Abstract
Given a set of data pxN$_{i}$, matrices X$_{i}$ observed from p-variate normal populations $\prod$$_{i}$~N($\mu$$_{I}$, $\Sigma$$_{i}$) for i=1, …, K, the test for equality form of the covariance matrices is to choose a hypothetical model which best explains the homogeneity/heterogeneity structure across the covariance matrices among the hypothesized class of models. This paper describes a test procedure for selecting the best model. The procedure is based on a synthesis of Bayesian and a cross-validation or sample reuse methodology that makes use of a one-at-a-time schema of observational omissions. Advantages of the test are argued on two grounds, and illustrative examples and simulation results are given.are given.