Application of an area level model to small area estimation

In this contribution we extend the Fay-Herriot model to an area level linear regression model with random intercepts having one of two possible variances. The model is applicable to data having a large subset of domains where direct estimates of the variable of interest cannot be described in the same way as in its complementary subset of domains. This is generally the case when domains are constructed by crossing geographical characteristics with sex. Estimation procedures for the variance components and regression parameters are considered and empirical best linear unbiased predictors (EBLUP) of domain parameters are derived as well as estimates of the mean square error of the EBLUP. Monte Carlo simulation experiments are presented to illustrate the gain of precision obtained by using the proposed model and to get some practical conclusions. A motivating application to Spanish Labour Force Survey data is also given.