arXiv.itsnick.co

an independent field guide to 23 million papers from arXiv & OpenAlex — not affiliated with arXiv.org

Mathematics · Statistics and Probability

Approved by.-/

Julio Da Motta Singer, Julio Da Motta Singer

See on the map

Abstract

direction of PRANAB K. SEN). Two types of M-estimators of the parameters of the multivariate linear model are considered: the first one is based on a coordinatewise scoring procedure while the second utilizes a scoring method based on a ~~halanobis-type distance. Under some general conditions on the error distribution, on the design matrix and on the score function, the asymptotic distribution of the coordinatewise 14-estimator is derived by means of an asymptotic linearity result. It is then indicated how a similar approach can be employed to obtain the asymptotic distribution of the second type of M estimators under the additional assumption of elliptical symmetry of the underlying error distribution. Robustness properties of both types of estimators are studied through their influence functions and breakdown points. Two families of robust M-estimators of the above types are compared through the Pitman asymptotic relative efficiency. Two types of M-tests for the general linear hypothesis are also suggested: one is a direct analogue of Wald's test and the other corresponds to a x2 test in which estimates under the more constrained hypothesis alone are required. Their asymptotic distributions are derived and it is shown that they are equi-efficient under local Pitman-type alternatives. It is indicated how the above ~l-methods can be extended to growth curve models and some more general linear models. Computational algorithms are proposed and the suggested procedures are illustrated by a numerical example with data from the statistical literature. •

More like this