A0282
Title: Novel robust estimator in logistic regression
Authors: Alfonso Garcia-Perez - Universidad Nacional de Educación a Distancia (UNED) (Spain) [presenting]
Abstract: Down-weighting the observations according to their outlyingness, understood as a large distance to the center of the data, is an old idea in Robustness. In fact, the famous robust regression estimator LTS (trimmed least squares) and, in general, all trimmed estimators weight with 0 or 1 the ordered observations. There are some problems with this approach, mainly related to the difficulty in multivariate statistics to order the observations. The aim is to propose to change the concept of outlyingness, understood here as the low probability of obtaining the value of the estimator in the problem considered. These ideas are applied to define the novel distribution-weighted estimator in logistic regression. Its main properties are also studied.
