Nonresponse Bias Analysis for National Survey on Drug Use and Health Using Small Area Estimation Methodology

• Activity Number: 135 - Nonresponse Adjustment and Weighting
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Akhil Vaish* and Matthew Williams and Kathy Spagnola and Ana Saravia and Neeraja Sathe
• Companies: RTI International and SAMHSA/CBHSQ and RTI International and RTI International and RTI International
• Keywords: nonresponse bias; small area estimation (SAE); National Survey on Drug Use and Health (NSDUH); survey-weighted hierarchical Bayes

Abstract

Increasing nonresponse rates continue to pose a threat to data quality in household surveys. If a survey's responses differ significantly between respondents and nonrespondents, then the estimates can be biased. Bias in estimates due to nonresponse can also be viewed as a function of the correlation between response propensity and responses to survey items. This paper uses the small area estimation methodology along with geographic predictors (at the state, county, and census tract level) to model response propensities at the screening and interviewing stages of the annual National Survey on Drug Use and Health (NSDUH). Using pooled 2014-2016 NSDUH data, small area estimates of nonresponse and key outcome measures are produced for all 750 of NSDUH's state sampling regions (SSRs) covering the 50 states and the District of Columbia. Using the 750 SSR-level estimates for the response propensities and prevalence of the outcome measures, various descriptive statistics and graphical displays are generated. Overall correlations between the response propensities and survey estimates are produced, which can provide insights into the potential risks of nonresponse bias in survey estimates.