unrelated to illegal drug use among its students. The principal investigators of MTF argue that this is the case (Johnston et al., 1998), but there is little empirical evidence to support (or refute) their claim. Perhaps the investigators are right, in which case school nonresponse does not affect identification of trends from MTF data. Alternatively, nonrespondent schools may have very different prevalence levels and trends, in which case MTF cannot be used to identify trends. We simply do not know.

Also troubling is the method of replacing nonrespondent schools. The replacement schools are not found using the original sampling scheme, but instead are purposefully selected to be “similar” to schools that decline to participate. If there are unobserved differences between the replacement and the dropout schools, the sampling properties are compromised. Replacing schools does not solve the nonresponse problem.

In the committee’s judgment, alternatives to the current MTF sample design warrant serious consideration as means to more effectively collect information on teenage drug use. A household survey of teenagers similar in design to the National Longitudinal Survey of Youth or the NHSDA would almost certainly achieve higher response rates and would, moreover, cover school dropouts. Or the present school-based design of MTF might be retained, but the manner of selecting respondent schools changed to enhance response rates. At a minimum, the survey design should explicitly account for school nonresponse, with any replacement schemes formally considered to be part of the sample design.

Inaccurate Response. Self-report surveys on deviant behavior invariably yield some false reports. Respondents concerned about the legality of their behavior may falsely deny consuming illegal drugs. Desires to fit into a deviant culture may lead some respondents to falsely claim to consume illegal drugs. Thus, despite considerable resources devoted to reducing misreporting in the national drug use surveys, invalid response remains an inherent concern.

This measurement problem is conceptually different from the nonresponse problem. The fraction of nonrespondents is known, but the fraction of respondents who give invalid responses is not. Thus, the methods used to investigate the effects of nonresponse are not applicable to incorrect response. Rather, one must obtain information or make assumptions about self-reporting errors.

There is a large literature that provides evidence on the magnitude of misreporting in some self-reported drug use surveys. Validation studies have been conducted on arrestees (Harrison 1992, 1997; Mieczkowski, 1990), addicts in treatment programs (Darke, 1998; Magura et al., 1987, 1992; Morral et al., 2000; Kilpatrick et al., 2000), employees (Cook et al.,



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