Size and Power Properties of Asymptotically Robust tests for Equality of two Covariance Matrices

Abstract

Some asymptotically robust test statistics for equality of two covariance matrices are discussed. The standard error test based on combined and separate estimator of asymptotic covariance matrices of vectors of second-order Sample moments is estimated with and without transformations. The untransformed test based on separate estimator is equally good as Layard (1972, 1974) [3,4] proposed transformed test based on combined estimator. The effect of transformations on the tests is examined. The size and power perform of the untransformed tests is compared: The standard error test based on separate estimator is found reasonable for moderate size or non-normal samples.