02-19-2016 01:44 PM
I've a GLIMMIX question in terms of how to specify the random effect. Specifically I've three-level data, with students nested in school, then schools nested in regions. And I've three level-1 variables and two level-2 variables - and level 2 variables are especially of interest. A random intercept model for level 2 and level 3 would do. Random intercept and random slope model is ideal. My outcome variable is not linear, so I used Gamma distribution with link=log.
What confused me is how I should let SAS know which variable are level 2 variables? What variables are for identifying different levels (StudentID, SchoolID, GeoID) in the random statement. How do I use the random statement to indicate levels? Where do I put the level 2 variables - if I assume 1) fix slope, or 2) random stlop.
I was told I have to have 3 RANDOM statement: /GeoID, then subject=SchoolID(GeoID), then subject=StudentID(SchoolID, GeoID), as below (but the model wouldn't run as it won't converge or ran out of memory), corrrect?
proc glimmix data=mydata METHOD=LAPLACE NOCLPRINT;
class PatientID HospID GeoID L1_var1 L1_var2 L1_var3 /*level 1 categorical vars*/ L2_var1 /*level 2 categorical vars*/
model Y = L1_var1 L1_var2 L1_var3 /*level 1 variables*/ L2_var1 L2_var2 /*level 2 variables*/
/dist= gamma link=log solution;
random intercept / subject = GeoID type=un solution;
random intercept / subject = SchoolID (GeoID) type=un solution;
random intercept / subject = StudentID(SchoolID GeoID) type=un solution;
What's the differences when I just have
RANDOM intercept /Subject=GeoID; RANDOM intercept /Subject=SchoolID(GeoID), or
RANDOM intercept /Subject=GeoID; RANDOM intercept Subject=SchoolID
If I want to assume random slope for level 2 variables, do I just
RANDOM intercept /Subject=GeoID;
RANDOM intercept L2_var1 L2_var2 / Subject=SchoolID
Sorry for those very basic GLIMMIX questions. Any reply is appreciated.Thanks.