Interval estimation in discrimination: The multivariate normal equal covariance case

SUMMARY A general approach to discrimination problems is described which emphasizes the viewpoint of individual cases and expresses the uncertainty with which individual decisions are made. It is applied to the particular case of two multivariate normal distributions with a common covariance matrix. An affine transformation, to which the key quantities are invariant, simplifies discussion of the distribution of the estimated log odds ratio and suggests a practically useful plot which generalizes the usual display of discriminant scores and has several attractive features. An illustration is given. Two approximate interval estimation techniques for the true log odds ratio are described and evaluated in a range of practical situations by a simulation study. Some further applications of our general approach are briefly indicated.