Bayesian Statistics for Small Area Estimation

Abstract. National statistical offices are often required to provide statistical information about characteristics of the population, such as mean income or unemployment rate, at several administrative or small area levels. Having good area level estimates is important because policies will often be based on this type of information. In this paper we describe how Bayesian hierarchical models can help in the task of providing good quality small area estimates. Starting from direct estimates obtained from survey data, we describe a range of Bayesian hierarchical models that incorporate different types of random effects and show that these give improved estimates. Models that synthesise individual and aggregated information are considered as well. Finally, we highlight some additional applications that further exploit the estimates produced, such as the classification and ranking of areas and how to approach the problem of having no direct information in several areas.