Hello, I have several clinical data set with binary response. For all data sets, subjects are enrolled from different sites and are taking either treatment A or B (parallel design). I am trying to assess the site variation for the response variable. Since there are 2 treatment, I assume that the site variation could be different for the 2 treatments and there could be a correlation. The following is my SAS code: proc glimmix data=dset; class sitenum trt; model resp= trt/dist=bin link=logit solution; random trt/SUBJECT=SITENUM type=UN solution G; run; For most data sets, Proc Glimmix doesn't converge. For one data set, Glimmix does converge and I got the following covariance matrix. I think here 0.00907 is the site variance of treatment A and 3.54E-18 is the site variance of treatment B. However, the covariance between treatments A and B are even bigger than the both variance, so the correlation will be >1. Do I understand it correct? I don't know why it can happened. Estimated G Matrix Effect Description of Actual Arm Row Col1 Col2 trt A 1 0.00907 0.02008 trt B 2 0.02008 3.54E-18
... View more