ADMISSIBLE LINEAR ESTIMATORS IN A LINEAR MODEL WITH THE NATURAL PARAMETER SPACE

It is known that the linear estimation, both with or without unbiasedness, may be reduced to a statistical game with a convex compact parameter set. Then all locally best estimators constitute a complete class and each locally best estimator being unique is admis sible. However, if the considered locally best estimator is not unique, then all known sufficient conditions for the admissibility work very hard. We derive a simpler sufficient condition for the admissibility in a linear model with the natural parameter space.