Robust estimation of mixed models

L.A. García-Escudero

{}^{1}

F. Laurini

{}^{2}

A. Mayo-Iscar

{}^{1}

Abstract

Parameter estimation for mixed models can be severely biased by outliers. The outliers can be in the conditional distribution of the response variable or even in the explanatory variables. We propose to mitigate the effect of outliers by introducing robust estimations by trimming a fixed proportion of observations. The level of trimming is an input hyper-parameter and it can be adjusted and monitored with proper diagnostics. We show some results with artificial data and compare with available competitor. We also apply the proposed robust method to real data, based on trade data of the European Union.

${}^{1}$

Department of Statistic and Operational Research and IMUVA, and RoSA, University of Valladolid, Valladolid, Spain [lagarcia@uva.es, agustin.mayo.iscar@uva.es]
${}^{2}$

Department of Economics and Management and RoSA, University of Parma, Parma, Italy [fabrizio.laurini@unipr.it]

Keywords: Mixed models – Robustness – Trimming