Admissibility of Nonnegative Quadratic Estimators for Error Variance with Respect to Restricted Ellipsoidal Parameter Space

This paper gives a necessary condition for the admissibility of a nonnegative, quadratic estimator for error variance in linear model with respect to restricted ellipsoidal parameter space. Moreover, a sufficient and necessary condition is given for the special case where the design matrix X = 1_n=（1,1,…,1）’ in linear model.