A Comparison of model-based methods for Small Area Estimation

Government agencies often provide small area estimates that rely on available data and some underlying model that helps to provide estimates in all areas, even in those that were not sampled. Several models have been well-established for the study of data coming from small areas. In this paper we have made a comparison of some of these methods paying attention to how different types of data sets can be employed efficiently and how to deal with the estimation in areas for which no direct individual data area available. We have considered design-based, regression and EBLUP estimators, which have been fitted using both a likelihood-based and a Bayesian approach. Spatial correlation among areas has also been considered. As in any study of the performance of different models, we have also compared different criteria for model comparisson and selection.