Percentage points for a test of rank in multivariate components of variance

Abstract

In the balanced one-way multivariate analysis of variance with random effects the likelihood ratio criterion for testing that the rank of the covariance matrix of the effect vector is not greater than a specified number is considered. The limiting distribution of −2 times the logarithm of the criterion is the distribution of the sum of squares of the positive characteristic roots of a certain random symmetric matrix. This paper presents exact and simulated quantiles of this limiting distribution for a number of cases depending on the difference between the dimensionality and the specified rank.