An unbiased likelihood ratio test for equality of the covariance matrices in several multivariate normal populations with partially known means (Q1083154)

Let samples from d multivariate normal populations be given with unknown covariance matrices \(\Sigma_ k\), \(k=1,...,d\), and with the mean of the ith sample \((i=1,...,n_ k)\) in the kth population given by \(B_ kz_{k,i}+a_{k,i}\) where \(B_ k\) is unknown and \(z_{k,i}\) and \(a_{k,i}\) are known. With unbiased estimates \(Q_ k\) of \((n_ k-r_ k)\Sigma_ k\), for \(k=1,...,d\) where \(r_ k=rank(z_{k,1}:...:z_{k,n_ k})\), the test which rejects the hypothesis of equality of the covariance matrices for large values of the test statistic \[ | \sum^{d}_{k=1}Q_ k|^{(n- r)/2}/\prod^{d}_{k=1}| Q_ k|^{(n_ k-r_ k)/2}, \] where \(n=\sum^{d}_{1}n_ k\) and \(r=\sum^{d}_{1}r_ k\), is unbiased against all alternatives.