[R-sig-ME] heteroscedastic models in lmer

[mikhail matz]

Apologies for repeated bugging on the same issue. I feel that I did not express my question well enough in the previous posts. Here I go again:

Is it possible to fit a heteroscedastic model in lmer, that is, a model in which the residual variances would differ between factor levels? More importantly, is there ever a need for that, given the way lmer's fitting algorithm works? 

These factor-specific variances are mere nuisance, I have no interest in quantifying them or even establishing their presence (even though in my case it is reasonable to expect that they would be there). So if lmer's estimates of fixed, as well as random, effects are robust to this type of heteroscedasticity, and there is a reference in the literature to back that up, I would greatly appreciate a tip. Of course, if not, please advise...

Misha Matz