topic Re: PCA standardization of input variables in SAS Programming

PCA standardization of input variables

RebeccaB_ — Tue, 10 Nov 2020 20:49:43 GMT

Hi,

I have a conceptual question about principal component analysis in SAS as compared to R. In R you must write the syntax PRCOMP to manually standardize the input variables. Does SAS automatically do this for you in the PROC FACTOR procedure?

Thank you

Re: PCA standardization of input variables

PaigeMiller — Tue, 10 Nov 2020 21:03:22 GMT

PROC PRINCOMP, by default, standardizes your variables. Or to say it another way, it works from the correlation matrix from your data. You can override this and use the COV option to work from the covariance matrix.

I don't use PROC FACTOR, you should check the documentation, but I believe it is similar.

Re: PCA standardization of input variables

Reeza — Tue, 10 Nov 2020 21:02:36 GMT

It does not, you can use PROC STDIZE to standardize the data ahead of time. Standardization requirements may vary slightly depending on the method used...and I'm not sure if you're doing it based on correlation versus covariance matrix if it matters.

Re: PCA standardization of input variables

RebeccaB_ — Tue, 10 Nov 2020 21:20:20 GMT

Thank you for your quick response. This is the code I used:

Proc Factor data= data-set-name

out= data-set-name

nfact=37

simple

method=prin

priors=one

mineigen= 0.99

scree

rotate=varimax

round

flag=.40;

var variables-to-be-analyzed

run;

So even using METHOD=PRIN that will not standardize our variables to make sure I'm understanding you correctly?

Re: PCA standardization of input variables

PaigeMiller — Tue, 10 Nov 2020 21:34:19 GMT

According to the PROC FACTOR documentation

COVARIANCE
COV

requests factoring of the covariance matrix instead of the correlation matrix. The COV option is effective only with the METHOD=PRINCIPAL, METHOD=PRINIT, METHOD=ULS, or METHOD=IMAGE option. For other methods, PROC FACTOR produces the same results with or without the COV option.

So if you don't use the COV option, you get the correlation matrix, which is created using standardized variables.

Re: PCA standardization of input variables

RebeccaB_ — Tue, 10 Nov 2020 21:44:29 GMT

Okay, to make sure I'm understanding correctly (I'm pretty much self-taught in PCA) because I specified METHOD=PRIN, and I did not use the COV option, that the input variables are standardized. We are using the correlation matrix (which I see in my output) and that is using standardized variables. Is that correct? So I would not need to change my code to make the input variables standardized?

Re: PCA standardization of input variables

PaigeMiller — Tue, 10 Nov 2020 21:47:52 GMT

@RebeccaB_ wrote:
Okay, to make sure I'm understanding correctly (I'm pretty much self-taught in PCA) because I specified METHOD=PRIN, and I did not use the COV option, that the input variables are standardized.

Yes

We are using the correlation matrix (which I see in my output) and that is using standardized variables. Is that correct? So I would not need to change my code to make the input variables standardized?

Yes