Area-level small area estimation with missing values

Model-based small area predictors are derived under the assumption that data files are complete. In application to real data, files may contain missing values. We introduce a variant of the bivariate Fay-Herriot model that takes into account for missing values in one component of the target variable and give fitting algorithms to estimate the model parameters. Based on the new model, we introduce empirical best predictors of domain means and derive an approximation to the mean squared error.