The Credibility of Drug Tests: A Multi-Stage Bayesian Analysis

Abstract

The authors show that even when drug tests are extremely accurate by conventional measures, under some circumstances they will yield a high “false accusation rate” (that is, a high percentage of those testing positive for drugs will not have drugs in their systems). For example, if a drug-testing process that produces only one false positive per 2,000 drug-free specimens, and no false negatives, is administered to a population in which 0.1% of the people use the targeted drugs, one-third of those identified as drug users will be falsely accused. The authors propose a multi-stage Bayesian algorithm—an approach commonly used in management science but novel to industrial relations—that assures that a drug-testing process will have a low enough false accusation rate to provide credible evidence of drug use. They also identify other types of employee evaluations to which Bayesian modeling could be applied.