Nonlinear mixed effects cross-sectional models for estimation of smoking proportions using the National Health Interview Survey

• Activity Number: 414
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Neung Ha*+ and Van Parsons
• Companies: University of Maryland and National Center for Health Statistics
• Keywords: Small Area Estimation ; National Health Interview Survey ; Hierarchical Modeling ; Cross-sectional time-series model ; MCMC

Abstract

The National Health Interview Survey (NHIS) is an annual health survey conducted by the National Center for Health Statistic. The survey design provides reliable annual estimates for health-related conditions for the nation and the four major geographical regions of the United States. However, direct estimates for some states or sub-state regions are unreliable due to insufficient sample sizes. We propose small area models that include both local area and short-term time random effects, which describe year-to-year variation over cluster samples, to estimate the prevalence of smoking for each of the fifty U.S. states and the District of Columbia. In particular, hierarchical Bayesian nonlinear mixed effect models, using the 2006 to 2010 NHIS data, is explored. Auxiliary variables will be obtained from the Area Resource File. Bayesian Markov Chain Monte Carlo (MCMC) approaches will be used for estimation. A major portion of this study is a discussion of various methods to estimate the time specific sampling covariances needed to implement the proposed models. Comparison of different models by model fits and model performance are discussed.