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: Chen, C. Y., & Anthony, J. C. (2003). Possible age-associated bias in reporting of clinical features of drug dependence: Epidemiological evidence on adolescent-onset marijuana use. Addiction, 98(1), 71–82. [PubMed: 12492757]
PURPOSE/OVERVIEW: As of 2003, the latest drug use reports showed a higher level of marijuana dependence for adolescents than for adults. This paper explored potential age-related differences in marijuana dependence using multivariate analysis and item-response biases.
METHODS: Marijuana dependence was measured using seven binary survey items from the 1995 to 1998 National Household Survey on Drug Abuse (NHSDA) questionnaires. Of the 86,021 respondents for these 4 years of the NHSDA, 2,628 (1,866 adolescents and 762 adults) were identified as recent-onset marijuana users. Multivariate response analysis using a generalized linear model (GLM) and generalized estimating equations (GEE) was performed to measure the age-related difference in reported marijuana use while taking into account the interdependencies of the yes-no responses and controlling for covariates. Further analyses were performed using the multiple indicators/multiple causes (MIMIC) multivariate response model to measure age-associated response bias.
RESULTS/CONCLUSIONS: The primary findings were that of the recent-onset marijuana users, younger users reported more drug dependency than older users. This association was found to be statistically significant in the multivariate analysis model, which controlled for other covariates. The MIMIC model found that there also were age-related biases in reporting of drug use. Younger users were biased toward reporting dependent behaviors. These findings support the notion that there may be differences in what constitutes “marijuana dependence” for adolescents compared with adults.
CITATION: Fendrich, M., Wislar, J. S., & Johnson, T. P. (2003). The utility of debriefing questions in a household survey on drug abuse. Journal of Drug Issues, 33(2), 267–284.
PURPOSE/OVERVIEW: Respondent’s discomfort in revealing answers to sensitive survey questions is often a cause of underreporting. This paper measures the effect that respondents’ comfort level has on their answers to sensitive questions and whether there are any differences in mode.
METHODS: Respondents were randomly assigned to receive either the control or experimental condition. The control condition replicated the design procedures from the 1995 National Household Survey on Drug Abuse (NHSDA), which used an interviewer-administered paper-and-pencil interviewing (PAPI) survey. The experimental condition had the choice of completing a computer-assisted personal interviewing (CAPI) survey or an audio computer-assisted self-interviewing (ACASI) survey. Following the interview, respondents were asked for a hair sample. At the end of the interview, interviewers administered two types of debriefing questions, subjective and projective, to elicit respondents’ level of discomfort during the study.
RESULTS/CONCLUSIONS: Respondents’ willingness to disclose sensitive information was correlated with their subjective and projective levels of discomfort with the survey. Contrary to the hypothesis, respondents who disclosed information revealed less comfort with the survey on the subjective debriefing items and more comfort on the projective debriefing questions. In addition, the degree of discomfort for survey respondents was greater for those in the experimental condition than the control condition. Levels of projective and subjective discomfort differed by subgroups for race, age, and education.
CITATION: Gfroerer, J., Penne, M., Pemberton, M., & Folsom, R. (2003). Substance use treatment need among older adults in 2020: The impact of the aging baby-boom cohort. Drug and Alcohol Dependence, 69, 127–135. [PubMed: 12609694]
PURPOSE/OVERVIEW: The purpose of this paper is to provide estimates of the number of older adults (defined as those 50 or older) needing treatment for substance use problems in the future as the U.S. baby-boom population ages.
METHODS: Data from the 2000 and 2001 National Household Surveys on Drug Abuse (NHSDAs) were used to estimate logistic regression models among adults aged 50 or older, predicting treatment need in the past 12 months, which was defined as being classified with substance dependence or abuse based on DSM-IV criteria. Separate regression models were estimated for individuals depending on whether they had used alcohol by age 30. For those who had not used alcohol by age 30, age was the only predictor. For those who had used alcohol prior to age 31, predictors in the regressions included age, gender, race/ethnicity, and substance use prior to age 31. Estimated parameters from these models then were applied to a pooled sample of adults 30 or older from the 2000 NHSDA and 31 or older from the 2001 NHSDA to produce predicted probabilities of treatment need. Weights were adjusted to match census population projections on age, race, and gender. An additional adjustment was made for expected mortality rates. Standard errors for the 2020 projections were computed using jackknife replication methods.
RESULTS/CONCLUSIONS: The estimated number of older adults needing treatment for a substance use problem is projected to increase from 1.7 million in 2000/2001 to 4.4 million in 2020. This increase is the result of a 50 percent increase in the population aged 50 or older (from 74.8 million to 112.5 million) combined with a 70 percent increase in the rate of treatment need (from 2.3 to 3.9 percent) in the older population. Increases are projected for all gender, race, and age groups. About half of the projected 2020 population needing treatment will be aged 50 to 59, and two thirds will be male.
CITATION: Gordek, H., & Singh, A. C. (2003). Comparison of linear and nonlinear models for generalized variance functions for the National Survey on Drug Use and Health. In Proceedings of the 2003 Joint Statistical Meetings, American Statistical Association, Section on Survey Research Methods, San Francisco, CA [CD-ROM]. Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: The generalized variance function (GVF) provides an approximation to the variance of an arbitrary domain estimate by means of modeling the relationship between the estimated variances of a selected set of domain estimates and a set of covariates, rather than through direct computation. The use of GVF is important in surveys with a very large number of characteristics, where publishing limitations and the immense amount of potential subgroupings may make several variance estimates of interest unavailable directly.
METHODS: For the National Survey on Drug Use and Health (NSDUH), GVFs were obtained using linear models with the log of the relative variance as the dependent variable, as well as a somewhat modified version that ensures the resulting design effect (deff) to be at least one. The authors consider a nonlinear generalization of these models. Numerical results on comparison of various models using the 2001 NSDUH data are presented.
RESULTS/CONCLUSIONS: The authors extended the idea of constant deff for a subset of estimates to constant deff-type parameters, such as variance-odds deff, and then proposed an alternative GVF model that overcomes the limitations of existing models. Ordinary least squares was used to fit these models to the NSDUH data, and results from different models were compared. It was found that different models performed quite similarly, and there seemed no compelling reason to change from simpler, commonly used models. However, the authors concluded that it would be useful to investigate the impact of using other models, as well as the use of weighted least squares in model fitting. For instance, the authors indicated that the effect of nonlinear modeling should be investigated. More specifically, rather than formulating a linear model for the transformed point estimate of variance, one can model linearly the mean of the variance estimate after transformation (such as log or logit).
CITATION: Kessler, R. C., Barker, P. R., Colpe, L. J., Epstein, J. F., Gfroerer, J. C., Hiripi, E., Howes, M. J., Normand, S. L., Manderscheid, R. W., Walters, E. E., & Zaslavsky, A. M. (2003). Screening for serious mental illness in the general population. Archives of General Psychiatry, 60(2), 184–189. [PubMed: 12578436]
PURPOSE/OVERVIEW: The Substance Abuse and Mental Health Services Administration (SAMHSA) was charged with the task of defining “serious mental illness” (SMI) in adults and establishing a method that States can use to estimate its prevalence.
METHODS: This paper compared three sets of screening scales used to estimate the prevalence of SMI: the World Health Organization (WHO) Composite International Diagnostic Interview Short Form (CIDI-SF) scale, the K10 and K6 psychological distress scales, and the WHO Disability Assessment Schedule (WHODAS). A convenience sample of 155 respondents received the three screening scales using computer-assisted self-interviewing (CASI). After the self-administered questions, respondents completed the 12-month Structured Clinical Interview (SCID), including the Global Assessment of Functioning (GAF). SMI was defined in the SCID as any 12-month DSM-IV disorder (defined in the 4th edition of the Diagnostic and Statistical Manual of Mental Disorders), besides substance use, that also had a GAF score less than 60.
RESULTS/CONCLUSIONS: All screening scales were found to be significantly correlated with SMI. The shortest scale, the K6, was the most statistically significant predictor of SMI. These results support the notion that short, fully developed, and carefully constructed screening scales can be strong predictors of the same results found in more lengthy and expensive clinical interviews. Another advantage of the K6 and K10 scales is that they can be administered in less time than 2 or 3 minutes, respectively.
CITATION: Odom, D. M., Eyerman, J., Chromy, J. R., McNeeley, M. E., & Hughes, A. L. (2003). The effect of interviewer experience on the interview process in the National Survey on Drug Use and Health. In Proceedings of the 2003 Joint Statistical Meetings, American Statistical Association, Section on Survey Research Methods, San Francisco, CA [CD-ROM]. Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: Analysis of survey data from the National Survey on Drug Use and Health (NSDUH) has shown a relationship between field interviewer (FI) experience, response rates, and the prevalence of self-reported substance use (Eyerman, Odom, Wu, & Butler, 2002; Hughes, Chromy, Giacoletti, & Odom, 2001, 2002; Substance Abuse and Mental Health Services Administration [SAMHSA], 2000). These analyses have shown a significant and positive relationship between the amount of prior experience an FI has with collecting NSDUH data and the response rates that a FI produces with his or her workload. These analyses also have shown a significant and negative relationship between the amount of prior experience of an FI and the prevalence of substance use reported in cases completed by that FI. In general, these analyses have been consistent with the published literature that FIs can influence both the success of the data collection process and accuracy of the population estimates (Martin & Beerteen, 1999; Singer, Frankel, & Glassman, 1983; Stevens & Bailar, 1976). Previous NSDUH analyses examined response rates and prevalence estimates independently. This made it difficult to determine whether the lower prevalence estimates for experienced FIs were a result of the change in the sample composition due to higher response rates or whether the lower prevalence estimates were a result of a direct effect of FI behavior on respondent self-reporting.
METHODS: This analysis combines these two explanations to produce a conceptual model that summarizes the authors’ expectations for the relationship between FI experience and prevalence estimates. The combined explanation from the conceptual model is evaluated in a series of conditional models to examine the indirect effect of response rates and the direct effect of FI experience on prevalence estimates. Earlier analyses using NSDUH data have shown a negative correlation between interviewer experience and substance use rates. However, these studies only examined the last step in the interviewing process, administering the questionnaire. This paper further explores the effect of interviewer experience by investigating a series of separate logistic models that are conditionally based on each step of the screening and interviewing (S&I) process. The S&I steps examined are contacting the household, gaining household cooperation, contacting the selected person(s), interviewing the selected person(s), and reporting of substance use. By separating the analysis into these steps, estimating the effect of interviewer experience on data collection at each stage of the survey is possible.
RESULTS/CONCLUSIONS: In conclusion, the analysis shows that increased FI experience simultaneously increases response rates and decreases prevalence estimates. In addition, the effect of increased FI experience on prevalence estimates cannot be fully explained by the adjustments based on earlier models (i.e., S&I level) to the final prevalence estimate model. In other words, the FI effect on prevalence cannot be fully attributed to the increase in response rates by experienced FIs. Furthermore, FI experience was significant in the final model, showing that the covariates also did not account for all the decrease in prevalence. Three hypotheses were given as possible explanations for the decrease in prevalence. As was shown in the statistical analysis, the marginal rates were too extreme to support the first hypothesis. This means that although some level of selection bias may be occurring, it is not the only cause of the decrease in prevalence estimates for experienced FIs. More likely, the relationship between FI experience and prevalence estimates is captured in hypothesis 3 (i.e., the decrease in prevalence estimates for experienced FIs is a function of lower substance use reporting by the additional respondents they obtain and also the remaining respondents that FIs with all levels of experience interview).
CITATION: Office of Applied Studies. (2003). Appendix C: NSDUH changes and their impact on trend measurement. In Results from the 2002 National Survey on Drug Use and Health: National findings (HHS Publication No. SMA 03-3836, NSDUH Series H-22, pp. 107–137). Rockville, MD: Substance Abuse and Mental Health Services Administration.
PURPOSE/OVERVIEW: This appendix presents the results of analyses designed to determine the degree to which increases in the rates of substance use, dependence, abuse, and serious mental illness (SMI) between the 2001 National Household Survey on Drug Abuse (NHSDA) and the 2002 National Survey on Drug Use and Health (NSDUH) could be attributed to important methodological differences between the two surveys. Major changes between the 2 years included (1) a change in the survey’s name, (2) providing a $30 incentive to all interview respondents, (3) improved data collection quality control procedures, and (4) switching to the use of 2000 census data as the basis for population weighting adjustments for the 2002 survey.
METHODS: Six types of analyses were used to assess the degree to which methodological changes could account for the various increases in prevalence rates: (1) a retrospective cohort analysis that examined the degree to which increases in lifetime use could be attributed to new initiates; (2) a response rate pattern analysis that examined changes in response rates between the fourth quarter of 2001 and the first quarter of 2002 for various geographic and demographic groups, reasons for refusal, and field interviewer characteristics; (3) a response rate impact analysis that attempted to determine whether the increased response rate between 2001 and 2002 resulted in “additional” respondents who in turn accounted for higher rates of substance use; (4) an analysis of the impact of new census data in which data for the 2001 survey were reweighted using 2000-based census control totals rather than 1990-based census data; (5) analyses in which measures of substance use were regressed on variables that included indicators related to the timing and occurrence of changes in data collection quality control procedures to determine the impact of these changes; and (6) a reexamination of differences in prevalence rates from the 2001 incentive experiment by applying response propensity adjusted weights to account for response rate differences and a comparison of these differences with differences in estimates between quarter 4 of the 2001 survey and quarter 1 of the 2002 survey.
RESULTS/CONCLUSIONS: (1) Increases in lifetime use cannot be accounted for by new initiates in 2002 or a cohort shift. (2) Response rate increases occurred across all geographic and demographic groups, with the exception of adults aged 50 or older. (3) “Additional” respondents from the response rate increase between 2001 and 2002 cannot account for the increase in prevalence rates between 2001 and 2002. (4) Reweighting the 2001 survey by using 2000 census-based control totals had minor effects on prevalence rates and a larger impact on totals. (5) Field interventions introduced during the 2001 survey appear to have had little effect on substance use estimates. (6) Differences in estimates between quarter 1 of 2002 and quarter 4 of 2001 were generally larger than differences in estimates from the incentive experiment between incentive and no-incentive groups, suggesting that incentive effects alone do not account for overall differences.
CITATION: Painter, D., Chromy, J., Meyer, M., Granger, R. A., & Clarke, A. (2003). Effect of incentives on data collection: A record of calls analysis of the National Survey on Drug Use and Health. In Proceedings of the 2003 Joint Statistical Meetings, American Statistical Association, AAPOR - Section on Survey Research Methods, San Francisco, CA (pp. 170–176). Alexandria, VA: American Statistical Association.
PURPOSE/OVERVIEW: Given the decline in response rates during the late 1990s, a randomized, split-sample experiment was conducted during the first 6 months of data collection for the 2001 National Survey on Drug Use and Health (NSDUH) to evaluate the effectiveness of monetary incentives in improving respondent cooperation. Based on the outcome of the 2001 experiment, NSDUH interviewers began offering a $30 incentive to all survey respondents in 2002. This paper analyzes the effect of the new $30 incentive on the data collection process as measured by record of calls (ROC) information.
METHODS: An ROC was generated each time an interviewer visited a household. It consists of a code that describes the outcome of the visit along with any notes an interviewer deems pertinent to a future visit. The data contain extensive information on the amount of effort taken to contact and obtain cooperation from respondents. Using these data, the authors analyzed the impact of the incentives on contact and cooperation patterns. The authors developed measures of the intensity of the follow-up, which include the number of calls, the timing of calls, special follow-up letters, the use of alternative interviewers, and similar special procedures used in trying to achieve contact and cooperation. These measures of the intensity of the follow-up were then related to household contact rates, household screening cooperation rates, respondent contact rates, and respondent cooperation rates. The authors first provided a brief description of the change in response rates between 2001 and 2002. Then they presented a model-based analysis assessing the changes in the patterns after the introduction of the incentives, while controlling for interviewer and respondent characteristics.
RESULTS/CONCLUSIONS: NSDUH data showed that the increases in response rates that accompany the use of incentives were also accompanied by the need for fewer call days and fewer calls to finalize sample dwelling units and sample persons. A review of the demographic data by call order showed that the sample distribution on demographic measures could be changed by prematurely cutting off the follow-up of pending cases. More study is needed to determine what effect such policies might have on the principal study measures on substance use addressed in NSDUH. The monitoring of ROC data provides a useful tool for ensuring that adequate follow-up procedures are being used within the limits of reasonable cost management.
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...
