Conditions under which a single iteration approximation to the maximum likelihood estimator dominates ordinary least squares are approximated analytically for the class of linear models for which the eigenvectors of the error covariance matrix are known.