Error Rates for Latent Fingerprinting as a Function of Visual Complexity and Cognitive Difficulty

[josher89]

On page 14 of this document, that's a right loop, correct?

[Steve Everist]

Looks like it to me...

https://www.ncjrs.gov/pdffiles1/nij/grants/249890.pdf

Capture.PNG

[Dr. Borracho]

Surely this must be the assessment of a REAL scientist, not some subjective latent print examiner.

[josher89]

On page 51 (last paragraph), they also indicated that a print used in the study was misclassified by 91 percent of the examiners evaluating it. I wonder if this is the print and they claim it is a left loop while the 91% claimed it was a (correctly so) right loop. What does that mean about the 9% that still classified it incorrectly?

If a typo, then okay.

Right above that statement they make a claim that it can be misleading or at least limited to talk about overall error rates for the field as a whole. I do agree with this as we know there are several types of errors. This study was limited (of course, like all studies) in that they used the images of the 2011 IAI participants and those were timed examinations. It's really hard to draw definitive conclusions about complexity and difficulty when the examinations are timed and they are forced to use same source or different source. Haber says inconclusives are errors and I would disagree with him largely (not wholly).

But it's Friday. I'll read it better when I have a clear head on Monday.

[Dr. Borracho]

. . . they also indicated that a print used in the study was misclassified by 91 percent of the examiners evaluating it. . .

I wonder if, in reference to classification, they meant ridge count, as in old Henry or NCIC classification formulae. That would depend on consistent application of delta and core rules, but even then two examiners with different cards could be a count or two different due to inking and pressure. If they were looking for exact count, conceivably 91% could be one or two ridge counts off from the author's ridge count. But the old formulae allow plus or minus three. So from the perspective of a fingerprint classifier, you could still have everybody consistent within the plus or minus three ridge counts.

It all depends on what the authors of the article meant.

[Steve Everist]

. . . they also indicated that a print used in the study was misclassified by 91 percent of the examiners evaluating it. . .

I wonder if, in reference to classification, they meant ridge count, as in old Henry or NCIC classification formulae. That would depend on consistent application of delta and core rules, but even then two examiners with different cards could be a count or two different due to inking and pressure. If they were looking for exact count, conceivably 91% could be one or two ridge counts off from the author's ridge count. But the old formulae allow plus or minus three. So from the perspective of a fingerprint classifier, you could still have everybody consistent within the plus or minus three ridge counts.

It all depends on what the authors of the article meant.

It looks like this result came from Experiment 1: Experts at a Conference, and it was a timed comparison test (I wasn't at the 2011 Milwaukee conference, but they describe the process thoroughly). The term classification wasn't used in the historical sense. From the results section, it looks like they were using the word classification to describe the match/no match result given by the people in the test.

[NRivera]

My tax dollars at work

[ER]

This study is easy to dismiss, but there are some good things about it.

Good things:
- "well-trained, experienced examiners are highly accurate when making positive identifications"
(whatever a positive identification is )
- The best predictors of accuracy, difficulty, comparison time, and examiner confidence are the presence of cores and deltas. And since most of the errors were false negatives, then relying on cores and deltas for exclusions are probably the best way to avoid false negatives.

Bad things:
- forced choice (no inconclusive)
- timed test
- at a conference
- on low res laptop screens
- all errors grouped together