05-13-2017 04:54 AM - edited 05-13-2017 04:56 AM
I am trying to graph the probabilities of an event happening in proc GLIMMIX. Does anyone know if the GLIMMIX procedure generates this type of graph?
Basically, I want to plot an S curve, as one does in logistic regressions.
05-13-2017 06:09 AM
I don't think GLIMMIX generates a FitPlot directly, but you can use the OUTPUT stmt to create an output data set that has the predicted values, then use PROC SGPLOT to create the graph. The trick is that you need to specify what sort of predicted values you want from (BLUP or NOBLUP) and (ILINK or NOILINK). See the GLIMMIX doc for what these terms.
Here is code for an "S-like curve" that you can study and modify:
proc glimmix data=sashelp.cars(where=(origin^='Europe')); model Origin = mpg_city / dist=binary; output out=modelout pred(noblup ilink)=prob lcl(noblup ilink)=lcl ucl(noblup ilink)=ucl; run; proc sort data=modelout; by mpg_city; run; proc sgplot data=modelout; band x=mpg_city lower=lcl upper=ucl; series x=mpg_city y=prob; yaxis min=0 max=1; refline 0 1 / axis=y; run;