Analysis of Basic Area-Level Models: The Extensions of the Fay-Herriot Model

• Activity Number: 78
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Abhishek Nandy* and Snigdhansu Chatterjee
• Companies: University of Minnesota and University of Minnesota
• Keywords: Multivariate Fay Herriot ; Mean squared prediction error ; time dependent

Abstract

Small Area Estimation has gained importance over the years in survey research methods for the purposes of effective allocation of government funds for economic, health and social planning. The direct survey based estimators are unreliable due to insufficient number of observations from each small area. The way to overcome this problem is typically to use model assisted predictors. Estimating mean square prediction error of these estimators is a crucial step to determining the quality of these predictors and is the first step towards developing other inference techniques. The situation gets very complicated when we consider predicting small area means of a bunch of correlated attributes that have possibly been measured over time as well. In this article, we work on developing theories and methods for the prediction of small area means, the estimation of their mean squared prediction errors and subsequent second order bias correction of these estimators in multivariate and multivariate-time series version of basic area level models under the assumptions of known and unknown sampling variances. We present simulation results and discuss applications of our developed methods.