An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Substance Abuse and Mental Health Services Administration . National Survey on Drug Use and Health: Summary of Methodological Studies, 1971–2014 [Internet]. Rockville (MD): Substance Abuse and Mental Health Services Administration (US); 2014 Nov.
CITATION: Biemer, P., & Wang, K. (2007). Using callback models to adjust for nonignorable nonresponse in face-to-face surveys. In Proceedings of the 2007 Joint Statistical Meetings, American Statistical Association, Survey Research Methods Section, Salt Lake City, UT (pp. 2889–2896). Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: In most sample surveys, weighting procedures attempt to compensate for nonresponse bias under the assumption of “ignorable nonresponse”; that is, the data that are known for both respondents and nonrespondents are sufficient to adequately adjust for nonresponse bias. For many surveys, call history data are available for all sample members, including nonrespondents, and because the level of effort (LOE) required to interview a sample member is likely to be highly correlated with response propensity, this method is ideally suited for modeling the nonignorable nonresponse. Biemer and Link (2007) provided a general method for nonresponse adjustment that relaxed the ignorable nonresponse assumption. Their method, which extended the ideas of Drew and Fuller (1980), used indicators of LOE based on call attempts to model the probability that an individual in the sample responds to the survey (referred to as the “response propensity”). Using data from the National Survey on Drug Use and Health (NSDUH), an annual in-person, cross-sectional study conducted in all 50 States and the District of Columbia to measure the prevalence and correlates of drug use in the U.S. population aged 12 or older, the authors investigated the feasibility to adjust for nonignorable nonresponse with callback models.
METHODS: The authors constructed several latent callback models consisting of four variables and compared them with an equivalent model that was constructed using the traditional logistic regression approach; that is, the model used was essentially the same the latent callback model with the special callback model features omitted.
RESULTS/CONCLUSIONS: Across a range of variables from NSDUH’s screener, the callback model showed improvement over the traditional model as hypothesized. Although the magnitude of the improvements was not dramatic, the results clearly showed that gains in accuracy were possible using the special features of the callback model in response propensity weighting.
CITATION: Frechtel, P., & Copello, E. (2007). Patterns of nonresponse for key questions in NSDUH and implications for imputation. In Proceedings of the 2007 Joint Statistical Meetings, American Statistical Association, Survey Research Methods Section, Salt Lake City, UT (pp. 3457–3464). Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: The idea of using “soft nonrespondents” to represent “hard nonrespondents” is not new to survey research. Callbacks are often used to adjust for nonresponse in surveys. The goal is to control nonresponse bias by assuming that the hard nonrespondents are more similar to the callback respondents than they are to the original respondents.
METHODS: The National Survey on Drug Use and Health (NSDUH), an annual nationwide survey involving approximately 70,000 subjects per year, does not make use of callbacks. However, for several key questions in NSDUH, follow-up questions, or “probes,” are presented to subjects who entered a response of “don’t know” or “refused” to the original questions. The probes are intended to increase item response rates by simulating an actual interviewer. The probe respondents can be considered soft nonrespondents, and the subjects who answer neither the original question nor the probe can be viewed as hard nonrespondents. An analysis from an earlier study was expanded to include data pooled from the 2000 to 2005 surveys. The values of auxiliary variables were compared between the original respondents, the probe respondents, and the nonrespondents to see whether the nonrespondents resembled the probe respondents more than the original respondents.
RESULTS/CONCLUSIONS: The probes offered a less costly alternative to callbacks for the mitigation of nonresponse bias. The response patterns suggested that, for illicit drugs, subjects who refused to respond to the original question but responded to the probes were more often lifetime users (and more often recent users) than subjects who responded to the original question. The presence of the probes seemed to be correcting for some of the bias, simply by adjusting the estimates relative to what they would be if the probes did not exist. However, the probes could be further used in imputation, which would enhance the adjustment for the nonresponse bias. A comparison of the predicted means of the different response patterns suggested that, at least for lifetime marijuana use and lifetime cocaine use, the imputation method was able to pick up some, but not all, of the difference between the original respondents and the probe respondents.
CITATION: Grucza, R. A., Abbacchi, A. M., Przybeck, T. R., & Gfroerer, J. C. (2007). Discrepancies in estimates of prevalence and correlates of substance use and disorders between two national surveys. Addiction, 102(4), 623–629. [PMC free article: PMC1924971] [PubMed: 17309538]
PURPOSE/OVERVIEW: The purpose of this research was to assess the degree to which methodological differences might influence estimates of prevalence and correlates of substance use and disorders by comparing results from two surveys administered to nationally representative samples in the United States.
METHODS: The authors conducted a post hoc comparison of data from the 2002 National Survey on Drug Use and Health (NSDUH) with data from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) administered in 2001 and 2002.
RESULTS/CONCLUSIONS: Prevalence estimates for all substance use outcomes were higher in NSDUH than in NESARC; ratios of NSDUH to NESARC prevalence rates ranged from 2.1 to 5.7 percent for the illegal drug use outcomes. In NSDUH, past year substance use disorder (SUD) prevalence estimates were higher for cocaine and heroin, but they were similar to NESARC estimates for alcohol, marijuana, and hallucinogens. However, prevalence estimates for past year SUD conditional on past year drug use were substantially lower in NSDUH for marijuana, hallucinogens, and cocaine. Associations among drug use and SUD outcomes were substantially higher in NESARC. Total SUD prevalence did not differ between the two surveys, but estimates for blacks and Hispanics were higher in NSDUH. The authors believed that a number of methodological variables might have contributed to such discrepancies; among plausible candidates were factors related to privacy and anonymity, which may have resulted in higher drug use estimates in NSDUH, and differences in SUD diagnostic instrumentation, which may have resulted in higher SUD prevalence among past year substance users in NESARC.
CITATION: Harrison, L. D., Martin, S. S., Enev, T., & Harrington, D. (2007). Comparing drug testing and self-report of drug use among youths and young adults in the general population (HHS Publication No. SMA 07-4249, Methodology Series M-7). Rockville, MD: Substance Abuse and Mental Health Services Administration, Office of Applied Studies.
PURPOSE/OVERVIEW: This report presents the results of a Validity Study conducted in 2000 and 2001 in conjunction with the National Household Survey on Drug Abuse (NHSDA), an annual survey to track the prevalence of substance use in the United States. The purpose of the Validity Study was to provide information on the validity of self-reported drug use in a general population survey by comparing the self-reports of respondents with the results of drug tests of urine and hair specimens obtained from those same respondents.
METHODS: The authors used the data from the Validity Study that was conducted as a supplement to the 2000 and 2001 NHSDAs. A separate national (excluding Alaska and Hawaii) sample of almost 6,000 individuals aged 12 to 25 was selected for the Validity Study, and more than 4,400 individuals completed an interview. These respondents were interviewed using the NHSDA methodology, with a slightly altered questionnaire to eliminate questions not needed for the study and to obtain key information needed for the study. Some of the questionnaire changes included adding questions about drug use corresponding to shorter time periods to be comparable with the window of detection of most drugs in urine and hair. In addition, a persuasion experiment was embedded in the study where half of the respondents were given a statement emphasizing the importance of accurate reporting before other questions specific to the Validity Study were asked. At the end of each interview, respondents were asked to provide a hair and a urine specimen, with an incentive of $25 for each specimen. Exactly 4,000 respondents provided at least one specimen. Specimens were mailed to a testing laboratory, which conducted the drug tests and sent the results to the study team at the University of Delaware. Urine and hair specimens were screened using an immunoassay test for the following drug classes: marijuana/hashish (cannabinoids), cocaine, amphetamines, and opiates. Urine specimens also were analyzed for the presence of cotinine, the principal metabolite of nicotine, using only an immunoassay test. For urine specimens, self-reported drug use (i.e., 30-day, 7-day, 3-day) was compared with confirmatory test results (i.e., positive or negative using the Validity Study cutoffs). Self-reported tobacco use (i.e., 30-day, 7-day, 3-day) was compared with the results of the urine test for cotinine (i.e., positive or negative using a 100 nanograms per milliliter [ng/mL] cutoff). Logistic regression models were developed to determine the correlates of overreporting and underreporting because logistic regression allows other variables that may affect the relationship between self-reports and drug test results to be controlled. After extensive bivariate and multivariate analyses were conducted, a set of variables was derived and used consistently to examine both underreporting and overreporting. These variables were gender, race (white, black, other), region of the country, religiosity, the privacy of the interview, whether the respondent received the experimental appeal to be truthful, difficulty remembering or understanding drug questions, truthfulness, friends’ smoking, and passive exposure to the drug.
RESULTS/CONCLUSIONS: The Validity Study demonstrated its possibility to collect urine and hair specimens with a high response rate from individuals aged 12 to 25 in a household survey environment. The results of tests conducted on hair collected in this study could not be used to compare with self-reports because there were technical and statistical problems related to the hair tests and unresolved issues concerning the interpretation of the analytical results. Most youths aged 12 to 17 and young adults aged 18 to 25 reported their recent drug use accurately. However, there were some reporting differences in either direction—with some not reporting use and testing positive, and some reporting use and testing negative. Biological drug test results can be used as objective markers of drug use to verify self-reports.
CITATION: Iannacchione, V., Morton, K., McMichael, J., Cunningham, D., Cajka, J., & Chromy, J. (2007). Comparing the coverage of a household sampling frame based on mailing addresses to a frame based on field enumeration. In Proceedings of the 2007 Joint Statistical Meetings, American Statistical Association, Survey Research Methods Section, Salt Lake City, UT (pp. 3323–3332). Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: Cost savings, timeliness, and geographic diversity are primary advantages of using mailing addresses instead of field enumeration as a sampling frame for household surveys. The question is whether the advantages of mailing addresses are accompanied by a decrease in the coverage of the household population.
METHODS: The research was based on a probability sample of 50 segments that were assembled from census blocks in North Carolina. Within the geographic confines of each segment, the authors constructed two frames: one based on locatable residential mailing addresses and the other based on field enumeration. The authors used Global Positioning System technology to match the housing units (HUs) from each frame, without presuming that either approach is the “gold standard.”
RESULTS/CONCLUSIONS: Overall, the authors found that field enumeration included approximately 98 percent of the HUs compared with 82 percent coverage for mailing addresses. When restricted to occupied HUs, however, the coverage increased to approximately 99 and 95 percent, respectively. Equal coverage was found in 59 percent of occupied HUs in urban areas. In rural areas, however, mailing addresses were found to have significantly lower coverage than field enumeration. Locatable mailing addresses were nonexistent for 0.4 percent of HUs in areas without home delivery of mail. The authors estimated that field enumeration combined with the half-open interval frame supplementation methodology would yield virtually complete coverage of occupied and unoccupied HUs. An analogous methodology based on a letter carrier’s delivery sequence would increase the coverage of locatable mailing addresses by at least 3.4 percentage points.
CITATION: Kroutil, L., Vorburger, M., & Aldworth, J. (2007, October). Evaluation of the effects of new noncore drug data on prevalence estimates in the National Survey on Drug Use and Health (NSDUH). In 2006 National Survey on Drug Use and Health: Methodological resource book (Section 18, prepared for the Substance Abuse and Mental Health Services Administration, Office of Applied Studies, under Contract No. 283-2004-00022, Deliverable No. 39, RTI 0209009). Research Triangle Park, NC: RTI International.
PURPOSE/OVERVIEW: New questions were added to the 2006 National Survey on Drug Use and Health (NSDUH) to capture information about the use of drugs that respondents were not directly asked about in “core” sections of the interview. These new questions appeared in the noncore special drugs module, and this report presents findings from an investigation of the effects of these questions on drug use prevalence estimates in NSDUH. The specific new questions added in the noncore special drug modules were GHB (gamma hydroxybutyrate), also called “G,” “Georgia Home Boy,” “Grievous Bodily Harm,” or “Liquid G”; Adderall® (a prescription stimulant); Ambien® (a prescription sedative); over-the-counter (OTC) cough or cold medicines; Ketamine (a hallucinogen), also called “Special K” or “Super K”; the tryptamine hallucinogens DMT (dimethyltryptamine), AMT (alpha-methyltryptamine), or Foxy (5-MeO-DIPT [5-methoxy-diisopropyltryptamine]); or Salvia divinorum (a hallucinogen).
METHODS: Measures of lifetime, past year, and past month use (or nonmedical use) of hallucinogens, stimulants, sedatives, prescription psychotherapeutic drugs, illicit drugs, and illicit drugs excluding marijuana that were based only on core data from the 2006 NSDUH were taken from the 2006 analytic file. The core variables that were used to construct these measures had been fully imputed to eliminate missing data. These measures were created in a manner consistent with the procedures in prior survey years. New measures also were created for the lifetime, past year, and past month based on the core drug use data and the noncore data from the new drug items in the special drugs module. Unlike the procedures for the core drug use measures described above, the noncore special drugs variables had undergone logical editing, but had not been statistically imputed. Consequently, these variables had missing data. For this study, missing data for the new drug variables were treated as being equivalent to lifetime nonuse or nonuse in the period of interest.
RESULTS/CONCLUSIONS: The inclusion of both core and noncore data from the 2006 NSDUH special drugs module for GHB, Adderall®, Ambien®, and hallucinogens had little effect on prevalence estimates for the use of hallucinogens, illicit drugs, and illicit drugs excluding marijuana compared with estimates based on core data alone. Inclusion of noncore data for Adderall® and Ambien® had some effect on estimates of the nonmedical use of prescription psychotherapeutic drugs, particularly for the lifetime period. In addition, inclusion of Adderall® and Ambien® data increased the estimate of past year nonmedical use of psychotherapeutic drugs by more than one percentage point for young adults aged 18 to 25. In contrast, inclusion of Adderall® and Ambien® data had notable effects on prevalence estimates for the more proximal measures of nonmedical use of stimulants and sedatives, respectively. In particular, many of the estimates of past year and past month misuse of sedatives more than doubled when estimates included both Ambien® data and core sedative data. Fewer than 1 million adults were estimated to be past year nonmedical users of sedatives based on core data alone compared with nearly 2.4 million based on core and Ambien® data.
CITATION: Shamblen, S. R., & Springer, J. F. (2007). Improving the sensitivity of needs assessment for substance abuse prevention planning: The measurement of differential severity of consequences for individual substance types. Journal of Drug Education, 37(3), 295–316. [PubMed: 18047184]
PURPOSE/OVERVIEW: There is an absence of systematic, comparative research examining the negative consequences that are experienced as a result of using specific substances. Further, techniques typically used for needs assessment (i.e., prevalence proportions) do not take into account the probability of experiencing a negative consequence as a result of using specific substances.
METHODS: An approximated severity index was proposed that (a) takes into account the probability of experiencing negative consequences as a result of using specific substances and (b) is comparable across substances. Data from the National Survey on Drug Use and Health (NSDUH) and the Alcohol and Drug Services Study (ADSS) were used to demonstrate these techniques. The authors defined severity as the intersection of three conditions. First, an individual experiencing a severe consequence must have used the specific substance during his or her lifetime, indicating consumption of the substance. Second, the individual must have been in treatment during his or her lifetime, indicating a recognized dependence or problem-use condition. Third, the individual must have experienced specified negative consequences. When the number in a population within this subset was divided with the number of lifetime users in the denominator, the resulting probability was a severity score directly comparable across substances. It was the probability of experiencing a negative consequence in one’s lifetime and having been in treatment in one’s lifetime, given that one used the specified substance in one’s lifetime.
RESULTS/CONCLUSIONS: The findings suggested that substances typically considered priorities based on prevalence proportions were not the same substances that had a high probability of causing negative consequences.
CITATION: Wright, D. A., Bobashev, G., & Folsom, R. (2007). Understanding the relative influence of neighborhood, family, and youth on adolescent drug use. Substance Use & Misuse, 42(14), 2159–2171. [PubMed: 18097997]
PURPOSE/OVERVIEW: A variety of programs have been developed to prevent substance use among youths in the United States. Oftentimes, these programs target youths directly and may have components that address the relational influence of families, schools, and communities. This article presents some new uses of variance components in understanding the influence of the neighborhood and family on individual drug use. In fields such as education, the use of variance components based on a nested hierarchical structure is more common, although it is rarely used in the area of substance use.
METHODS: The paper focused on marijuana use among youths, and the data came from the 1999 National Household Survey on Drug Abuse (NHSDA), which included a sample of approximately 25,000 youths aged 12 to 17. Using the data, the authors calculated variance components for a binary variable (0 or 1 for “yes” or “no”), which were similar to the methods discussed by Goldstein and Rasbash (1996). Specifically, they used several outcome variables for analysis, one being a direct measure of marijuana use and two others that were indirectly related to drug use.
RESULTS/CONCLUSIONS: The authors estimated variance components when the outcome was dichotomous and found that, for the use of marijuana in the past year, the role of the individual-level variables (individual adolescent vs. role of household vs. role of neighborhood) was quite prominent (79 percent of variation). A similar result was observed for the continuous scale variable of individual positive attitudes toward drug use (83 percent). For continuous constructs related to either the household level (parental monitoring) or the neighborhood level (neighborhood disorganization), the majority of variation still occurred at the individual level (67 and 51 percent, respectively), although they revealed significant percent variation (about 30 percent) at the corresponding family or neighborhood levels as well.
All material appearing in this report is in the public domain and may be reproduced or copied without permission from SAMHSA. Citation of the source is appreciated. However, this publication may not be reproduced or distributed for a fee without the specific, written authorization of the Office of Communications, SAMHSA, HHS.
Your browsing activity is empty.
Activity recording is turned off.
See more...
