Properties of plug-in predictors in case of some model misspecification in small area analyses

The article treats the problem of predicting characteristics in domains using plug-in predictors. The simulation studies analyse the properties of the considered plug-in predictors in the case of selected types of model misspecification. The studies were conducted assuming a linear mixed model with correlated random effects specific for domains, and the considered characteristics were global values in the domain. A comparison of the properties of the analysed plug-in predictors was also made with selected statistics of the studied characteristics. Analyses were conducted based on real data on municipalities’ income in a million zlotys for three years from the Local Data Bank (Statistics Poland). The R language was used to prepare the simulation studies.