ABSTRACT
Similarly for the oblique factor model with cov(Y)=R (correlation matrix) we obtain the following theorem.
Theorem 12.3.2. The maximum likelihood estimates of ? , R, D, respectively, for the oblique factor model are given by
(1)
(2)
(3)
For numerical evaluation of these estimates, standard computer programs are now available (see Press, 1971). Anderson and Rubin (1956) have shown that as N? 8 ,
has mean 0 but the covariance matrix is extremely complicated. Identification. For the orthogonal factor analysis model we want to represent the
population covariance matrix as
