Although PCA is only a rotation of axes through a point cloud I've always
thought the algorithim assumed normalty, ot at least symmetry. I thought this
came from the fact that the mean is usually subtracted from all the variables
first (or am I getting PCA mixed up with factor analysis?).
However, recently I was taken to task over this by someone pointing out
that a rotation of axes through a point cloud doesn't need to assume
normality, and I think I agree with them.
Could someone please tell me if PCA assumes normality, and why? Or is it
just some of the eigenvalue tests for significance that require normality.
Christopher G Howden
(Work) 02 9895
(Mob) 0410 689 945