08-30-2011 05:43 PM
Hi; I'm working in a multi-center cohort study and I'd like to control for the effect of study center. I identified Proc SurveyLogistic as one possible approach and ran this model;
proc surveylogistic data=one ;
class gender disease_final (ref='0') exposure (ref='exp-') /param=ref;
strata ccenter /list;
model cvd(event='1') = age_enroll ats_packyears disease_final gender exposure;
my statistician asked me how the ccenter variable was being handled in the strata statement and I cannot locate an answer in the SAS documentation. I assumed that it would be a random effect and that I could duplicate that model with GLIMMX like so;
proc glimmix data=one ;
class gender disease_final exposure ccenter;
model cvd(event='1') = age_enroll ats_packyears disease_final gender exposure /dist=binomial link=logit solution;
This yealds estimates and p values that are similar but not the same. So could someone please direct me the technical documentation that describes how the strata statement handles the variable it's handed?
Thanks in advance.