Abstract

We consider the problem where estimates of proportions are needed for domains with typically very small sample sizes in the sense that it might be difficult to construct any standard small area model. Assuming the availability of good proxy variables, we propose simple design-based small area estimators supported by area-level models and evaluate their accuracy. We compare these estimators with the empirical best linear predictors based on the Fay—Herriot model in the simulation study using data from surveys of official statistics.