Propostas de testes multivariados para comparar matrizes de covariâncias de populações normais dependentes

Abstract

A number of tests may be found in the literature to compare k-covariance matrices, assuming independent multivariate normal populations. For the cases that this assumption is violated there is a reduced number of such tests in the literature mainly with a small number of populations. The motivation of this work is to propose new tests based on generalized variances ratios for comparing covariance matrices of dependent normal multivariate populations. The covariance matrices were generated using frequentist (bootstrap) and bayesian theory. The tests that used bootstrap for generating the covariance matrices were more powerful than the bayesian approach for obtaining the covariance matrices.