topic Re: Principal Components Regression in SAS Programming

Principal Components Regression

gstratis — Wed, 21 Feb 2018 12:15:13 GMT

Hello to everyone,

I have carried out principal components regression using the PROC PLS procedure with 'method=pcr'. I have concluded that the number of principal componentns to keeo is 2.

My question is:

Can I derive a model for my response variables and the 2 principal components I have found? If yes, waht is the command I should use in order to derive the coefficients for each PC? I have the intercept of the model.

Regards,

George

Re: Principal Components Regression

PaigeMiller — Wed, 21 Feb 2018 13:21:43 GMT

The SOLUTION option in the model statement will give you the model terms.

As a side issue, it was never clear to me why PCR was an option in PROC PLS, as the two component PLS model should always fit as well or better than the two component PCR model.

Re: Principal Components Regression

gstratis — Wed, 21 Feb 2018 14:04:58 GMT

The thing is that I have to carry out different methods of Modern regression with a data set where the predictors are highly collinear. I have also carries Partial Least Squares Regression and the objective is to compare the different kinds of regression.

Hope this clears your question.

George

Re: Principal Components Regression

PaigeMiller — Wed, 21 Feb 2018 14:10:18 GMT

It's kind of hard for me to accept that PCR is a "modern" regression method, I would claim it is an antique method that should not be used in the 21st century. Nevertheless, you might want to take a look at the paper by Frank and Friedman, who did such a comparison of many different regression methods.

Re: Principal Components Regression

gstratis — Wed, 21 Feb 2018 14:11:55 GMT

I totally understand what you are saying but I only follow orders here. Nevertheless thank you for your advice.

George

🙂

Re: Principal Components Regression

Reeza — Wed, 21 Feb 2018 15:47:48 GMT

but I only follow orders here

I didn't think AI had come that far...