topic Re: Proc factor: principle component scores in Statistical Procedures

Proc factor: principle component scores

Eileen18 — Tue, 28 Aug 2018 13:30:51 GMT

Hi! I am using proc factor to run a principle components analysis (to note: I went with this procedure over princomp bc I wanted more of the output...if i need to switch to princomp then I will!)

Here is my question: based on the SAS documentation below, I am confused whether proc factor produces principle component scores (discussion accompanying 33.1.8) or if one has to use proc princomp to have the principle component scores (discussion accompanying 33.1.19).

https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_factor_sect028.htm

Again, per this documentation, it seems that the factor1 generated automatically by proc factor is not equivalent to principle component scores (33.1.8) but I'm confused as this seems to be the correct equation. Caution against using factor1 if one is wanting to use the components in a subsequent regression model? or is it correct?

thank you SO much for the support!

Re: Proc factor: principle component scores

WarrenKuhfeld — Tue, 11 Sep 2018 14:35:46 GMT

When in doubt, ask the procedures themselves. These two analyses produce the same scores.

proc factor data=sashelp.iris out=f nfactors=4 prefix=Prin; run;

proc princomp data=sashelp.iris out=p n=4 std; run;

proc compare error note briefsummary criterion=1e-10
   data=f(keep=prin: label='') compare=p(keep=prin: label='') method=relative(1);
run;

Re: Proc factor: principle component scores

PaigeMiller — Tue, 11 Sep 2018 15:11:18 GMT

Again, per this documentation, it seems that the factor1 generated automatically by proc factor is not equivalent to principle [sic] component scores (33.1.8)

That link comes to the conclusion that is the opposite of your conclusion. It says: "As you can see, these standardized scoring coefficients are essentially the same as those obtained from PROC PRINCOMP, as shown in Output 33.1.9. This example shows that principal component analyses by PROC FACTOR and PROC PRINCOMP are indeed equivalent."

It also says: "If a principal component analysis of the data is all you need in a particular application, there is no reason to use PROC FACTOR instead of PROC PRINCOMP."