topic Re: How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby& in Statistical Procedures

How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby" options)?

jrbrauer — Sat, 04 Nov 2017 15:06:57 GMT

Hi folks,

I want to show predicted event rates for Y by X at two different values of Z1 and two different values of Z2 in a single plot using "effectplot" in proc genmod. I can plot at two different values of Z1 (or Z2) using the "sliceby" command, and I can compare these curves across two different values of Z2 (or Z1) across two different plots by adding a "plotby" command. Is there any way to stack or overlay these two plots in SAS? Conceptually, I think it would be equivalent to simultaneously requesting two "sliceby" commands, though that does not appear to be a possibility. Here is an example of my code:

ods graphics on ;
proc genmod data=tempd plots=(resraw resdev);
model prjdlqad = female age16p famsplit famstep famoth finsesza acachfs
impulsza allblfza morctxza impulsza*allblfza impulsza*morctxza allblfza*morctxza
/ dist=nb link=log; store nbmodel1; run;
ods html style=statistical ;
title "Impulsivity & Projected Delinquency, by Moral Beliefs & Context";
proc plm source=nbmodel1;
effectplot slicefit(x=impulsza sliceby=allblfza= -8.12 8.12 plotby=morctxza= -4.42 4.42)
/ CLM at(female=0 age16p=0 famsplit=0 famstep=0 famoth=0);
run;
ods graphics off;

Thanks in advance for any help!

Jon

Re: How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby"

Ksharp — Sun, 05 Nov 2017 10:12:00 GMT

One way I could think is useing ODS TRACE to get the name of graphic and using ODS OUTPUT to get their data and finally using proc sgplot to overlap the graphic.

Re: How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby&

jrbrauer — Mon, 06 Nov 2017 16:17:02 GMT

Thank you for your helpful reply. I do not have much experience with sgplot but I will give this a try.

If you are aware of an example of this being done, please share!

Re: How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby&

Ksharp — Tue, 07 Nov 2017 13:41:44 GMT

ods trace on;
          ods output FitPlot=plot;
          proc genmod data=sashelp.class;
          model sex=age weight height;
          effectplot fit(x=age);
         run;
  ods trace off;
        
         proc sgplot data=plot;
          band x=_xcont1 lower=_lclm upper=_uclm;
          series x=_xcont1 y=_predicted/lineattrs=(color=red);
          run;

Re: How do I overlay plots with effectplot command in proc genmod (e.g., multiple "sliceby&

jrbrauer — Tue, 07 Nov 2017 14:20:21 GMT

Thank you! This is exactly what I needed. After your initial post, I was able to save the predicted values using ods output. Then, I created a new "group" variable using the "_plotby" and "_group" variables and modified the sgplot code you provided here by adding "/group=z" after the series. Now I just need to learn how to tweak the colors! Here is the full code in case anyone else ever has this question:

ods graphics on ; ods trace on;
proc genmod data=tempd plots=(resraw resdev);
model prjdlqad = female age16p famsplit famstep famoth finsesza acachfs
impulsza allblfza morctxza impulsza*allblfza impulsza*morctxza allblfza*morctxza
/ dist=nb link=log; store nbmodel1; run;
ods listing style=statistical;
title "Impulsivity & Projected Delinquency, by Moral Beliefs & Context";
proc plm source=nbmodel1;
effectplot slicefit(x=impulsza sliceby=allblfza= -8.12 8.12 plotby=morctxza= -4.42 4.42)
/ CLM at(female=0 age16p=0 famsplit=0 famstep=0 famoth=0) ;
ods output SliceFitPanel = mygraph;
run;

data mygraph2; set mygraph;
if _plotby = "morctxza=-4.42" & _group = -8.12 then morfiltr = 1;
if _plotby = "morctxza=-4.42" & _group = 8.12 then morfiltr = 2;
if _plotby = "morctxza=4.42" & _group = -8.12 then morfiltr = 3;
if _plotby = "morctxza=4.42" & _group = 8.12 then morfiltr = 4;
run;

proc sgplot data=mygraph2;
title "Impulsivity & Projected Delinquency, by Moral Beliefs & Context";
band x=_xcont1 lower=_lclm upper=_uclm / group=morfiltr;
series x=_xcont1 y=_predicted/group=morfiltr lineattrs=(color=red);
run;

Thanks again. I really appreciate the help!

Jon