SMALL AREA ESTIMATION: AN APPLICATION OF A FLEXIBLE FAY-HERRIOT METHOD

Abstract

The importance of small area estimation in survey sampling is  increasing, due  to the growing demand for reliable small area estimation from both public and private sectors. In this paper, we address the important issue of using  statistical modeling techniques to compute more reliable small area estimates. The main aim is to assess the use of a flexible methodology for small area estimation. We formulate a new flexible small area model by incorporating a tuning (index) parameter into the standard area-level (Fay-Herriot) model. We achieve this using a combination of two methods namely, empirical Bayes (EB) approach and hierarchical Bayes (HB) approach. Our results suggest that the proposed model can be seen as advancement  over the standard Fay-Herriot model. The novelty here is that we have developed a flexible way to handle random effects in small area estimation. The Implementation of the proposed model is only mildly more difficult than the Fay-Herriot model. We have obtained results for both EB approach and the HB approach. Compared with the corresponding HB procedure, the EB approach saves a tremendous computing time and is very simple to implement.