Reliable Robust Regression Diagnostics

• DOI: 10.1111/insr.12103
• Publication Date: 2015-04-24T05:51:14Z
• Authors: Silvia Salini; Andrea Cerioli; Fabrizio Laurini; Marco Riani
• Keywords: Robust regression; Robust Statistics

Summary Motivated by the requirement of controlling number false discoveries that arises in several application fields, we study behaviour diagnostic procedures obtained from popular high‐breakdown regression estimators when no outlier is present data. We find empirical error rates for many available techniques are surprisingly far prescribed nominal level. Therefore, propose a simulation‐based approach to correct liberal diagnostics and reach reliable inferences. provide evidence our performs well wide range settings practical interest variety robust techniques, thus showing general appeal. also evaluate loss power can be expected corrections under different contamination schemes show this often not dramatic. Finally, detail some possible extensions may further enhance applicability method.