R individual pca scores different than SAS and SPSS when scale is applied

2
votes

I had previously posted this as a question to my original question regarding scaling but this was deemed inappropriate and deleted with a request to submit as a new question, so here we are!

Myself and my colleagues were performing PCA using r (prcomp), SAS, and SPSS. We all used scale and centering. All three produced the same variable/factor loading (different terminology between programs). No issue there.

However, the individual’s scores produced in R are different than SAS and SPSS. When I applied scale(to.my.pca) the results were the same as SAS and SPSS. So, it appears that scale.=TRUE in R isn’t passing to the individual scores. Can anyone using r and SAS or r and SPSS confirm this?

Many thanks!

1
rsaspcaspss

1 Answers

0
votes

R, SAS, and SPSS use different algorithms to find optimum coordinate system to set up the Eigen values and corresponding Eigen vectors for PCA. Thus, results could vary from software to software.

For example, in a 2 dimension coordinate system (applying PCA on two features) rotating x-axis counter clockwise or rotating y-axis clockwise can give you different Eigen value / vectors hence the features of PCA.