The Behavioral Risk Factor Surveillance Systems (BRFSS) is designed to produce accurate estimates for each state and for the entire United States but not for small geographic areas or subpopulations. This paper summarizes our investigations on small area estimation techniques to obtain sub-state estimates from the BRFSS and the strategies to elect appropriate models.
Our research concerns estimates of proportions such as the probability of having any health care coverage. To include all sources of variation in the model, we apply the multi-level regression models. First, for each respondent, we include individual-level demographic classifications in multivariate regression. Second, for each cluster (county), the model includes local-level covariates. We apply two basic models: a hierarchical model and a random effect model. We describe the numerical methods needed to obtain model parameter estimates. Then we discuss the strategies for model inference and model selection. Finally, we summarize two different numerical methods using results from two examples: (a) health service area level mammography rates, and (b) county level severe work disability rates.
Learning Objectives: At the conclusion of this talk, the listener should be able to describe how personal and regional covariates may be combined in fixed and mixed effects multilevel regression models using BRFSS data to estimate local proportions of subjects with specific risk factors.
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.