With proc mixed one can specifiy a covariance structure for ranomd effects that has the same structure but with different values for each level of a categorical variable. For example, suppose Time is continuous, Period is a 0/1 variable that signals the transition from one period of time to another, SubjectID identifies the indiviudal subjects. Using proc mixed we can fit a random slope and intercept model where the covarnace struture is different for perods 0 and 1 wiht the following code. Proc Mixed data=D; class SubjectID Period; model y = Time/s; random intercept time/type=un subject= SubjectID group= period; run; Because period changes within subjects, the covariance matrix for the random effects is block diagnonal. Mixed is used as an example but for the model I really want to fit I will need NLMIXED. Can anyone explain how this covariance strucutre can also be implemented with NLMIXED? Thanks, Greg
... View more