• Reweighted estimators for the common principal components model: Influence functions and Monte Carlo study

BOENTE, GRACIELA; PIRES, ANA M.; RODRIGUES, ISABEL

Beijing, China

Congreso; International Conference on Robust Statistics (ICORS 2004); 2004

Several authors, as Flury (1988), have studied models for common dispersion structure. Those models have been introduced to overcome the problem of an excessive number of parameters, when dealing with several populations, in multivariate analysis. As it is well known, reweighted estimators allow to improve the asymptotic efficiency of the initial estimators. We consider a reweighted estimator of the scatter matrices of each population, where the weights do not depend on the Mahalanobis distance as usually, but on the outlier detection measures defined in Boente, Pires and Rodrigues (2002a). In order to avoid masking, those measures depend on the influences on the classical functionals but with the unknown parameters robustly estimated. As it is well known, reweighted estimators allow to improve the asymptotic efficiency of the initial estimators. We consider a reweighted estimator of the scatter matrices of each population, where the weights do not depend on the Mahalanobis distance as usually, but on the outlier detection measures defined in Boente, Pires and Rodrigues (2002a). In order to avoid masking, those measures depend on the influences on the classical functionals but with the unknown parameters robustly estimated.