topic Re: proc reg - assessing models with blocks of variables in Statistical Procedures

proc reg - assessing models with blocks of variables

mmjohnson — Wed, 10 Feb 2016 17:06:15 GMT

Hi,

I am trying to run a regression analysis in SAS using proc reg with two sets of variables (second set includes the first set). But I don't want to use selection=stepwise because I don't want any variables discarded based on their contribution. I use SAS 9.4

Basically want to first run

model intention = var_a var_b var_c

and then run another model

model intention = var_a var_b var_c var_d var_e

and see the change in R2, F, significance between the two models.

I ran it in SPSS (output attached) and this is what I want to get in SAS too. Thank you.

Re: proc reg - assessing models with blocks of variables

SteveDenham — Wed, 10 Feb 2016 20:27:10 GMT

Since PROC REG is interactive, you can include as many model statements as you want. The only thing you will not get in the output is a likelihood ratio test, but that can be calculated from the Type I sums of squares. See the following:

http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_reg_examples03.htm

In that example there are only two variables entered into one of the two model statements, and one into the other, but you should be able to easily expand this example to your code.

Steve Denham

Re: proc reg - assessing models with blocks of variables

mmjohnson — Wed, 10 Feb 2016 21:23:07 GMT

Thank you, that is helpful. But I was also looking to get a table showing the change in R-squared, F-statistic and p-value between the two models. Do you know where I could find a reference for that option?

Re: proc reg - assessing models with blocks of variables

SteveDenham — Mon, 15 Feb 2016 19:32:20 GMT

ODS output statements can put all of the fit statistics into a dataset, and from there you can do about anything.

Steve Denham