topic Effectplot from PROC MIXED in Statistical Procedures

Effectplot from PROC MIXED

lauragrau — Thu, 10 Dec 2020 15:10:35 GMT

Hello,

I raw this model:

proc mixed data=grouping ;
class id b;
model y=time|b/outpred=Pred;
random intercept /sub=id type=un;
store out=MixedModel;
lsmeans b/ at time=(0) diff;
lsmeans b/ at time=(3) diff;
run;

I can use PROC PLM to visualize the effectplot:

proc plm restore=MixedModel;
effectplot fit(x=time plotby=b)/clm;
effectplot slicefit(x=time sliceby=b)/clm;
run;

Two questions:

1) Is there a way to change how PROC PLM's visuals look? I want to change the formatting!

I also tried this

proc sgplot data=pred;
band x=timelower=lower upper=upper/group=b transparency=.75;
series x=time y=pred/group=b;
run;

However, this graph looks different than the PLM graph.

Re: Effectplot from PROC MIXED

PaigeMiller — Tue, 10 Nov 2020 20:04:34 GMT

Show us the plots. Describe what you don't like. Provide screen captures by clicking on the "Insert Photos" and not by attachments.

Re: Effectplot from PROC MIXED

lauragrau — Tue, 10 Nov 2020 20:10:38 GMT

Unfortunately, I can't! It's protected data. Basically, there are only two time points. PROC PLM makes a super nice graph with two lines, clearly showing the interaction. The SGPLOT has vertical lines at 0 and 3, which are then connected.

Re: Effectplot from PROC MIXED

Rick_SAS — Tue, 10 Nov 2020 22:25:02 GMT

There are two kinds of predicted values. The OUTPRED= option outputs the predicted that incorporate the random intercept estimates for each subject. The OUTPREDM= option outputs the marginal (averaged) predictions. These are the same predicted values that you obtain from the STORE statement and PROC PLM.

So if you want to get the PROC PLM graph, use OUTPREDM=.

For details, an example, and more explanation, see "Visualize a mixed model that has repeated measures or random coefficients," especially the last section.