A Monte Carlo Study of Robust Location Estimates with Non-gaussian Multivariate Data

Four robust estimators of multivariate locations are compared in a Monte Carlo study by examining their empirical efficiency performance on a range of bivariate distributions with heavier tails than the Gaussian distribution. Using three different multivariate measures of relative efficiency, the simulation results show that the Hodges-Lehmann estimator is "safest" among the robust estimators, in that it has uniformly high efficiencies across distributions ranging from the uniform distribution on a square to the bivariate Cauchy distribution.