A Multilevel Modeling Approach to Explaining State-Level Variation in Response Rates
Safir, A., Murphy, J., Park, H., & Wang, K. H. (2007, May). A Multilevel Modeling Approach to Explaining State-Level Variation in Response Rates. Presented at American Association for Public Opinion Research Conference, Anaheim, CA.
This paper explores the effects of area, household, interviewer, and respondent level factors on survey contact and cooperation in the National Survey on Drug Use and Health (NSDUH). The NSDUH is Federal Government's primary source of data on the use of alcohol, tobacco, and illicit substances in the civilian, non-institutionalized population of the United States aged 12 or older. This annual cross-sectional survey collects data by screening up to 160,000 households and administering a questionnaire to a representative sample of approximately 67,000 persons at their place of residence. In the analysis, we fit multi-level models to measure the correlation between screener contact, screener cooperation, and interview cooperation, and factors such as county, segment, household, respondent, and field interviewer characteristics. The primary purpose was to determine which correlates of nonresponse can most effectively be used to explain variation in response rates within a state over time and across states at a single point in time, taking into account the interaction between the various covariates. A secondary purpose was to develop methods, within a framework of responsive design, to facilitate strategic resource decisions for minimizing response rate declines due to suboptimal calling patterns, variation in selected segments, or interviewing staff turnover from one survey phase to the next. The findings reveal data patterns which may inform future operational efforts to target specific correlates of nonresponse. The results also demonstrate the potential utility of process data review for the explanation and prediction of state-level response rate changes.