Robust Fay Herriot Estimators in Small Area Estimation Seminar

Due to reasons of confidentiality only aggregated data may be accessible, i.e. direct estimators of area means and variances. In such cases the analyst can still be confronted with outlier contaminated data which may influence the analysis in an unintended manner. Robust techniques for area-level models can be beneficial. However on the aggregated level of information outlying observations do not influence the results as strong as they would on the individual level. So a discussion of outliers on the area-level is motivated by a unit level perspective. On unit-level we may consider outlying areas and outlying observations which result in different compositions on the area-level. To justify the interest in this topic it is shown how the Fay-Herriot model and a robust extension perform in different model-based simulation scenarios and what we can conclude for practical applications.