For binomial data with fixed and random effects, I have been using Proc Glimmix with the events/trials syntax, e.g., class block trt; model events/trials = trt/ solution ddfm=Satherth; random block/ group= block*trt; lsmeans trt/ adjust=tukey; However, I am wondering what the difference is from this syntax (difference bolded): class block trt; model state = trt/ solution ddfm=Satherth dist=binomial; random block/ group= block*trt; lsmeans trt/ adjust=tukey; where the state variable is a 1 or 0 in the dataset. Using the same data, I have tried both ways and get different output, specifically with regards to estimations of differences in Least Squares Means (esp. degrees of freedom) and the subsequent means separation letters. There are also differences in the co-variance parameter estimates. Why is this, and most importantly, how do I determine which syntax is most appropriate? Thanks!
... View more