Dear all, I am doing unblanced longitudinal analysis (the number of individuals declines with time). I want to consider the decreasing cases with time by adding a weight which assumes a different variance function at each time. However, if i run 'm1' (shown below), the error message ' weights must be like glm weights for generalized case' is shown. If i delete 'family=poisson' which is 'm2', the model works. m1<-gamm(y<-s(t1)+t2,random=list(id=~1),data,weights=varIdent(form=~1|t1),family=poisson,methods='REML') m2<-gamm(y<-s(t1)+t2,random=list(id=~1),data,weights=varIdent(form=~1|t1),methods='REML') Does anybody know how to keep 'weights=varIdent' working in gamm model with poisson or binomial family without specifing a numeric weights (because i dont know the real population variance)? Or any suggestions? Thank you very much in advance. I will be really appreciated for your time and help. Best wishes, He