My responses assume that I understand your study design correctly. I think I do, but there also could be some important detail that you have not yet provided. It does seem that year (and gene) are whole plot factors in this study, and that treatment (and sex) are subplot factors.   The second error term can be specified as the interaction of the whole plot unit and the subplot factor(s).    For easy reference, here is the model I suggested previously:   proc glimmix data=survival method=laplace; class year pen gene sex treatment; model survival6w (descending) = year gene treatment sex / dist=binary; random intercept treatment*sex / subject=pen(year gene); lsmeans year gene treatment sex / ilink; run; For your study, the whole plot error is variance among pens, specified by the combination of intercept and subject=pen(year gene) in the RANDOM statement. The subplot error is variance among groups of animals within pens treated alike, specified by the combination of treatment*sex and subject=pen(year gene). An equivalent syntax for the RANDOM statement that you might find more intuitive is   random pen(year gene) treatment*sex*pen(year gene);   treatment*sex*pen(year gene) is an interaction of the subplot factors treatment and sex and the whole plot unit pen(year gene).   I'm not particularly enthused about the results from the model above, but the data are so sparsely allocated across the design and most animals survived, so there is just not much to work with.    ... View more