CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Small area estimation with partially missing direct estimates Authors:  Jan Pablo Burgard - Trier University (Germany) [presenting]
Domingo Morales - University Miguel Hernandez of Elche (Spain)
Anna-Lena Woelwer - Trier University (Germany)
Abstract: A common problem in small area estimation is that direct estimates are not reported for some areas or domains. This typically occurs when the areas or domains of interest are not planned domains in the sampling design. But also too imprecise estimates, e.g. because of too low sample sizes, lead statistical agencies to suppress their publication. This inhibits the use of classical small area estimation methods like the Fay-Herriot model. We propose an empirical best predictor for the prediction of area and domain parameters for the situation of partially missing direct estimates. Further, we propose a REML-like algorithm to estimate model parameters. Besides the parameter estimation we also show a first approach for the estimation of the MSE of the empirical best predictor. The performance of this method is evaluated in a model-based simulation study, showing both advantages and caveats of the newly proposed estimator.