Title: Robust estimation and efficient estimates of partition and model parameters: A step beyond the normality assumption
Authors: Francesca Greselin - University of Milano Bicocca (Italy) [presenting]
Andrea Cerioli - University of Parma (Italy)
Luis Angel Garcia-Escudero - Universidad de Valladolid (Spain)
Agustin Mayo-Iscar - Universidad de Valladolid (Spain)
Marco Riani - University of Parma (Italy)
Abstract: An iteratively reweighted approach is extended to cluster partitions recently introduced in the literature, and based on multivariate normal clusters, to the case of leptokurtic cores. Assuming multivariate t-distributions for the cluster cores, we estimate their parameters, starting from a high proportion of trimming, and iteratively increasing the active sample size in a controlled fashion. To guarantee consistency at the t-distributed model components, we employ consistency factors to inflate the covariance matrices estimates, based on non-trimmed data. Simulation results and examples on real data show the effectiveness of the approach.