Alcohol and Tobacco Use

Limitations of Survey Studies
by National Institute of Health

(Page 3 of 3)

When discussing epidemiologic survey estimates, such as the ones presented in this article, researchers must consider several study limitations. First and foremost among these limitations is the generally consistent observation that self-reported alcohol and tobacco consumption tends to understate consumption levels when compared with other data sources (sales-tax revenues and data on production or supply). Second, the estimates for alcohol and tobacco dependence are based on a lay diagnostic survey method, rather than a stringent psychiatric diagnostic evaluation, and use diagnostic criteria that are not perfect in their validity. Consequently, people who are willing to report symptoms of tobacco dependence might also be more likely to disclose information confirming their symptoms of alcohol dependence and vice versa. As a result, the prevalence estimates of concurrent alcohol and tobacco dependence may be overstated compared with the prevalence estimates for dependence on either drug alone. This limitation, also called "shared methods covariation," merits greater attention in future studies on alcohol and tobacco use and on co-occurring alcohol and tobacco dependence, particularly in light of the ever-growing social disapproval of tobacco smoking.

Survey Methods

The National Household Survey on Drug Abuse

The National Household Survey on Drug Abuse is currently conducted annually by the Research Triangle Institute under contract with the United States Substance Abuse and Mental Health Services Administration, Office of Applied Studies. The survey methods used have been described in detail in each year's published report and in the scientific literature. In brief, the investigators in recent years have used a complex sampling procedure, called a multistage area probability sampling procedure, to draw each year's sample of noninstitutionalized U.S. residents age 12 and older. This procedure generates a representative sample, but also allows for oversampling of some population subgroups in order to yield a sufficient number of respondents from those groups for valid statistical analyses. This complex survey design necessitates attention to variation in sample selection probabilities and to survey design effects.

Of the designated survey respondents, approximately 75 to 82 percent typically agree to participate. The estimates derived from the survey are statistically modified (weighted) using certain procedures that help take the varying sampling weights and the nonrespondents into account. These procedures generally assume that people who do not respond to the survey are similar in their characteristics (drinking and smoking behaviors) to the people who respond. This assumption, however, may not always be correct.

The NHSDA methods used for eliciting selfreports about licit and illicit drug use have been designed to promote self-disclosure of potentially sensitive and illegal behaviors, such as the use of controlled substances (heroin and cocaine). For example, the respondents mark their answers on special answer sheets that are not seen by the interviewers. In recent years, the survey has introduced an audio, computer-assisted self-interview procedure in which the respondents listen to the questions via audio headphones and reply by entering the response on a laptop computer.

Following the data collection, an initial public report is released and the data are prepared to ensure adequate levels of respondent confidentiality when they are analyzed. Subsequently (typically 2 to 3 years after the data have been gathered) a public-use data file is released and made available to researchers for analysis. Throughout the preparation of the public-use data files, standard procedures for survey quality control and quality assurance are implemented. Moreover, additional information (primary sampling unit and strata designators) is supplied with the publicuse data files to aid in the statistical analysis.

At the time of preparation of analyses for this article, data from the 1995 to 1997 surveys were available in the public-use data files. The sample sizes for these years were 17,747 respondents in 1995; 18,269 respondents in 1996; and 24,505 respondents in 1997. In this article, the authors chose to show estimates from all 3 survey years to display year-to-year variation. The figures display point estimates from each survey, based on a generalized linear model and smoothing functions designed to generate less variable patterns across age groups.

The National Comorbidity Survey (NCS)

The University of Michigan Institute of Survey

Research conducted the NCS with primary sponsorship by the National Institute of Mental Health and supplementation by the National Institute on Drug Abuse. Numerous scientific articles have described the NCS survey procedures in detail.

As with the NHSDA, the NCS survey procedures were designed to draw a nationally representative sample of noninstitutionalized U.S. residents, although with a somewhat narrower age range. Thus, the NCS included 8,098 respondents ages 15 to 54. The NCS used the NHSDA procedures to elicit self-reports about alcohol and other drug use with the addition of a special supplement to elicit reports on daily tobacco use and tobacco dependence. The investigators then used standardized self-report diagnostic procedures to assign respondents to one or more categories of mental disorders as defined in the American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders. To generate weighted estimates and 95-percent confidence intervals, the researchers used variance estimation procedures that thoroughly considered the survey sample selection probabilities and complex survey design. For clarity, however, the figures in this article display point estimates.

The NCS did not estimate the number of tobacco smokers. Therefore, NHSDA estimates of the number of tobacco smokers have been used to prepare the figures in this article.

Conclusion

Despite the limitations mentioned in the previous section, the epidemiologic evidence regarding the co-occurring use of alcohol and tobacco is consistent with other clinical, biobehavioral, and neuroscientific evidence. Thus, the evidence points toward patterns of co-occurring use and co-occurring dependence syndromes that deserve greater scrutiny in all of these research domains.

The patterns of co-occurring alcohol and tobacco use in the population motivate a continued search for mechanisms underlying both types of drug use, including mechanisms of reciprocal process, in which alcohol use promotes continued smoking and smoking promotes continued drinking. Similarly, the patterns of co-occurring alcohol and tobacco dependence prompt investigations into underlying shared genetic and neurobiological vulnerabilities to both forms of dependence as well as social and experiential processes that result from or lead to sustained consumption of these drugs.

About the Author

NIH is the nation's medical research agency - making important medical discoveries that improve health and save lives. The National Institutes of Health (NIH), a part of the U.S. Department of Health and Human Services, is the primary Federal agency for conducting and supporting medical research.