Robustness of the Minimum Norm Quadratic Unbiased Estimator of Variance in Linear Model

摘要

Robustness of the minimum norm quadratic unbiased estimator of variance in terms of error distributions is discussed in linear model. We explore the maximal distribution class of error term, where the minimum norm quadratic unbiased estimator of variance σ2　holds its optimality. Furthermore considering robustness of the best linear unbiased estimator of estimable function χβ, the maximal distribution class of error term is obtained, where the minimum norm quadratic unbiased estimator of variance and the best linear unbiased estimator of χβ keep optimality simultaneously.