Date:         Sun, 7 Sep 2008 17:10:01 -0400
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: interpreting SAS output
Comments: To: "globalreviewer@gmail.com" <globalreviewer@GMAIL.COM>
Content-Type: text/plain; charset=UTF-8

"globalreviewer@gmail.com" <globalreviewer@GMAIL.COM> wrote

>Thank you for helping me with your inputs. I am not expecting
>different SEs rather I am expecting 'similar' SEs once the values are
>back-transformed (not after transformation).
>

OIC, that makes more sense, but backtransforming doesn't restore the original values ... Not sure how dissimilar they 'ought' to be, though

>My mixed code is given below. I used log transformation on my data
>because they were not normally distributed:
>

The key is not that the *data* be normally distributed, but that the *residuals* are normally distributed.  But I see below that you seem to have done this.

>proc mixed data=PlotObs_Trans ;
>class Month Sex ;
>Model LogVRate=Month Sex Month*Sex  /solution ddfm=satterth
>outp=OutOldTrans1;
>Lsmeans month*Sex/slice=Month;
>Lsmeans month*sex/slice=sex;
>LSmeans Month Sex month*Sex/PDIFF CL ADJUST=TUKEY;
>Run;
>

this looks right to me, but others are more expert in MIXED coding.


>Data OutOldTrans1;
>Merge PlotObs_Old_Trans OutOldTrans1;
>aresid = ABS(resid);
>Drop StdErrPred Lower Upper;
>Run;
>Proc plot data=OutOldTrans1 vpercent=50;
>plot LogVRate*Month LogVRate*Sex;
>Plot Resid*Month resid*Sex /vref=0;
>plot Resid*Pred aresid*Pred/Vref=0;
>Quit;
>

I know very little of proc plot ....

>TITLE7 'Examination of variance homogeneity using rank correlation';
>TITLE8 'between absolute residuals vs predicted values';
>PROC CORR SPEARMAN DATA=OutOldTrans1;
>VAR aresid Pred;
>QUIT;
>TITLE8 'Examination of normality of residuals';
>PROC UNIVARIATE NORMAL PLOT DATA=OutOldTrans1;
>/*tests for normality W shud be closer to 1 and p shud not be <.05*/
>VAR Resid;
>Quit;
>
>Also, what is the best method to show the interaction graphically..
>plotting the means from the raw data or the estimates from proc
>mixed ? I thought one would back-transform the estimates and the SE
>and plot them on the graph (???).
>

Since both your IVs are categorical, I think I'd draw a series of parallel box plots.  I think you said that one IV has 2 categories and the other has 6.  So, I would do a graph with 12 boxplots, in two rows of 6 each.

Not sure at all how to do that in SAS.  I do my graphics in R, usually

Peter

Peter L. Flom, PhD
Statistical Consultant
www DOT peterflom DOT com