Robust Bayesian Small-Area Estimation for Area-Level Data

• Activity Number: 634
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Adrijo Chakraborty* and Gauri S. Datta and Abhyuday Mandal
• Companies: NORC at the University of Chicago and University of Georgia/U.S. Census Bureau and University of Georgia
• Keywords: Hierarchical Bayes ; Fay-Herriot model ; Small area estimation

Abstract

The Fay-Herriot model has received great appreciation among model-based survey researchers in the past few decades. The classical Fay-Herriot model assumes normality for the random small area effects, which may not be appropriate in various circumstances, particularly in the presence of outliers. We discuss different cases where the normality assumption for random effects may lead to poor small area estimates and subsequently propose two robust extensions of the Fay-Herriot model. In this paper we focus on a hierarchical Bayesian approach in order to build the models. The performance of these models is examined through an extensive simulation study.