortion and 3.14 could even be up to 3.2 or as low as 3.1. I'd say with latent prints it's the same...the skin may have infinite variety down to the skin cells, molecules, atoms, etc...but as you said, with latents you become bounded and then have to allow for more degrees of freedom because of distortion....As a result, never mind that each ridge ending in the skin is "unique", because in my latent, now I can barely differentiate ridge endings, b/c you lost me back at the photocopy of the crappy livescan card that barely gives me L1D and L2D to do my job...

g.

Interesting discussion, although my initial impression was that there was a bit of 'apples and oranges' when attempting to compare fingerprint comparison to DNA.

Although the papillary ridge structure of skin itself may be as close to infinitely variable as we get (due at least in part to biological diversity and random variation), the observable features of sufficient reliability as represented in a transferred image of some portion of that skin will consist of a subset of the actual data, thereby of necessity being finite. However, that does not mean that the available data is not sufficient to reliably individualize. Statistical analysis and probability modeling may have a place in providing evidence in support of such a premise, and provide a path to a better understood and more definable level of sufficiency.

If it is agreed that the formation of friction ridge skin is initially dependent on DNA coding, then is the data at that point finite or infinite? Has an individualizing process already occurred, or is it entirely dependent on random events influencing the cell by cell building of the friction ridge skin? If there are multiple layers of random events effecting the final structure and appearance of the epidermal layer of friction skin, how is that accounted for in probability models?

I do not follow g's argument regarding DNA and infinite variation (perhaps I am not well enough versed in current DNA comparative analysis & the underlying science). It would seem though, that there would be a constraint in what can be detected and measured with current technology, similar to what has been pointed out in fingerprint analysis.

There have been a vast amount of resources devoted to historical weather data gathering & predictive modeling, most notably related to the current controversy over 'global warming'. While the jury may still be out on the ultimate outcome, it has become abundantly clear that the probability models used to date have been severely inaccurate. We should strive to avoid the same fate when contemplating a reliance on probability models related to fingerprint identification.

RLT

RL Tavernaro wrote: However, that does not mean that the available data is not sufficient to reliable to individualize. Statistical analysis and probability modeling may have a place in providing evidence in support of such a premise, and provide a path to a better understood and more definable level of sufficiency.

I've so far identified three (maybe four) possible applications of the knowledge, should we manage to become able to speak with authority about those probabilities.

1. An actual probabalistic conclusion of examination. Something along the lines of stating the likelihood of a specific set of particular characteristics occurring in particular relationships with each other. A most ambitious goal of a genuinely scientifically valid conclusion. After trying to think through even the most obvious issues in designing the studies in this direction, I suspect it would take a generation to produce valid results.

2. A lesser product of the same kind of work required to validate (1.). This would provide probabilities that would not be sufficiently precise or sufficiently widely applicable to produce reliable conclusions but would be direct scientific argument of the power of the current conventional expert statement of belief conclusion. Essentially, it would be a demonstration that the current general standard of practice justifies an expert belief in a source, rather than "we must be doing it right, because we've not been shown to be wrong." This should be clearly distinguished from probabalistic conclusions. The examination conclusion here remains as it has been, a statement of the expert's willingness to rely upon the evidence to name the source. Using the probability research products would simply bolster the experience-based argument that it's reasonable to hold that opinion. It does not so far seem that such additional support is needed. The arguments in favor of fingerprint evidence and the courts' favorable opinions of it have been essentially unchanged for a long time.

3. To allow examiners to form their conclusions from from a lesser quantity of data. Perhaps we discover through statistical analysis that some arrangement of features that we knew was not frequently seen was actually very unlikely to occur, far more unlikely than just the generally unusual status we were otherwise assigning to it. That would be new, specific knowledge that could go toward reaching a threshold of conclusion. We can't know what might pop out of the research, because we really don't understand the dynamics of ridge formation at that level. Maybe particular areas of simple unbroken ridges at some locations are sufficiently unlikely as to become themselves a kind of super-detail.

There's fourth kind, a sort of hybrid application in which the examination conclusions is conventionally reported, and the known likelihood of some array of features is appended. This is not so outlandish as it might sound. Very large and very small numbers, such as the kind of likelihoods we're talking about being useful, don't mean much in any concrete way to lay fact-finders and take on reliable meaning only to expert interpreters. The expert knows the nature of the field and the accumulated experience and can say why such numbers are convincing.

When thinking about in any way changing or adding to the nature of conclusions, you can't say, "Oh, no, no. That's not in the guidelines" or "that's not the way we do it." We're kind of spoiled in that way, too, having gotten to say how we say. But the reality is that whatever is learned will be applied not according to our traditions but according to legal decisions on whether it's more probative or prejudicial, and the fact that the genie can't be stuffed back in the bottle isn't reason not to rub the lamp.

Of course, I pretty much believe any currently working examiner would be fortunate indeed to live long enough to see these things become validated and in common use. But I'm aware, too, that predictions like that are historically among the most unreliable, since I'm a child of the 50's and still don't have my flying car, have not found anyone offering permanent homes under the sea, and have been unable to book my vacation on the moon, and I remember that in 1977 the president of Digital Equipment Corporation couldn't see why anyone would ever want a computer at home.

at some point, you lose the ability to differentiate between these things as you go smaller and smaller

I agree. At some level you cannot tell the difference anymore, especially in the latent. However, the infinite variation still exists. (While it does not in DNA.)

I think that exploring the probability models is a worthwhile exercise. Eventually, I foresee the probability models becoming a theoretical support of fingerprint identification, but not becoming a day-to-day part of practical fingerprint analysis.

I think that's right for the foreseeable future. For one thing, it's just such a massive task to do any kind of comprehensive research. I note that the FBI paper in the other thread mentions the use of statistical research, too, in this vein. It would be interesting if it might discover some specific, very rare events that might allow an identification with something less than is currently accepted as sufficient. I don't think that's all that unlikely. I think we would agree that all our observation tells us that ridge features don't just form entirely randomly. By that, I mean we see almost exclusively a rather limited number of ways in which they form. We don't see every imaginable arrangement. That tells me the forces that are at work produce irregularities in certain ways that conform to a predictable set of forms. Just as an example, we do not expect to see a kind of "feathered" or "fringed" ridge, either having a string of bifurcations, one after another close to each other down a ridge or a ridge that ends and becomes a series of chevron-like short diagonals before again becoming a continuous ridge. Such things, if they occur, must be very, very rare and might be very powerful factors. One of the nice things about statistical analysis is that, once you work out how to code information, it reveals what's significant. It can tell you what's significantly unusual without you depending on an feeling based on experience.

Neither fingerprints nor DNA are infinitely variable when dealing with a finite set, which we are. It is exactly as Gerald put it

Only if the area of skin were infinite could it vary infinitely.”

And the same applies to DNA.

People have a very hard time conceptualizing “Infinity”, because it is boundless and therefore impossible to fully grasp. And so it is often mistakenly applied to extremely large, but still finite numbers.

If I had a single epithelial cell and an infinite number of cells to compare it to I would eventually find one that was identical. And by identical, I am not just talking about the gross structures of the cell like mitochondria, golgi bodies, rough and smooth ER, etc., but down to the atomic and subatomic levels. And not only would I find that identical cell, but I would find an infinite number of them. Likewise, if I was looking for a finite patch of skin such as a fingerprint with an infinite number of patches to compare it to I would find an infinite number of identical ones.

We are dealing with a limited number of cells/molecules and a limited population to compare them against so there is no way they are infinite.

It the variability large? Sure…but it is infinitely far from being infinite

PG

g.

Thanks for the reply! I wasn't trying to get into a stone throwing match between DNA and fingerprints - I always see the two as being partners in crime scene and criminal investigation so no offense to DNA people. They both have their merits and limits and I tend to disagree as to the characterization of the new 'gold standard' versus the 'old'... it is really case by case as to which is more probative in my opinion. But back to the issue though-

The very fact that we can model and capture L1D L2D and soon/recently L3D, means that this variation can be expressed at some level. And we will only get better at it.

I have often asked colleagues, when the technology is finally here and validated and can express L1D, L2D, and L3D in terms of probabilities, what then will be the excuse for not using this technology.

Yes!!! And keep up the good work- it is a necessary step in how we are received by the scientific community as we, as a science, are evolving alongside the rest of the forensic community. I think if and when the technology and research to capture L1-3 catches up you are going to have a statistical model that will rival or surpass DNA's probability model. I'm going to quote Gerald because he puts it into perspective pretty well here:

The examination conclusion here remains as it has been, a statement of the expert's willingness to rely upon the evidence to name the source. Using the probability research products would simply bolster the experience-based argument that it's reasonable to hold that opinion. It does not so far seem that such additional support is needed. The arguments in favor of fingerprint evidence and the courts' favorable opinions of it have been essentially unchanged for a long time.

But foresee a Fingerprint statistical model will probably be a bit bolder than the decision making process of many latent print examiners. And I concede I am speculating on this point because of what we were just talking about - what is infinite variability?? I am speculating that a probabalistic model for FA would be bolder than the DNA probability model because of the disciminatory power of FR skin. As far as nailing down what is infinite variability I was speaking in terms of the entire fingerprint population as it applies to our body of empirical evidence rather than a latent fragment and/or formation. However I am a bit hesitant as to implementing the idea of a probabilistic match for a couple of reasons:

1) What is we come up with a statistical model based on L1-3....is an examiner still encoding the print and marking the points? Is the model it just going to be a calculator or is going to be discriminatory on its own?

2) What is it gives us a 70% probability and its a print you or I don't want to call? What is the detective or prosecutor gets ahold of the report and decides to run with it on his or her own?

I understand I am saying the discriminatory power of fingerprints would make for a stronger, more discriminatory database and I am contradicting myself at the same time when saying a probabalistic model might be a slippery slope but thats kind of the state we are in; the examiner is still the most discriminatory factor in the latent print process and I don't think you can divorce the examiner from the decision making process without serious repercussions. I think the role of research in stat modeling should be a central focus in future research within the fingerprint community but never at the expense of the examiner or the integrity of the conclusion(s).

Great debate on this subject btw...its worth talking about even if we have to hijack a few threads to do it

If I had a single epithelial cell and an infinite number of cells to compare it to I would eventually find one that was identical. And by identical, I am not just talking about the gross structures of the cell like mitochondria, golgi bodies, rough and smooth ER, etc., but down to the atomic and subatomic levels. And not only would I find that identical cell, but I would find an infinite number of them. Likewise, if I was looking for a finite patch of skin such as a fingerprint with an infinite number of patches to compare it to I would find an infinite number of identical ones.

Sorry, Peter. This cannot be true.

First, there may be a large number of cells that appear identical to you with limited means of measuring the differences. However, infinitely small differences in the distance between objects in the cell make them different. The mere fact that there are an infinite number of measurements between 1 nm and 2 nm and an infinite number of possible angles gives us an infinite amount of variation. Again, I'm not saying that you can tell the difference, but that they exist. In addition, since Heisenberg says you can't know both the location and momentum of a particle, you can't compare to the level that you describe.

Secondly, since we live in a finite universe, there are not an infinite number of cells or patches of skin to look at. There are a finite number of stars and they will all eventually burn out (theoretically), therefore we cannot find an infinite number of identical patches of skin that match, or an infinite number of cells that match.

Finally, infinite variation can exist in a finite set. Again, pick a number between 1 and 10. There are an infinite number of possibilities. Skin does not have to be infinite to have infinite variation. (Much of this is theoretical and therefore meaningless to real work, because as g. pointed out, eventually you can't tell the difference.)

I would agree with you that DNA has a finite number of possible arrangements because it has precisely 4 choices at each point and the DNA molecule has an eventual end. (It can't be infinitely long.) This is essentially limiting us to 'whole numbers' in the 'pick a number' analogy.

ER
I am going beyond our ability to measure and speaking completely hypothetical. And yes there are infinite distances between the objects in the cell, but I am using an infinite number of cells to compare. So I suppose “identical” could be better be stated as “infinitely the same” since I suppose identical implies an ending which infinity does not have.

I agree with your second statement and mentioned there are a limited number of cells.

As for skin not having to be infinite to have infinite variation, I actually agreed with, but would then also be able to apply the same argument to DNA. It too can be infinitely different in that sense. But, both skin and DNA are products of structure and function and I think in that sense there is a finite number of arrangements for both when looking at finite quantities.

PG

Not sure how this interesting theoretical discussion on 'infinite' is going to help.... but the answer to the discussion of 'infinite' maybe resolved (or at least move forward) if we recognize that 'infinite' is not one thing; there are different kinds of 'infinite'. See for example, numerable infinite vs innumerable infinite (also called countably infinite. vs uncountably infinite), and Cantor's distinctions and classic theorem of 'infinity of infinities'.

Hope this helps,

Itiel

Peter,

Glad to see we're on the same page.

I think the main difference between between DNA and fingerprints in this discussion is that there are a theoretical infinite number of possibilities at an infinite number of places in a fingerprint, while DNA, by its structure, has only 4 possibilities. You cannot get the infinite and unmeasurable variation. There is no G.00001. It has to be just G, or just A.

I would therefore classify fingerprints as infinitely variable, but DNA as not. (Which is good for those DNA folks and their statistics.)

Oh, and this all reminds me of a comic in the newspaper with one astrophysicist saying to another astrophysicist, "I may not fully comprehend infinity, but I comprehend more of it than you do."

kevin wrote:
1) What is we come up with a statistical model based on L1-3....is an examiner still encoding the print and marking the points? Is the model it just going to be a calculator or is going to be discriminatory on its own?

If we assume there will one day be sufficient valid study to say how likely it is that some features in a specific arrangement will occur, the nature of our application still requires someone to interpret what is impressed and characterize the physical reality of the skin that made the impression. That's the same task then and today. You can't compare two impressions under any process unless you can confidently characterize the skin that made them. All known probabilities would do would be to provide a scientifically validated way to make a statement of likelihood. We're making a statement of likelihood today. We just don't measure it numerically. Now, that doesn't mean the conclusions are the same kind of conclusions.

Today, a conclusion of identification is a statement of our personal belief in the identity of the source. We're saying that our high standard of proof has been met and we're offering an opinion backed by expertise. In a sense, we're saying, "I believe, and my expertise is such that you, as a legal fact-finder, should share my high confidence. If I say the likelihood of any person matching this latent is one in ten million, I'm telling the fact finder how many people in the set of people who might have left this mark could have left this mark. Is that good or bad? Is it more probative or more prejudicial, and is the prejudice in favor of or against the proposed identification? I don't know. It's not my role to say. It's certainly not my role to preempt the courts and refuse to reveal what I know to be valid knowledge. Like just about everything else in our legal realm, it's a matter of making a best guess that's so much more likely to be correct that we will rely upon it in determining whether the question being posed in the matter can be answered with the degree of confidence required, which is in criminal matters beyond a reasonable doubt.

2) What is it gives us a 70% probability and its a print you or I don't want to call? What is the detective or prosecutor gets ahold of the report and decides to run with it on his or her own?

Well, first, I think when we're talking about probability conclusions, we're talking very large probabilities, perhaps in the kind of numbers used in DNA conclusions. DNA analysts could make one-in-ten statements, and that information would be useful to investigators, if only to express how weak the DNA evidence is, but I think it would be held to be misleading in court and not helpful to a jury.

I believe it is not for us to say what knowledge may or may not be used. We feel like it is, because we've for so long been largely unopposed and have been the default gatekeepers of our own field. But it is entirely a matter for the courts and litigators to determine the admissibility of evidence. They will make that determination as they always have, by judging where it is more probative than prejudicial. In this issue, the court would have to decide if such a statement of probability was (1) valid, in that it was sufficiently supported by research and logic, and (2) whether it could assist the finders of fact to adress the trial issue or whether it was unjustifiably persuasive. For example, prior to any specific research into eyewitness identification, experts in the general fields of psychology and perception were offered to talk about why identifications might not be reliable. The trial courts, supported by appeals courts, including federal circuits, held that the courts were right to exclude their testimony. The reasoning was that (1) their knowledge was not sufficiently specific to eyewitnesses, and (2) such testimony therefore unjustifiably called into question the accuracy of the identifications. It would hurt more than it would help. Now, as more specific research is being done at places like the UT El Paso Eyewitness Lab, circuits are beginning to hold that it's error to exclude the new research testimony. It's deemed specific to the actual task of the eyewitnesses, so it helps the jury judge the reliability of the witnesses.

And keep in mind that what prosecutors can say about what evidence means is closely watched. Many cases have been reversed because a prosecutor told a jury something was "a match" or something "proved" some fact, when the testimony did not reflect that. It's kind of the same test. Did what the prosecutor say properly remind the jury of some testimony they could use, or was it more prejudicial, in that it works against their proper use of the testimony?

This is something of the tack taken in opposing fingerprint evidence. The argument is that fingerprints are unduly powerful, that the conclusions are insufficiently supported, that various experts have something helpful to say about the nature of examinations, and that their knowledge should be admitted to dampen the undue prejudice of fingerprint evidence. So far, they've been experts like Cole and others whose research has been held to be insufficiently specific. That, of course, is changing. Not in the direction of admitting the previous experts, but in the direction of admitting new, specific research that's on the horizon.

But I think there's a real difference between what an investigator uses and what a prosecutor tries to sponsor. I've used polygraph to good effect. The courts don't allow it as evidence. I've made many cases using hearsay, even anonymous hearsay and direct information that I cannot talk about in court. I know I differ with many examiners in believing that everything we can say about a piece of evidence has value. I have to be responsible and accurate in how I report it, and that includes providing information on the significance of the information. How many examiners have told an investigator that they want very much to obtain a better set of prints from suspect X? in effect, that's saying the examiner suspects a match might be had, if a little better exemplar can be obtained. It's essentially a statement of probability.

Other fields have measures to prevent what they believe is unjustified use of or confidence in some lesser conclusion. CODIS requires a minimum amount of DNA information before they will run, even though running with less might produce something useful. Others routinely make use of class conclusions. A tire mark expert may conclude to the make and model of a tire or make and model and indicated mechanical defect, without concluding the specific source. We do that, too, when we exclude based on L1. Would you withhold the information that you could determine a rather uncommon sequence of L1 classes across ten fingers, the positions being clear from the impressions, and that a known suspect shared that sequence? Do you guess that it won't be useful and thereby insert yourself into the investigative process? Do you say, we just don't trust how you'll use that information, so we'll make an investigative decision and withhold it? How do you justify withholding that information when it may well be that the investigation has shown that the actor was one of an identified set of people, and only one of them could have made those latents? Baaaad analyst. Baaaad scientist.

Remember that many things that are not admissible at trial are quite properly useful in probable cause and that acting on probably cause is often a very important investigative move. Even anonymous information and past behavior are good in a PC affidavit. And pretty much anything goes in grand jury. You have to decide if you're an analyst or an investigator. Obviously, an examiner is bound to be an analyst. Even when the examiner is also an investigator, the mindset must change with the roles. This was strikingly demonstrated in a crime analysis course. The instructor told us on the first day that half of us would not be able to do it. The individual results of our analysis of actual data from a real case in which we were asked to provide the characteristics to be used in selecting a decoy and time and place of offering for a serial rapist, showed he was correct. Half the students were unable to shed the investigator mindset, and it prevented them from just letting the data speak. The successful half described exactly the decoy characteristics that were chosen in the actual case in which the decoy attracted the actor on the first attempt, at the time and in the place specified. The "investigator's" decoy, different in age, dress, apparent occupation, and hair color and offered at the wrong time, would have missed. Bias. Clear unconscious bias, even after a week of rigorous training in analytical methods. That's why we have investigators and analysts and why, when as investigators, we do analysis, we have to know if we can effectively shift roles, and why we have to stay suspicious of ourselves. The analyst mind properly doesn't care if the case is ever cleared. Doesn't care who it turns out to be. The job is to analyze and provide information, without regard to its effect on the investigation. If your "analyst mind" starts caring, you have then invested in the outcome.

These are not comfortable questions. They are the same concerns that people like medical researchers have. What if we publish what we believe are our valid results, and it leads lay patients to make poor decisions? What do we do with a drug trial that hasn't been completed but which shows dramatic benefits and could save many lives today? But they are easy questions to answer when the roles are clear.