Robust estimation in linear regression models with fixed effects

In this work we extend the procedure proposed by Peña and Yohai (1999) for computing robust regression estimates in linear models with fixed effects. We propose to calculate the principal sensitivity components associated to each cluster and delete the set of possible outliers based on an appropriate robust scale of the residuals. Some advantage of our robust procedure are: (a) it is computationally low demanding, (b) it is able to avoid the swamping effect often present in similar methods, (c) it is appropriate for contamination in the error term (vertical outliers) and possibly masked high leverage points (horizontal outliers). The performance of the robust procedure is investigated through several simulation studies.