Geek Answers Handbook

The main component and factor analysis

I have a question regarding the analysis of the main components and factors.

For the PCA, it matters whether the eigenvalues ​​are calculated from the covariance matrix or the correlation matrix É A As for FA, then the results of the eigenvalues ​​are the same if I use the covariance or the correlation matrix É

+4
source share
2 answers

PCA will be affected by data rescaling, so you will get different answers from covariance and correlation matrix. FA (I assume that you mean canonical FA) does not affect scaling, so it does not matter.

+1
source

Units of variables influence PCA results. In addition, if any dispersion variable is much larger than the other, this variable tends to coincide with the first major component.

A way to overcome these problems is to use correlation instead of the covariance matrix - provided that the variance differences do not contain valuable information for the problem at hand.

The previous position is also for FA, if the factoring type is the "main component". Conversely, if you use “maximum likelihood” factoring, the choice of covariance or correlation matrix does not affect the results.

+1
source

More articles:


All Articles