Testing for variance homogeneity of correlated variables

Abstract

Four large-sample methods are proposed for testing the equality of the variances of a p -variate normal population. Two of the methods are asymptotically distribution robust against departures from the multinormality assumption of the parent population. In applying the tests no restrictive assumptions are made about the off-diagonal structure of the covariance matrics and no random splitting of the sample into subsamples is required. An example of the use of the tests is provided.