Date:         Fri, 19 Oct 2007 15:17:03 -0000
Reply-To:     ginanicolosi@HOTMAIL.COM
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Gina Nicolosi <ginanicolosi@HOTMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: Use all numeric variables as predictors in proc reg
Comments: To: sas-l@uga.edu
In-Reply-To:  <1192567485.658959.211720@k35g2000prh.googlegroups.com>
Content-Type: text/plain; charset="us-ascii"

On Oct 16, 3:44 pm, ginanicol...@hotmail.com wrote:
> I am writing a program which randomly selects predictor variables to
> be included in a regression.  Since the predictors during any one
> iteration can change, I can't write them out in the MODEL statement.
> Therefore, I was wondering if there was any way to specify that all
> numeric variables in the data set act as regressors? Kind of like
> putting _ALL_ or _NUMERIC_ after the equals sign when running PROC REG
> (but this approach doesn't work).  Any guidance would be greatly
> appreciated.

Thanks so much for the suggestions.

At first, I found a solution using IML, but since I've never used that
PROC before, I had to canibalize someone's posted IML regression code,
and I wasn't terribly comfortable with running something I didn't
understand.

I am also embarrassed to say that my experience with macro programs
and variables is also limited.  However, I was able to work the code
below (written by Stuart Long) into the macro I wrote:

data whatever;
if 0 then set mydata;
length numerics $32767;
array num_vars {*} _NUMERIC_ ;
do i = 1 TO DIM(num_vars);
numerics=CATX( " " , numerics , VNAME(num_vars{i}) );
end;
call symputx("regressors",numerics);
stop;
run;
%let dependent = y;
%let regressors= %SYSFUNC(TRANWRD(&regressors,&dependent,));
proc reg data=mydata;
model &dependent = &regressors ;
run;
quit;

Finally, thank you Howard and Mary for your suggestions.  I have never
used DICTIONARY.COLUMNS before, so I will investigate what it does.

Again, thanks so much! - gina