[R-sig-ME] GLMM & lack of linearity on the logit

[Fabian Scheipl]

Simon Blomberg:
> If there is important nonlinearity in egg volume, you could
> consider a Generalized Additive Mixed Model (GAMM) with a smoothing
> term for egg volume. I recommend package mgcv.

I second that, and I would add that you really can't lose by including
a smooth term for important continuous variables as long as you have a
decent number of observations: if the relationship is indeed linear,
the smooth term will be estimated to be (close to) linear as well and
you can then proceed to simplify your model.

In terms of software to fit this, also have a look at my package amer
(http://cran.r-project.org/web/packages/amer/index.html),
which avoids the PQL-approximation used in mgcv and instead relies on
lme4's Laplace integration to fit generalized additive mixed models.