Date:         Wed, 27 May 1998 21:14:46 GMT
Reply-To:     jshi@KUHUB.CC.UKANS.EDU
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Jingren Shi <jshi@KUHUB.CC.UKANS.EDU>
Organization: Deja News - The Leader in Internet Discussion
Subject:      Re: Model selection in PROC REG

In article <35696FA2.2F06@ukc.ac.uk>,
  "M.M.Iqbal" <mmi1@ukc.ac.uk> wrote:
>
> Hello all,
>
> I am using following code to select a model of size 7 using
> SELECTION=RSQUARE in model statement.
>
> proc reg data = dataset;
>   model y =  list of independent variables/selection=rsquae
> start=7                   stop=7;
> run;
>
> It works for me and give me a list of models in descending order of R^2.
> In my particular case first model is reported to have R^2=0.89. When I
> try to fit the selected model using PROC REG or PROC GLM it gives me R^2
> as low as about 0.29.
>
> I am wondering what makes SAS to calculate different R^2 for the same
> model.
>
> I tried to figure out the reason behind this difference by calculating
> residuals for every intermediate model but remained fail to get it as
> use of OUTPUT OUT=MYOUTPUT P=PREDICT R=RESID; gives the information for
> the last model only.
>
> Any help would be appreciated.
>
> Iqbal
>
 Iqbal,
I did a sample dataset and ran the approach you did. I got the same
 result --- R-sqs are the same. If you can e-mail me your data, I would
 like to take a look.
Good luck.
 
data t1;
  do i = 1 to 100;
    x1=i+sin(i);
    x2=-sin(i);
    x3=30+2*cos(i);
    x4=30+i**2;
    y= 5+3*i+rannor(0);
    output;
  end;
run;
 
proc reg data=t1;
l1: model y = x1 x2 x3 x4/ selection=RSQUARE
    start=2  stop =2;
quit;
run;
 
proc reg data=t1;
l1: model y = x1 x2;
quit;
run;
 
proc glm data=t1;
   model y = x1 x2;
quit;
run;
 
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/   Now offering spam-free web-based newsreading