Saturday, June 11, 2011

The Inspector General's Follow-up Report on the FBI Latent Fingerprint Unit

The Department of Justice's internal watchdog issued a 53-page report on the FBI's progress in meeting recommendations made in the wake of the infamous Mayfield misidentification. Here are the bottom lines (literally):
In conclusion, we believe that the FBI Laboratory has made significant progress in implementing many of the recommendations from our Mayfield Report. Changes to the FBI Laboratory’s procedures and training materials made in the course of this review, such as providing further guidance on the use of Level 3 detail and improving documentation of explanations for differences, represent substantial steps toward reducing the likelihood of errors. We encourage the FBI Laboratory to continue funding and conducting research aimed at validating latent fingerprint analysis and at creating objective criteria for determining sufficiency and declaring an identification. We also strongly support the efforts by the FBI and the CCU to identify the remaining capital cases in which the FBI Laboratory conducted latent fingerprint analysis before December 31, 2006, and to complete the Capital Case Review. We believe that the FBI Laboratory must continue to address these issues to avoid future errors, improve the reliability and accuracy of the latent fingerprint discipline as a whole, and ensure public confidence in the results of its examinations. We believe that full implementation of our recommendations will help the FBI Laboratory in this effort.
The new report is generally thorough, thoughtful, and informative. However, it has a few quirks.

I. Federal Courts?

A very minor but surprising mistake is the statement that "In at least two federal district court cases, however, courts have limited or excluded testimony about latent fingerprint identification based in part on the Mayfield error" (Page 20 n.22). The two cases cited are United States v. Zajac, No. 2:06-cr-00811 (D. Utah Sept. 13 and 16, 2010), and Maryland v. Rose, Case No. K06-0545, Mem. Op. at 5-9, 24-25, 28-31 (Balt. Co. Cir. Ct. Oct. 19, 2007). Obviously, a Baltimore County trial judge is not a federal district judge. As I recall, after the state court threw out the fingerprint identification in the state case, the US Attorney's Office made it into a federal case and got the U.S. District Court for the District of Maryland to write an opinion on the admissibility of the analyst's testimony. The OIG's footnote ends: "But see United States v. Rose, 672 F. Supp. 2d 723, 725-26 (D. Md. 2009) (rejecting argument that the Mayfield error called into question the reliability of the ACE-V methodology)."

So as far as I know, there is only one "federal district court case[ to] have limited or excluded testimony about latent fingerprint identification based in part on the Mayfield error." That is Zajac. This case might have led to more such rulings, but I should think that the Noblis-FBI experiment (described in earlier postings) will inhibit any such trend.

II. Linear Thinking?

The new report also states that "[u]nlike some forensic laboratories, the FBI Laboratory uses a “linear” approach to ACE-V, requiring examiners to complete and document their analysis of a latent fingerprint before viewing any known fingerprints or moving to the comparison and evaluation phases" (Page 5). The OIG adds that under new FBI operating procedures:
[E]xaminers must complete and document analysis of the latent fingerprint before looking at any known fingerprint; examiners must separately document any data relied upon during comparison or evaluation that differs from the information observed during analysis; and verifiers or blind verifiers must separately complete and document their ACE examination. The FBI Laboratory refers to this approach as “linear” ACE-V, and Unit Chiefs Soltis and Wieners stated that it eliminates circular reasoning and “cherrypicking” by requiring examiners to identify the characteristics present in a latent fingerprint during analysis before moving to the comparison phase ... .
(Page 27). In its entirety, however, the FBI's examination process is not linear -- at least not unidirectionally linear. It is "oscillatory" -- it goes back and forth. After studying an exemplar, an analyst can return to the latent print and use features not documented the first time through in making the comparison. Indeed, the examiner can go back and forth as many times as desired.

This is not to condemn what some analysts call "recurring examining." If the process is fully documented, the value of using initially undiscovered information could outweigh the risk of bias. The point is that one must view the post hoc comparisons more skeptically.

Neither is the oscillation a form of circular logic. The report defines "circular reasoning" as "the use of data from the known fingerprint to influence the characteristics observed in the latent fingerprint." (Id.) However, it is not necessarily "circular" to use a feature in an exemplar as a clue to what is present in a latent print. Thus, I am not so sure that oscillating between the prints inherently "is a form of confirmation bias or 'mindset' that can lead to unintentional false identifications." (Id.) It would seem more precise to say that the use of the data can result in confirmation bias, which can lead to error.

This does not mean that oscillation is clearly a desirable protocol. Note 28 explains that:
In the Mayfield error, for example, the original examiner encoded seven Level 2 details in the latent fingerprint before being exposed to any candidate fingerprints. After running an IAFIS search and viewing Mayfield’s fingerprint, the examiner changed his interpretation of five of these seven points. Additionally, similarities between the Madrid latent fingerprint and Mayfield’s known fingerprint led the examiner to see other similarities that were not actually present.

The question, as indicated earlier, is whether the benefit of oscillation outweighs the danger.

III. What's independent?

My final nit is the use of "independent" to describe a verification in which the verifier knows the result reached the first time around. To call this "independent," as the FBI does (p. 11), is misleading. If I solve a math after you give me the answer to help me along, my work is not independent. It is more independent that it would have been had you given me a full, written solution that I could have copied, but it is not truly independent.


U.S. Dep't of Justice, Office of the Inspector General, A Review of the FBI’s Progress in Responding to the Recommendations in the Office of the Inspector General Report on the Fingerprint Misidentification in the Brandon Mayfield Case (2011), available at

No comments:

Post a Comment