CLPEX Chat Board

Well?

Anybody want to bounce ideas around?

My latest credible information on such an idea is Prof Andre A Moenssens who makes mention (in his early 1970's edition of Fingerprint Techniques) of one Prof James W Osterburg and others who started work on this line in the 1960's but that is about it.

No detail, simply mention being made of it.

I think this concept fell by the wayside as there have been bigger fish to fry?

antonroland

Based on this thread and the one previous that you posted do you believe that level 2 ridge characteristics can have different values or must they all have the same value?

In a book by someone retired from the FBI they state that all points are equal.

Even SWGFAST to my knowledge has never stated anything on whether L2D is all equal or some unusual combinations have more value.

With South Africa requiring 7 points I just wondered if that was 7 common points or 7 unusual points.

Some will go along with Locard and say rarity in event and relationship is certainly a consideration while others will say 'A point is a point is a point and they are all the same----just like there is no disagreements at all between LPE'.

Just curious as to what you belive---there is unusual variety that should be considered OR all things are equal because you just cannot put a value on UNIQUNESS?

Doesn't this come under the "no scientific basis" category? What does that mean? For example, there's much made over the quasi-official statement that a point count threshold has no demonstrated scientific basis, but does that mean no expert can apply a point count standard? Why not? There's no scientific basis to invalidate it, either. For that matter, there's no demonstrated scientific basis for the usual expert belief conclusion. The point is that, when there's no conclusive science, it becomes a clinical experience based issue, and experts may differ on whether or not they weight different structures differently.

I should note that I don't mean there's no scientific basis for distinctly different frequencies of occurrence of different structures. There certainly is, from Kingston, Osterburg and others. ( http://www.latent-prints.com/images/Osterburg.pdf ) I mean validation of the concept of considering different weights according to the different frequencies.

But my distinct feeling is that treating every kind of detail as having equal weight is tantamount to declaring a point count standard.

Gerald, I always enjoy reading your responses!

The thing to me is, once you acknowledge certain structures / features / details / points as "high value" you open a door that is best left closed...or better yet, dont even break a hole into the wall...leave it be...

If I am prepared to say that minutia A is of a higher value than minutia B, I must surely be able to back up that statement, right? Now then, what do I base this on other than my personal experience?

I'll see your high value points and raise you one location.

What I mean is that I believe not only do points have varying value, but that value can be imparted onto a point due to it's location. For instance, we expect a high level 2 count around cores and deltas, we expect an abundance of ridge endings underneath and flowing out from the core in a loop. Often times for me the 'tipping point' to sufficiency can be a ridge event in an area approaching the upper boundary of the latent. Perhaps there's data from AFIS feature extraction or from Glenn and his statistical data on how accurate my suppositions are, but it's been my experience nonetheless.

Backtracking a bit, I'd also say that it's untenable to say that all events carry equal weight. We have the smiley files page for more than the mere fact that it's 'cute', this combination of characteristics in this cumulative formation is relatively rare, which is the exact thing we say when we make an individualization.

Osterburg developed frequency numbers for various structures and then made some probability calculations based on those numbers. It sounds reasonable, but it hasn't been shown through vigorous study that those probabilities conform to the many possible variations of arrangements of features in the population. And it lacks the informal but persuasive validation of the admittedly foggy but apparently more than sufficiently conservative community standard threshold among examiners, that validation being in the form of noting no revealed errors not accounted for by fraud or paper error or clear and demonstrable misinterpretation of features. In short, the traditional examination is validated, if not in a very scientifically satisfying way, then at least in a very pragmatic way that has been accepted in legal affairs. We don't even know what kind of probabilities a valid statistical model would produce in cases judged just sufficient in the traditional way, so we can't apply the historical lack of error to the statistical model. What probability could you adopt as a threshold for declaring a sufficiently likely fact? Would you even do such a thing? I don't think you can mix frequencies with conventional examination, because you can't statistically validate the conventional mode, and when you insert probabilities, you abandon a conclusion of belief in favor of a statistical statement.

And I think it would be less than persuasive to try to apply individual experience to any treatment of features as having relative weights. An individual experience isn't that relevant. It takes a very much larger pool of experience to convince that whatever threshold the whole of practice and training seems to impose on examiners is very much more than sufficient for accuracy. It's not that the individual doesn't reliably judge some structures to be relative rare. It's that the individual experience can't validate using that known rarity. We just haven't submitted (and will likely never submit) a mode of traditional examination with relative rarity of features considered to the historical lack of error type validation.

Thanks again Boyd and Gerald for your responses.

You so beautifully put into words what I believe. I am also happy that there are no recent studies that suddenly introduced silver bullet values to minutiae.

Note from a 'Dinosaur':

Osterburg gave a presentation & had a roundtable discussion at the IAI Conference in Phoenix, Arizona in 1979. I'm pretty sure I have some documentation of that buried in a storage garage. Most of that Conference was audio-taped, so there may even be an audio tape still floating around. My recollection is that Osterburg gave information about his frequency studies, & I believe that it included the co-relation with both other features & location in a print. He was requesting technical assistance from the fingerprint identification community, but was not particularly well received at the conference (that may be a bit of an understatement!). Of course, at that time, the name Ashbaugh or the term 'Ridgeology' were still years away from being widely known, and there was significant resistance to change.

Some within our ranks now, including those coming from a scientific background ( & especially proponents of statistical models), may well believe that some things never change.

If interested, I could attempt to dig out the documentation, although I believe that the most relevant information was published by Osterburg & should be available elsewhere.
Regards, RLT

Something may be found in these references:

Fingerprint Probability Calculations Based on the Number of Individual Characteristics Present, James W. Osterburg, Identification News, 24:10 (1974)

Development of a Mathematical Formula for Calculation of Fingerprint Probabilities Based on Characteristics, J.W. Osterburg, T. Parthasarathy, T.E.S. Raghavan and S.L. Sclove, Journal of American Statistics Association, 72:360 (1977)

The Ninth Ridge, A.P. Rienti, Fingerprint Whorld, 13:51 (1988)

That would be grand but let's first see what turns up on this thread and thank you so much for offering...would hate to see you unpacking stored boxes and the info comes up here.

http://www.latent-prints.com/images/Osterburg.pdf

"Development of a mathematical formula for the calculation of fingerprint probabilities based on individual characteristics" from the Journal of the American Statistical Association (1977)


http://www.cse.msu.edu/biometrics/Publi ... PAMI02.pdf

Also, this paper titled "On the Individuality of Fingerprints" has a table on page 13 that lists the various probability models over the years and gives rise to the fact that people have used minutae location, type and orientation among others to calculate probabilities. This is an implicit recognition that these variables have qualitiative value considering it's part of the methodology for their calculations.

Gerald Clough wrote:And I think it would be less than persuasive to try to apply individual experience to any treatment of features as having relative weights. An individual experience isn't that relevant. It takes a very much larger pool of experience to convince that whatever threshold the whole of practice and training seems to impose on examiners is very much more than sufficient for accuracy. It's not that the individual doesn't reliably judge some structures to be relative rare. It's that the individual experience can't validate using that known rarity. We just haven't submitted (and will likely never submit) a mode of traditional examination with relative rarity of features considered to the historical lack of error type validation.

That's a relative statement for sure. Of course it would be less persuasive in an IAI presentation or JFI submission titled "What Boyd thinks about stuff relating to level 2 characteristic values" . However, in a real world office environment, the unspoken sentiment when you have a case with individualizations is "based on my experience, this combination of information has enough relative weight to individualize". This sentiment is further exemplified by what critics love to assail as 'verification shopping'. If you have two examiners with different tenures (one long and one short) and a complex print, you run the risk of experience preventing a verification if the new examiner does not have the requisite ability to discern the value of the elements used in the comparison.

On the flip side, statistics can be just as subjective. I was reading an article from the 2010 Boston WVU conference last night titled "The Forensic Analysis of Footwear Impressions" where the author goes to state:

As stated earlier, Stone (2006) developed a hypothetical probability model for identifying characteristics. A metric grid with divisions of 1 mm2 was used to help quantify the occurrence of identifying characteristics on a shoe outsole. By placing the grid over a size 8 1/2 shoe, Stone created a 16,000 mm2 grid on the surface area of the shoe outsole. Stone used the following simple formula to calculate the probability of a single characteristic (point):

Pe = m/n

where Pe = probability of an event occurring,
m = number of ways of success, and
n = total number of possible outcomes.
The probability that another outsole contains a single characteristic (point) in the same position is 1 in 16,000

Those links to articles I provided in my previous post use a similar method, in that a grid is placed over each print. However, I can derive a smaller probability by decreasing the
size of the squares in the grid. If I'm able to diminish the size of the grid squares to 25000 instead of 16,000 then I now have a 1 in 25,000 chance that two impressions will share one characteristic. What's the scientific reason behind a square milimeter being used as the standard of measurement? There is none.

Boyd Baumgartner wrote:

Gerald Clough wrote:And I think it would be less than persuasive to try to apply individual experience to any treatment of features as having relative weights. An individual experience isn't that relevant. It takes a very much larger pool of experience to convince that whatever threshold the whole of practice and training seems to impose on examiners is very much more than sufficient for accuracy. It's not that the individual doesn't reliably judge some structures to be relative rare. It's that the individual experience can't validate using that known rarity. We just haven't submitted (and will likely never submit) a mode of traditional examination with relative rarity of features considered to the historical lack of error type validation.

That's a relative statement for sure. Of course it would be less persuasive in an IAI presentation or JFI submission titled "What Boyd thinks about stuff relating to level 2 characteristic values" . However, in a real world office environment, the unspoken sentiment when you have a case with individualizations is "based on my experience, this combination of information has enough relative weight to individualize". This sentiment is further exemplified by what critics love to assail as 'verification shopping'. If you have two examiners with different tenures (one long and one short) and a complex print, you run the risk of experience preventing a verification if the new examiner does not have the requisite ability to discern the value of the elements used in the comparison.

My point was that while an experienced examiner (who happened to have access to a valid statistical model) could make some credible statements about the calculated likelihoods in his cases where he concluded identification, that one examiner's cases are (1) not numerous enough, considering that many or most involve lavish quantities of details and are therefore not good threshold tests, and (2) we don't know how many cases in how many places it might take to have any hope of the interesting kind of error being discovered. We have that wide historical community experience to validate conventional conclusions. We obviously don't have that with declaring belief in identification based on a modeled likelihood because we don't have believable models.

I think it is very likely that, in the event (to me unlikely) of having a valid statistical model, the likelihoods calculated in cases where identification had been concluded in the conventional way would be infinitesimally small in world population terms. In fact, given that the conventional mode has been informally validated in the historical record, one might reasonably claim some degree of validation for a statistical model that predictably generated extremely small likelihoods from traditional identifications. Logically, if any proposed model did not generate very convincing likelihoods, even in the barely sufficient identifications, we should over 100 years have seen at least a few errors from two sources creating indiscernible impressions.