Indicators and in particular the Laeken indicators are vulnerable to outliers. Outliers may not only bias the indicators but also introduce a high additional variability. Therefore, outliers and other deviations from theoretical distributions may undermine the quality of indicators thoroughly. Robust procedures remain stable in those situations but are more complex to handle.
The work package on robustness will develop robust imputation procedures, in particular for multivariate data, including detection of outlying and influential observations. The work package will develop and evaluate robust small area estimation procedures, which shall make the results less sensitive to deviations from the models needed for small area estimation (strong interaction with WP2). The procedures will be implemented in the statistical programming language R and a corresponding package will be developed. The work package will evaluate the robustness of classical indicators, in particular the Laeken indicators, and of robust procedures. Quality measures on the robustness of indicators and on the impact of robust procedures will be developed. The work package will formulate proposals for recommendations on the use of the procedures and for the interpretation of the quality measures by the users of the indicators. These proposals will be validated in work package 7 on Analysis.