(Massey University, 2005) Chu, Ka Lok; Isotalo, Jarkko; Puntanen, Simo; Styan, George P.H.
We consider partitioned linear models where the model matrix X = (X1 : X2) has
full column rank, and concentrate on the special case whereX0
1X2 = 0 when we say
that the model is orthogonally partitioned. We assume that the underlying covariance
matrix is positive definite and introduce the efficiency factorization multiplier which
relates the total Watson efficiency of ordinary least squares to the product of the
two subset Watson efficiencies. We illustrate our findings with several examples and
present a literature review.
