sites for processing. For NIBIN partner agencies with low volume of entry of crime scene evidence, the ATF should continue to develop its procedures for reallocating NIBIN equipment to higher performance environments.

6–C
TECHNICAL ENHANCEMENTS

Several of them deal with the specific functionality and interface of the current IBIS platform; others are broader in scope and speak to the type of information that should be recorded for the NIBIN system as a whole. Put another way, these recommendations are not a “to do” list for the current IBIS or its developers, but will require collaboration between system developers, NIBIN management, and the program’s user base.

A common theme of our technical recommendations extends from our general assessment of the IBIS platform in Section 4–F: that it is a sorter and a tool for search that is commonly, and unfortunately, confused with a vehicle for verification; the two are very different functions. The recommendations we offer are meant to improve the system’s effectiveness as an engine to search and process large volumes of data and to give its users more flexibility to explore possible connections between cases.

6–C.1
The Language of “Correlation”

We begin with a matter that is inherently technical, even though it does not deal directly with computer hardware or software: It is an issue of nomenclature, of what to call the basic process performed by the IBIS technology. As described in Chapter 4, Forensic Technology WAI, Inc., and the IBIS user base describe the process as “correlation,” even though system training materials repeatedly stress that the actual correlation “scores” are of little consequence and that what matters is the rank of particular exhibits. We avoid using “correlation” throughout this report, describing the algorithm and process as “comparison” instead. In statistics, and as has seeped into common parlance, the correlation coefficient measures the strength of linear association between two random variables. Scaled to fall between 0 (no relationship) and 1 (perfect linear relationship), the correlation coefficient provides a clear and easy to understand measure of association. That IBIS uses the same term in labeling its scores imparts to the process—however subtly—an undue degree of quantitative confidence. This is not to say that the IBIS procedures are either unreliable or unsophisticated; indeed, we argue quite the opposite in Chapter 4.

To fully warrant the term correlation, the scores reported by ballistic imaging systems would have the same easily understood interpretation as a



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement