Tuesday, June 12, 2012

Who Is Nelson Acosta-Roque? (Part III)

As we have seen, the fingerprint analyst who testified about a match between the two ten-print cards from the man named Acosta-Roque and the two from the man named Pecheca-Aromboles said next to nothing about her examination in the particular case. Anthony Capece, Assistant Chief Counsel for DHS, only asked only the following, very general questions, about the methodology:
Q. Okay now, how do you — if you can answer this, how do you actually identify someone by means of fingerprints?
A. I have two impressions that I'm comparing, I use — I personally use two fingerprint magnifiers and two pointers is what we call them, and I'll have a glass over each impression, and I use the pointer to what I call walk through the ridges and count my points.
Q. Okay now, is there a set number of points that needs to compare in order for you to be able to say that this is the same person?
A. There is no established number. I normally would not want to sign off on anything with less than eight, but there is no set established policy for minimum.
Q. Okay, is there an amount that's required by criminal courts?
A. No, there's not.
Q. Okay, so eight would be acceptable in criminal courts then?
A. Yes. Yes, if you sign off on it, you're saying that that's the same person.
It is impossible to discern from the answers to these poorly framed questions how complete the comparisons among the four cards were. They could have been exceedingly thorough or they could have been disturbingly cursory. The fingerprint analyst, Susan Blei, has yet to respond to emails seeking clarification

The amicus brief (2012) of "scientists and scholars" offers the following guidance:
Although the record is not altogether clear on this point, it is possible that the examiner based this conclusion on eight corresponding ridge details from a single finger. Amici are aware of no research that would support concluding from eight corresponding details that “this is the same person.” Indeed, recent research suggests that while some comparisons showing eight corresponding characteristics may well have significant probative value, the probability of observing such number of characteristics in agreement between fingerprints from unrelated individuals is not negligible.28 The scientific data collected to date clearly support the general value of fingerprint characteristics for constituting very strong evidence in favor of a same-origin hypothesis, but are in significant contradiction with the claims made by the expert during her testimony.
I wonder if, like the examiner’s testimony, this description of "scientific data" is subject to a charge of overclaiming. Just what do the data show about “the probability of observing such number of characteristics in agreement between fingerprints from unrelated individuals” and the probative value of most eight-point matches? Indeed, how would one measure the probability of a false eight-point match and the probative value of an eight-point match?

An experiment could investigate how often fingerprint examiners detect eight (or more) matching points among impressions of the same fingers (mates) as opposed to different fingers (non-mates). This experiment could provide an estimate of the probability of an eight-point match for different fingers and show that it "is not negligible” (if that is the case). Moreover, if eight-point matches turned out to be not much more common among pairs of impressions from the same finger than among pairs from different fingers, an eight-point match would not give strong support to the hypothesis that the marks come from mates. Thus, the ratio of the probability of a match with a mate to the probability with a non-mate—a quantity known as the “likelihood ratio” (LR)—could tell us whether the probative value of such matches is high or low.

The amicus brief cites no data on eight-point matches of human examiners. The footnote accompanying the claim that the probability of eight-point marches from different fingers is “not negligible” cites three studies. The researchers computed quantities akin to a likelihood ratio according to an algorithm and model of distortion in impressions for different numbers of points (even if a human would not call them matches). This is great research, but it does not investigate the properties of an eight-point matching rule as employed by fingerprint examiners—at least not directly.

Still, it might be suggestive, which is all that the brief claims. So what do the papers reveal about the probabilities for eight-point non-mates versus eight-point mates? Are only “some” likelihood ratios large? Are probabilities of non-mate matches “not negligible”? The first paper (Neumann et al. 2006) is confined to three minutiae. Hence, it is largely beside the point.

The second paper (Neumann et al. 2007) reports that computed LRs with 8 minutiae rarely exceeded 1 for prints from different sources (non-mates). The percentages of LRs in this range in different types of patterns are 1.2%, 0.5% 0.3%, 0.1% (Table 3). In contrast, the LRs exceeded 1 for prints from the same fingers (mates) 99.6% of the time (Figure 11). Of course, an LR only slightly greater than 1 has little probative value, but the LRs for mates exceed 10,000 nearly 90% of the time, judging from Figure 11.

In the third paper (Neumann et al. 2012), the separation in the distributions of the LRs for eight minutiae as between mates and non-mates was even sharper. The LRs almost always exceeded 1,000 for mates, almost never reached this level for non-mates, and never reached 10,000 for non-mates. The text accompanying Figure 3 refers to only a “few occasions when LR(Hd) [the LR for non-mates] exceed[ed] 1" and suggests that even this small number is inflated “when it is borne in mind that the prints have been retrieved as the closest among 600 million.” Later, the paper states that “very few instances of LR(Hd) > 1 were observed” and “the probability of misleading evidence ... is very low.” Moreover, the “wider range of features [used] by examiners would make their examination more discriminating.”

Thus, the research seems more supportive of the value of an eight-point match than the brief's suggestion that significant probative value is merely a sometimes thing for such a match. “Significant probative value” seems to characterize not just “some comparisons” but the vast majority of them. Likewise, the research indicates that different fingers match at eight points at a low rate. There could be some false eight-point matches, but this hardly means that an eight-point matching rule would produce a major proportion of false positives. (And even a small proportion would be of concern.)

So is it true that “[t]he scientific data collected to date are in significant contradiction with the claims made by the expert during her testimony”?  Well, the answer is both yes and no. The research actually seems to support the examiner’s good feelings about eight matching minutiae when making a positive identification. At the same time, the studies do limit how positive a knowledgeable examiner can be about an eight-point identification. Two of the studies suggest that the examiner’s degree of confidence in the claim that no one else in the world has a finger that could produce eight or more similar minutiae was unfounded.

Of course, for the reference in the testimony to eight points of comparison to have much bearing on the specific identification in issue, the principal examiner would have had to have made her identifications among four ten-print cards using only one of the ten fingers and stopping upon finding eight corresponding points in the impressions of that one finger. In addition, the second verifying examiner also would have had to have used only eight features—and the very same ones at that. The amicus brief treats this as a possibility worth considering because the record does not “affirmatively” show otherwise. At the same time, the brief recognizes the distinct possibility “that the examiner relied upon a much greater number of corresponding details than the figure of eight mentioned in the record” (but never mentioned as having been used in the case itself).

The brief's response to this obvious possibility is curious. The scientists and scholars state that “[i]f this is the case, however, there is no information available regarding how the forensic expert witness adjusted her assessment of the probative value of the evidence in light of this greater amount of discriminating information.”

But why would an expert who believes that there is zero probability of a false match for 8 points have to “adjust her assessment” when there are, say, 32 points? Suppose I assert that (1) the probability that all the air molecules around my mouth and nose will happen to move away and stay away for one second is essentially zero, and (2) the chance of a temporary vacuum lasting two seconds also is virtually zero. The second statement is neither false nor misleading because of the first statement. Similarly, it makes little sense to demand that a fingerprint examiner who would give an opinion that the chance of false match is negligible for eight features of the sort seen here has to “adjust[] her assessment” when there are more such features. Naturally, her confidence would increase with more data pointing to the same conclusion. Once again, the real problem with the testimony is not the examiner’s personal rule of thumb. The problem lies with the protestation of 100% confidence in the first place.

What can an examiner who is sensitive to the growing base of scientific research into her craft do to explain the striking similarities in a pair of fingerprints? Here is a hypothetical dialogue:
Q. What did you find when you compared the prints?
A. I observed that the overall patterns and the minutiae were very similar. The prints had the same general appearance and the same detailed features.
Q. And what does that signify?
A. Well, all these similarities strongly support the conclusion that the marks come from the same finger rather than from two different fingers.
Q. How strongly do they support the fact they are from the same finger?
A. I cannot give you a simple number, but patterns like these are much more likely to arise from prints of one and the same finger than from impressions of different fingers.
Q. How certain are you that they are from the same finger?
A. I did not say they came from the same finger. All I can tell you is that they have the definite appearance of coming from the same finger. They do not look like a pair of marks coming from different fingers.
Q. How certain are you of this?
A. I am very, very certain

Brief of Scientists and Scholars of Fingerprint Identification as Amici Curiae in Support of Petitioner and in Favor of Reversal, Acosta-Roque v. Holder, No. 11-70705 (9th Cir., Mar. 8, 2012).

Cédric Neumann et al., Computation of Likelihood Ratios in Fingerprint Identification for Configurations of Three Minutiae, 51 J. Forensic Sci. 1 (2006).

Cédric Neumann et al., Computation of Likelihood Ratios in Fingerprint Identification for Configurations of Any Number of Minutiæ. 52 J. Forensic Sciences 54 (2007)

Cédric Neumann et al., Quantifying the Weight of Evidence from a Forensic Fingerprint Comparison: A New Paradigm, 175 J. Royal Stat. Soc’y: Series A 371 (2012).

No comments:

Post a Comment