Estimation of the Errors of Misclassification on the Criterion of Asymptotic Mean Square Error

Abstract

Several estimators of the errors of misclassification of the linear descriminant function (Anderson's classification statistic [1]) are compared on the criterion of asymptotic mean square error (AMSE). The AMSE predictions concerning the relative superiority of the estimators are found to coincide with the conclusions, based on Monte Carlo experiments, of Lachenbruch and Mickey [4]. It is concluded that the AMSE criterion gives a reliable assessment of the performances of the estimators in practical situations.