Because of budget constraints a survey has two major limitations when it comes to availing data on prevalence of diabetes in small areas as Counties. First, it is costly for a survey to cover all relevant areas. And second, a survey often comes short of taking large samples for adequate representations. Examining such limitations and shortcomings of a direct method of estimation which uses data from such surveys, this dissertation attempted to apply Bayesian Hierarchical Model of estimation to provide reliable data on prevalence of Diabetes in small areas (counties). In doing so a range of Bayesian Hierarchical models which provide reliable data on prevalence of diabetes for small areas as counties were explored.

The Estimation Models used data of Behavioral Risk Factor Surveillance System (BRFSS [1]) survey. In total the analysis examined survey data made on 1,497 counties (including the 644 counties in the CDC diabetes belts [2] in 16 states in the US.

The statistical models used in this analysis are aimed at reducing estimation error of diabetes prevalence in direct estimation methods, so as to help an efficient policy formulation and budget allocation. In this regard we generated estimates on the prevalence of diabetes for 1,188 Counties having a complete set of information and another 295 which were not covered in BRFSS survey and among the 1188 Counties 824 Counties that have smaller sample size (Healthy people 2020 data suppression for BRFSS [3]).

Unlike the direct method usually applied for such estimation the result in this analysis brought about statistical significance of the estimates in our study.