(Received October 14, 1996; revised November 17, 1997)
Abstract. Our main concern is about second order admissibility under mean squared error. A sufficient condition and a necessary condition for a modified maximum likelihood estimator to be second order admissible regardless of parametrization are obtained. In addition, some procedures for characterizing such estimators are provided.
Key words and phrases: Second order admissibility, parametrization invariance, mean squared error.