[R-sig-ME] Random vs. fixed effects

[dave fournier]

I think the point is that lme only addresses the simplest case of a
linear mixed model and that by such specialization it is more effective
in this case. lmer attempts  to handle slightly more complex mixed
models and must use a different algorithm and this algorithm
seems to fail. AD Model Builder on the other hand uses
a much more general algorithm intended to work with arbitrary nonlinear
mixed models. To stick with the current simple example suppose that you
want to have the regression be robust such as assuming that the random
variable e has a fat-tailed student's distribution. This can easily be
done in ADMB, but not in lme, I believe.