An empirical evaluation of small area estimators

This paper investigates the comparative performance of five small area estimators. We use Monte Carlo simulation in the context of both theoretical and empirical populations. In addition to the direct and indirect estimators, we consider the optimal composite estimator with population weights, and two composite estimators with estimated weights: one that assumes homogeneity of within area variance and square bias, and another one that uses area specific estimates of variance and square bias. It is found that among the feasible estimators, the best choice is the one that uses area specific estimates of variance and square bias.