Wednesday, April 27, 2011

Part II of Fingerprinting under the Microscope: Examiners Studied and Methods

The Noblis-FBI study mentioned yesterday offers insights into the validity and reliability of the latent fingerprint examination process, but inferences about actual casework are necessarily limited by the design of the study. Before reaching conclusions about latent print examinations in general, one needs to consider the representativeness of (1) the sample of examiners studied, (2) the fingerprint pairs they examined, and (3) the conditions of the examinations.

The description that follows is modified and condensed from the more complete supplement to the PNAS study. My impression is that the examiners tested were better than average at their work and that they were motivated to do well in the exercise. On the other hand, they had unusually challenging pairs of prints to examine, and they had to conduct the examinations with a somewhat confining procedure.

1. Examiners Tested

By soliciting the examiners at the 2009 International Association for Identification (IAI) International Educational Conference, at SWGFAST, and by direct contact with various forensic organizations, the researchers obtained 169 volunteers. (Three sent in partial results that were excluded from the analyses. Employers encouraged or required some "volunteers" to participate.) Because the subjects were not randomly selected, generalizing the results to all latent print examiners (LPEs) would be dangerous. The "healthy volunteer" bias is well known in epidemiology, and the volunteers here are likely to be an above-average group. Based on the reports provided by 159 out the 169 participating LPEs, the group had the following characteristics:

AgeMedian 39
Mean 42
EducationCollege degree 50%
Graduate or professional degree 25%
EmployerFederal gov't 48%
State or local gov't 44%
Accredited 83%
Experience as LPELess than 5 yrs 21%
Less than 10 years 49%
trainees 4%
Certification Not certified 18%
Testified in courtNever 11%
Within past year 60%

2. Fingerprint pairs

Fingerprint impressions for the study were collected at the FBI Laboratory and at Noblis (from employees?). Latents and mated exemplars were selected from the resulting set of images to include a range of characteristics and quality and to be broadly representative of prints encountered in casework. But the exemplar data included a abnormally large proportion of poor-quality exemplars and the latents included many of relatively low quality, to evaluate the consensus among examiners in making value decisions.

The fingerprint data included 356 latents, from 165 distinct fingers of 21 distinct subjects; and 484 exemplars. These were combined to form 744 distinct image pairs, with each pair including one latent and one exemplar. There were 520 mated pairs and 224 non-mated pairs. The large proportion of mated pairs was intended to compensate for the higher proportion of poor quality latents among the mated pairs.

Prints to be compared were selected for difficulty. A subset, with disproportionately poor quality, of mated pairs were selected from the latents and exemplars. The non-mated pairs were designed to yield difficult comparisons. Unusually similar exemplars in the nonmates came from searches of the FBI’s Integrated Automated Fingerprint Identification Ssystem (IAFIS), which included exemplars from over 58 million persons with criminal records.

3. Conditions of the Examinations

Noblis developed software that presented latent and exemplar fingerprint images, provided limited image processing capabilities, and recorded test responses. Image processing capabilities were limited. The tests were distributed on DVDs to examiners who were given several weeks to complete the test. Each of the 169 examiners was initially assigned 100 image pairs.

They spent a median of 8 total hours on the test, over multiple sittings. Participants were not told what proportion of the image pairs were mated. Participants were instructed that it was imperative that they conduct their analyses and comparisons in this study with the same diligence that they would use in performing casework.

For each image pair, the latent was presented for analysis, and the examiner was asked if the image was of value for individualization (VIn); if the image was not VIn, the examiner was asked if the image was of value for exclusion only (VExO). If the image of the latent print was neither VIn nor VExO, the exemplar was not presented for comparison; otherwise, the exemplar was presented and the examiner was required to make a decision of individualization, exclusion, or inconclusive. Examiners were able to review and correct their responses before proceeding to the next comparison, but could not revisit previous comparisons, or skip comparisons and return to them later.

Stay tuned for the results.

No comments:

Post a Comment