I would also account for within subject correlation, with a repeated statement. My preferred code would be: proc glimmix data=hw5.imp1; class id time; model implant = time/solution; random time/residual subject=id type=un; run; This provides the most direct between-within subject approach. Under the unstructured covariance approach, a random subject effect is completely confounded with the residual parameterization. For other structures, such as autoregressive structures for equally spaced in time measurements, the random effect of subject can be captured as: proc glimmix data=hw5.imp1; class id time; model implant = time/solution; random intercept/subject=id; random time/residual subject=id type=un; run; Steve Denham
... View more
