Sample Size Determination in Bayesian Discriminant Analysis

Abstract

Abstract The preposterior analysis of the two-population classification problem for multivariate normal densities with equal covariance matrices is presented. Consideration is given to the case that arises when the common precision matrix is known and the mean vectors have normal prior densities, and to the case that arises when the precision is unknown and the prior densities are normal-Wishart. The expression for the preposterior expected loss is used to find the expected value of sample information and hence the optimal sample size.