Application of Hierarchical Bayesian Models with Poststratification for Small-Area Estimation from Complex Survey Data

• Activity Number: 129
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Vladislav Beresovsky*+ and Catharine W. Burt and Van Parsons and Nathaniel Schenker and Ryan Mutter
• Companies: Centers for Disease Control and Prevention and Centers for Disease Control and Prevention and Centers for Disease Control and Prevention and Centers for Disease Control and Prevention and Agency for Healthcare Research and Quality
• Keywords: health care utilization ; small area estimation ; multilevel logistic regression ; poststratification ; hierarchical Bayesian model

Abstract

Small area estimation from stratified multilevel surveys is well known to be challenging because of extreme variability of survey weights and the high level of data clustering. These challenges complicate county- and state- level estimates of healthcare indicators such as proportions of visits with asthma and injury diagnoses at emergency departments (ED) from the National Hospital Ambulatory Medical Care Survey (NHAMCS). In this study, proportions of visits with asthma and injury diagnoses to hospital EDs were predicted by various multilevel logistic regression models and then aggregated to state level estimates. County level population covariates from the Area Resource File, hospital level covariates from Verispan Hospital Database and survey design information were used for modeling fixed effects. Aggregation of predicted hospital proportions to state level estimates utilizing the available number of ED visits to each hospital amounts to poststratification with cells defined at the state level. We evaluated models by comparing predictions with estimates based on administrative data from the Healthcare Cost and Utilization Project (HCUP) databases.