2015 Statistical Inference Report

2014 NSDUH MRB Statistical Inference Report

The focus of this report is to describe the statistical inference procedures used to produce design-based estimates as presented in the 2017 detailed tables and the 2017 FFR, which are based on restricted-use data.

The focus of this report is to describe the statistical inference procedures used to produce design-based estimates as presented in the 2019 detailed tables and the 2019 FFR, which are based on restricted-use data.

The focus of this report is to describe the statistical inference procedures used to produce design-based estimates as presented in the 2020 detailed tables and the 2020 FFR, which are based on restricted-use data.

This report describes the statistical inference procedures used to produce design-based estimates for the 2012 National Survey on Drug Use and Health (NSDUH). These design-based estimates are presented in the 2012 national findings report detailed tables, as well as the 2012 mental health findings report and detailed tables. The report covers missingness rates, sampling error, degrees of freedom, statistical significance differences, confidence intervals, incidence estimates, and suppression of estimates with low precision.

The focus of this report is to describe the statistical inference procedures used to produce design-based estimates as presented in the 2013 detailed tables, the 2013 mental health detailed tables, the 2013 national findings report, and the 2013 mental health findings report. Thestatistical procedures and information found in this report can also be generally applied to analyses based on the public use file as well as the restricted-use file available through the data portal. This report is organized as follows: Section 2 provides background informationconcerning the 2013 NSDUH; Section 3 discusses the prevalence rates and how they were calculated, including specifics on topics such as mental illness, major depressive episode, and serious psychological distress; Section 4 briefly discusses how missing item responses of variables that are not imputed may lead to biased estimates; Section 5 discusses sampling errors and how they were calculated; Section 6 describes the degrees of freedom that were used when comparing estimates; and Section 7 discusses how the statistical significance of differences between estimates was determined. Section 8 discusses confidence interval estimation, and Section 9 describes how past year incidence of drug use was computed. Finally, Section 10 discusses the conditions under which estimates with low precision were suppressed. Appendix A contains examples that demonstrate how to conduct various statistical procedures documented within this report using SAS® and SUDAAN® Software for Statistical Analysis of Correlated Data (RTI International, 2012) along with separate examples using Stata® software.

Given the analytic benefits of having comparable indicators of SMI and AMI for additional years of NSDUH data, this report explores whether new models could be developed from the MHSS data to generate 2005 to 2007 estimates of SMI and AMI that would be comparable with the 2008 and later estimates.

The Mental Health Surveillance Study (MHSS) clinical follow-up was conducted as part of the National Survey on Drug Use and Health (NSDUH) from 2008 to 2012 for the primary purpose of developing models that estimate the prevalence of serious mental illness (SMI) and any mental illness (AMI) in the adult (18 or older) U.S. population (Center for Behavioral Health Statistics and Quality [CBHSQ], 2014). In each of these years, the MHSS clinical study consisted of a subsample of eligible adults selected from the main NSDUH study for follow-up clinical interviews. A prediction model was developed that included SMI status (collected in the clinical interview) as the dependent variable. The predictor variables were variables such as psychological distress and impairment measures that were collected in the main NSDUH questionnaire for adults. The resulting prediction model developed from the MHSS clinical data was then applied to all adult NSDUH respondents to obtain the predicted probability of SMI for each respondent.

Tables showing the 2008-2010 and 2016-2018 substate area estimates and the percent Bayesian confidence intervals along with an indication of statistical significance of the difference.