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

[mart]

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

Probably need to have this translated !! Will try and answer what I think your asking....

I take each print on its own merit, weigh up all the available information (1st, 2nd and 3rd level) then evaluate whether there is sufficient..... as simple as that !

[Neville]

maybe Tavernaro means metrology the 'science of measure' rather than metric as in the normal method of measuring. I am sure Tavernaro is not suggesting you measure between points with a metric ruler in millimetres. Though that does not really make it too much clearer.

Maybe Charles or Pat can translate I think they are from around that area and I can understand them.

[Charles Parker]

Definition: Metrics are a set of measurements that quantify results. Performance metrics quantify the units performance. Project metrics tell you whether the project is meeting its goals. Business metrics define the business' progress in measurable terms.

metric - A measure for something; a means of deriving a quantitative measurement or approximation for otherwise qualitative phenomena (esp. ...
en.wiktionary.org/wiki/metric

I am not from Arizona. I do know some nice people from there but I am not from that area.

I think what Bob is referring to is:

Q At what point do you know you have an ID or Exclusion?

A1. When I see it I will know!
A2. My training and experience tells me when I have one!
A3. When the Ouija Board tells me!
A4. When the investigator tells me!
A5. When I get that fuzzy feeling all over!
A6. (Blank – Provide your own answer)!

Now I could be wrong and Bob please correct me if I am.

Metrics – what are your measurements (friction ridge detail, type, relationship, rarity, amount) that quantify (determine) your results (conclusions).

[16_Points]

Lots of great comments. But I stand by my statements. If "friction ridge skin impression identification" is based on a quantitative and qualitative assessment of detail (any detail), then by definition you must have some quantitative measure of L2 & L3 details (even if its not a number though I usually put a fat zero next to L3 in most cases). For those of you that don't "count" even in an abstract fashion, how do you fulfill the "quantitative" portion of Quantitative-Qualitative Assessment (2QA)?

And while everyone's experience with latent prints is different, I have found that a majority of identifications can be justified based on the quantity of minutiae alone (which does NOT mean there is no additional information that supports the identification such as L3 detail).

If this is true, then a probability model based on minutiae alone will be helpful in a majority of comparisons which result in an identification. Perhaps not all comparisons all the time, but a majority of them certainly.

[RL Tavernaro]

I'll try to translate to a more easily understood format. How about:

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

TRANSLATED: How do you measure the amount of information (matching features including 1st, 2nd, and 3rd level detail) that you find as you do a fingerprint comparison? And do you have a standard method of 'weighting' the different types of features present in the prints being compared that can be reliably repeated?

Does that help? I think both Neville and Charles were on the right track.

I take each print on its own merit, weigh up all the available information (1st, 2nd and 3rd level) then evaluate whether there is sufficient..... as simple as that !

I believe that you've given an overview of your process. What I'm interested in, is what kind of objective measurements you use in the 'weigh up' and evaluation portion, and then how do you use those results to determine sufficiency.
Regards, RLT

[Neville]

How do you measure the amount of information (matching features including 1st, 2nd, and 3rd level detail) that you find as you do a fingerprint comparison?
A4. When the investigator tells me! Then I know it is not an ident! Bad answer, in theory I don't; in practice I keep a running total in my head. When I get to twelve it is off to court no worries. However that is not to say I do not id it way before that I just don't get excited. But then I have been accused of being very creative in my idents and I guess that is the artistic streak coming out in me, I know something is happening in there as 3 ridges go in and 4 come out.

.. do you have a standard method of 'weighting' the different types of features present in the prints being compared that can be reliably repeated?
A2. My training and experience tells me when I have one!

Now that is being honest, thats me and is no reflection on what others do here in NZ. I personally would rather see a criminal go free than make a bad ident, no matter what the offence or what pressure is applied and by whom.

Sorry Charles I guess it is a bit like me being asked by someone in your nick of the woods is New Zealand a State of Australia or in Europe and that is the intelligent ones.

[JohnT]

Given that the quote function won't work on my PC I will not refer specifically to any of the previous posts and rather try and give a general statement on my views re measuring, Quantitative & Qualitative and ojective criteria for how much is enough.

I consider fingerprint identification to fall into the area of physical fit evidence. If a piece of broken glass 'fits' into the pieces left in the window, it came fom that window. A LR ratio is not helpful at that point - it just fits. The challenge we have with ridge detail are the effects of skin flexibility and distortion in making contact with a surface. These effects need to be understood and accounted for as we attempt to 'visually & mentally' test whether or not we have a physical fit, or rather agreement of friction ridge features.

How much is enough? Counting points is not the same as having a quantity of features. Reaching a number such 7, 10 or 12 does not magically make a latent print identical to the print from a known source. Something I learnt from Pat Wertheim a number of years ago sums up how I know when I have sufficient ridge features in agreement to make an identification - Reliable Predictability - I go from seeing a feature in the latent and checking to see if it is in the same relative position in the exemplar, to seeing a feature in the latent and predicting where it will be in the exemplar. At that point I have enough, but I still couldn't tell you how many, because I wasn't counting, I was looking for a physical fit. I certainly wouldn't go back and then count the points and write them down, because not only is that number meaningless to me, but it can potentially create the wrong meaning to the lay person. Clearly this process is subjective, but it is a qualified subjective approach. It is certainly testable and repeatable.

I was once asked how certain am I really when I reach a conclusion. My brief analogy was this; when you are doing a jigsaw puzzle and are finished except for the last piece and you are about to place that last piece in the puzzle, how certain are you that it fits and is correct. I was answered, absolutely certain. That is the feeling I have when I make an identification. And I also have no doubt that many on this board would suggest that is overly confident and statistics would show that we should not have that degree of confidence in our identifications. There in lies the whole argument.

I am fascinated by the concept from DNA scientists, statisticians and similar that if I had only 2 people in a room and asked one to touch a piece of glass and then compared this partial latent with their respective exemplar prints, that if I identified the first person I checked, to complete the process, I would need to also check the second individual. From my perspective by identifying the source of the latent there is no need to check the second person, or billionth person for that matter.

I have individualised the latent print.

[JohnT]

And I do apologise for being so long winded - it is Monday here.

[Neville]

Hi John
your last paragraph sums it up 100%.

[16_Points]

JohntT, you are correct by saying that the number of minutiae doesn't necessarily tell you anything useful. But sometimes it does. If there discernible 20 minutiae in a latent print, then that certainly tells you something. If I make an ID and state that the crime scene mark and known exemplar have 20 matching minutiae, I suspect most examiners would find the conclusion credible based on (the accuracy of) that statement. Whereas, if I state that I made an identification based on 4 minutiae, I suspect that most examiners would want to know what else I had (level 3 etc.). Minutiae counts become meaningless when you have an intermediate value (e.g. 8-12 minutiae) in which cases the ancillary components (such as qualitative factors) become very important.

More important than that, counting minutiae fails because minutiae have a weighted value based on their rarity. For example, an ending ridge does not have the same evidential value as a trifurcation. I believe most examiners would agree. It just so happens that the statistical models currently being tested are capable of assigning a weighted value to minutiae.

[Neville]

Hi 16 points,
one problem I immediately see is that ident that the SCRO made, didn't some of them suggest that there was over 20 points.

I am not agreeing nor disagreeing with you, I am just making the point, so to speak.

Also agree the Trifurcations and spurs are less common

[16_Points]

Hi 16 points,
one problem I immediately see is that ident that the SCRO made, didn't some of them suggest that there was over 20 points.

I am not agreeing nor disagreeing with you, I am just making the point, so to speak.

Also agree the Trifurcations and spurs are less common

Sure, you're absolutely right. I've seen non-matching prints with 14, 15, even 16 minutiae that agree very well with no visible disagreements. These are rare occurrences, but they do happen. I have not seen the SCRO prints. As I put in the parenthetical, claiming 20 minutiae in agreement is only meaningful if the examiner's observations are accurate.

[Taggart]

16_points,

I am fascinated by your latest posting where you stated:

I've seen non-matching prints with 14, 15, even 16 minutiae that agree very well with no visible disagreements. These are rare occurrences, but they do happen.

I wonder if I can ask what may be a stupid question? In the prints where you have seen up to 16 minutiae that 'agree very well' with no minutiae in ‘visible disagreement’ on what basis were you or the experts concerned able to positively exclude that individual as the donor? Was there any other evidence which excluded that individual perhaps? Or was it down to further examination of the fingerprints? If so what was it, given it is not further minutae in disagreement?

I could understand this if you were able to find visible disagreements, but am interested how an expert could exclude an individual if there were 'no visible disagreements'?

And for accuracy it was Peter Swann who stated he found 21 characteristics in agreement in print Y7 with none in disagreement.

[Neville]

Thanks for putting me right on that Tagget, I must admit I have not bothered much with those blogs as it got too personal.
16 points are occurances where the direction of the minutiae is not accounted for, as I have never seen this.

[16_Points]

16_points,

I am fascinated by your latest posting where you stated:

I've seen non-matching prints with 14, 15, even 16 minutiae that agree very well with no visible disagreements. These are rare occurrences, but they do happen.

I wonder if I can ask what may be a stupid question? In the prints where you have seen up to 16 minutiae that 'agree very well' with no minutiae in ‘visible disagreement’ on what basis were you or the experts concerned able to positively exclude that individual as the donor? Was there any other evidence which excluded that individual perhaps? Or was it down to further examination of the fingerprints? If so what was it, given it is not further minutae in disagreement?

I could understand this if you were able to find visible disagreements, but am interested how an expert could exclude an individual if there were 'no visible disagreements'?

And for accuracy it was Peter Swann who stated he found 21 characteristics in agreement in print Y7 with none in disagreement.

Actually, you raise an excellent point! I suspect most examiners would have made the identification with as few as 14 points and moderate quality. I knew that it wasn't an identification because the questioned print was actually an experimental control of known origin. Hence we know with 100% certainty that it is an exclusion. The exemplar was obtained via an AFIS search optimized to produce a close non-matching candidate (the actual AFIS search was conducted by a distinguished colleague who showed me the results at an IAI lecture).

This is the difference between the knowledge gained via casework and the knowledge gained via controlled experimentation. Furthermore, it would be very interesting to see the likelihood ratio generated by a 14 point close non-match vis-a-vis other 14 point matches. Everyone would learn something useful.