Journal Published Online: 08 December 2017
Volume 46, Issue 1

Bayesian Estimates and the Effectiveness of Metal Detection Devices



When a small number of observations is used to estimate the performance of a walk-through metal detector (WTMD), it is difficult to compare the observations made with different numbers of tests on different machines. Standard tests for metal detectors, whether hand-held (wand) or walk-through, are expressed in terms of specific test objects and a requirement [1], typically of the following form: “the object is to be detected at least 19 times in 20 trials.” Complementary requirements for suppressing false positives take various forms, depending on what is to be “not detected.” An example is the case of “no object at all,” for which the requirement may be “20 trials with no alarms.” Practitioners use much shorter tests to assess whether a machine is working on the day of the event, and they have no principled guidance about how to interpret that information, particularly if the information from each test is logged and thus might be aggregated to detect that a specific machine is slipping out of calibration. This article develops a principled Bayesian analysis that can convert a test of any size into a common language of expected odds ratio, which can then be used to compare tests with different sample sizes. Some technical issues of forming a “reasonable prior” are discussed. This research has the potential to turn quick checks, conducted in operational settings, into a rigorous and accumulated source of useful information about detection devices and procedures.

Author Information

Kantor, Paul B.
Department of Industrial and Systems Engineering, University of Wisconsin, Madison, WI
Pages: 5
Price: $25.00
Reprints and Permissions
Reprints and copyright permissions can be requested through the
Copyright Clearance Center
Stock #: JTE20170259
ISSN: 0090-3973
DOI: 10.1520/JTE20170259