I've done it after bootstrapping eigen coefficients so that I have some
confidence in a rejection of isometry when dealing with multiple groups
(Common PCAs). I've got an unpublished manuscript (revising it now
actually) explaining my methods if you would like to see it. But, you
can probably als
I've seen this "normalization" before (although I don't remember where -
probably in 1980s physical anthropology papers). As Dr. Rohlf notes,
the only real point of dividing PC1 eigenvector coefficients by the
isometric expectation (1 over sqrt of the number of measurements) to put
them into more
I guess what is meant is the fact that if you compute a normalized PC1
from a correlation matrix based on p variables for data in which PC1
represents size and growth is isometric then the loadings will all be
equal to 1/sqrt(p). That is because "normalized" means that the sum of
the squared elemen
