A GEOMETRIC APPROACH OF THE GENERALIZED LEAST-SQUARES ESTIMATION IN ANALYSIS OF COVARIANCE-STRUCTURES

In this paper, the generalized least-squares estimation of the covariance structure model is studied from a geometrical point of view. General definitions of the intrinsic curvature and the parameter-effect curvature are defined for the model. Based on the general result, the second-order approximations of the bias and the covariance matrix of the generalized least-squares estimator are established. The information loss of the estimator is also computed under the multivariate normal assumption.