It is suggested that factor matrices be scaled for presentation—not just computation—so that the test vectors in the common factor space all become and remain of unit-length, the termini of which all lie on the surface of a unit hypersphere. This implies that factor matrices representing uncorrelated (“orthogonal”) factors are row-normalized. [Factor matrices representing correlated (“oblique”) factors would also have test vectors of unit-length, but the row vectors in such a matrix would not be normalized because of the nonorthogonality of such vectors.]
KaiserH. F. (1989) A possibly interesting intercorrelation matrix to accompany a varimax solution. Perceptual and Motor Skills, 68, 1076–1078.
3.
KaiserH. F.HunkaS.BianchiniJ. C. (1971) Relating factors between studies based upon different individuals. Multivariate Behavioral Research, 6, 409–422.
