Cochran theorems for a multivariate vector-elliptically contoured model. Part II

Let Y be a vector-elliptically contoured distributed n by p random matrix associated with a gamma (alpha, beta) random variable, where the mean of Y need not be zero and the covariance of Y may be singular and need not be the Kronecker product of the design covariance and the population covariance. A Cochran theorem about the quadratic forms of Y is obtained, when alpha=np and beta=2, it yields the most general Cochran theorem for a normal Y. (C) 1999 Published by Elsevier Science B.V. All rights reserved.