You can, but be sure to include the random statements as in the current model, especially the group=group*compound option to deal with the heterogeneity of variance issue.  I don't know why you would drop it, though. It is a known design factor, the F tests, lsmeans and lsmean difference p values for tests of group, compound and group by compound are no different, and you get some insight into which whole plot factors have the largest effect on the repeat to repeat variability.   SteveDenham ... View more