An imputation model database and its relevance to analysis
The National Survey on Drug Use and Health (NSDUH) is sponsored by the Substance Abuse and Mental Health Services Administration and provides national, state and substate data on substance use and mental health in the civilian, noninstitutionalized population age 12 and older. The NSDUH is a continuous survey, with approximately 67,500 interviews completed annually. As part of the NSDUH imputation procedures, over 400 regression models are fit each year. These models are used to match each item nonrespondent with a "neighborhood" of similar item respondents in order to identify a donor. The response variables in these models are variables of primary interest to analysts. After the procedures are complete for each year, an imputation model database is populated which stores covariate-level information such as the p-values associated with the regression coefficients. This database is used both by staff working on the NSDUH imputation and by staff analyzing the NSDUH data. This paper illustrates how such a database can be used not only by those conducting the imputation, but also by those making decisions during the analysis of NSDUH data.