Dear R users, I want to fit a model with within-subject correlations but with no random effects. So I could use gls to set correlation=corSymm(~1 | group). Here the problem I encountered is that I have 90 groups, and for the first, second and third 30 groups I want them to have three different unstructured correlation structures. Can I achieve that in gls() of the nlme package? Thank you! [[alternative HTML version deleted]]