Hi, I have some question regarding principal component and factor analysis.
For PCA, does it matter whether the eigenvalues are computed from the covariance matrix or the correlation matrix É And what about FA, are the results of the eigenvalues the same if I use the covariance or the correlation matrix É
PCA will be affected by rescaling of the data, so you will get different answers from the covariance versus the correlation matrix. FA (I assume you mean canonical FA) is not affected by rescaling, so it doesn't matter.
PCA results are affected by the units of the variables. Apart from that, if some variable's variance is much greater than the others, that variable tends to coincide with the first principal component.
A way to overcome those problems is using correlation instead of covariance matrix - provided that the differences in variances do not contain valuable information for the problem in hand.
The previous stand for FA also, if the type of factoring is "principal components". Conversely, if you use "maximum likelihood" factoring, the choice of either covariance or correlation matrix does not affect the results.