later, the prevalence of marijuana use and cocaine among people on active duty was substantially lower than it was among their classmates. Although it is possible that the self-reports by active-duty personnel are less trustworthy than those of other respondents, the data are suggestive of a deterrent effect and also tend to show, somewhat surprisingly, that there was no substantial self-selection (i.e., high school drug users were no less likely than their peers to enlist in the military), even though such a self-selection effect was evident for smokers after the military toughened its smoking policies in the late 1980s.

The second study, conducted by Mehay and Pacula (1999), used data from the National Household Survey of Drug Abuse and the Department of Defense’s Worldwide Survey of Health Related Behaviors to compare military and civilian populations before and after adoption of the military’s zero-tolerance policy in 1981. They concluded that, in 1995, military employees were about 16 percent less likely to report using drugs during the past year as their civilian counterparts, and that very little of this difference appears to be attributable to self-selection bias. Noting, however, that selection bias might more heavily influence younger age groups, they estimated that the deterrent effect might be as low as 4 percent.

Mehay and Pecula observed that a zero-tolerance, frequent testing protocol, similar to the military approach, probably would not be cost-effective in the civilian sector, but they suggest that a less frequent testing policy, with a more lenient second-chance sanction, might yield a 10 percent deterrent and “can be expected to reduce drug use in a cost-effective manner.” In the committee’s view, however, extrapolating in this way from the military experience to the civilian sector is questionable. Even if the prospect of being subjected to drug testing does not affect drug users’ initial decisions to seek employment with a particular firm, employees have an opportunity for exit not available to people on active duty in the military when they find out about a firm’s testing program. As a result, self-selection is likely to be a much more substantial factor in civilian employment than in the military setting, especially when employment opportunities without drug testing are plentiful.

Until recently, nationally representative data that could be used to examine the association between drug use and drug testing outside the military have not been collected (National Research Council, 1994). However, the 1994 National Household Survey of Drug Abuse included a special workplace module that contained a series of questions about workplace drug testing programs. It therefore became possible to ascertain the