Date:         Fri, 17 Oct 2003 11:12:56 -0400
Reply-To:     "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
Subject:      Re: R-square from a cubic regression
Content-Type: text/plain

Just a few opinions on it:
1) Why do you need to use GLM not REG?
        Unless I'm missing something, you'll have to define x2 = x**2 and x3
= x**3 in a data step, then you can do:
                Proc reg  data=datafile outest=outfile;
                model y = x1 x2 x3;

2) If you do want to use GLM:
        You're trying to get the R2 in output, I think that's the question?
        You can do an ODS statement -
                 ODS output FitStatistics = outfile;
        in the GLM code and that will output the R2 to the output file.

3) You can do a similar ODS statement with Proc Reg.

Hope this helps!
-Venita


> ----------
> From:         Nigel Pain[SMTP:Nigel.Pain@SCOTLAND.GSI.GOV.UK]
> Sent:         Friday, October 17, 2003 10:58 AM
> Subject:      R-square from a cubic regression
>
>
> The following question is from a statistician that I provide IT support
> to.
> It doesn't mean a thing to me so I'm just reproducing her query word for
> word:
>
> Simple linear regression we use PROC REG. The model statement is the y=x.
> So its fitting Y=Mx+c
>
> PROC REG DATA=work.reg11
>          OUTEST=work.ncon4l(KEEP=year subject level intercept avoth _rsq_
>                              RENAME=(avoth=slope)
>                              LABEL='Concurrent analysis at S4 by subject
> and level') EDF;
>    MODEL award=avoth ;
> RUN;
>
>
>
> With the use of the Simple linear regression OUTEST option in PROC REG you
> get the following which is the estimate of the slope and intercept and the
> _RSQ_ VALUE.
> So this is really good
>
>                                                  data from PROC REG
> -OUTEST
>
>
>                                                Obs    Intercept     slope
> _RSQ_
>
>                                                 1      0.36136     1.01971
> 0.65916
>
>
> However we need to fit a cubic regression with quadratic and linear terms
> and the intercept!! So need to use PROC GLM as PROC REG is not sufficient.
>
> PROC REG syntax goes as follows where the terms B1 is the linear term, b2
> is
> the quadratic term and B3 is the cubic term
>
> Y= b1x+ b2x +b3x +c
>
> proc glm data=work.reg11  outstat=work.model1 ;
>
> model AWARD=b1 b2 b3 /solution ss1 ;
> quit ;
> run ;
>
>
> I would really like the R-square value above and the estimates for
> intercept, b1, b2 and b3.
>
> The OUTSTAT datasets gives you :-
>
>
>                         Obs    _NAME_    _SOURCE_    _TYPE_       DF
> SS            F       PROB
>
>                          1     AWARD      ERROR      ERROR     58111
> 47596.19          .               .
>                          2     AWARD      b1         SS1           1
> 103556.33    126433.69             0
>                          3     AWARD      b2         SS1           1
> 5880.38      7179.46             0
>                          4     AWARD      b3         SS1           1
> 71.53        87.34    9.4628E-21
>
>
> I could calculate the r2 = 103556.33+5880.38+71.53/
> 103556.33+5880.38+71.53+47596.19=0.69, from above SS values but really
> need
> the estimates??!
>
> I'd like a dataset which comes automatically from the SAS PROCEDURE which
> will contain the estimates and the R squared value from the cubic
> regression
> analysis.
>
> Currently I read in the output as a text file, and find the appropriate
> columns!!!! I've also tried PROC MIXED!!!!
>
>
> ***************************************************
> Nigel Pain
> Scottish Executive
> Analytical Services Team
> Victoria Quay
> EDINBURGH
> EH6 6QQ
> UK
> Tel +44 131 244 7237
>
> Mailto:nigel.pain@scotland.gsi.gov.uk
> <Mailto:nigel.pain@scotland.gsi.gov.uk>
> Website: http:\\www.scotland.gov.uk <http:\\www.scotland.gov.uk>
>