[R-sig-ME] GLMM - selection of models

[Philippi, Tom]

In addition to what Dr. Bolker wrote, note that for family=poisson, as you
add or drop fixed effects, there will be different amounts of
overdispersion among the models fits.  This can be an issue that affects
both the mechanics and interpretation of any form of model selection or
averaging.  May you be luckier with your data and overdispersion than I
tend to be with mine.

Tom 2


On Thu, Mar 16, 2017 at 2:31 PM, Ben Bolker <bbolker at gmail.com> wrote:

> There are strong theoretical arguments for **not** doing stepwise
> selection on models (any kind, at all), e.g.
> http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/
> .
> If you must, you can use `drop1()` to semi-automate the process, or
> you can use MuMIn::dredge to do all-subsets fitting.
>
> On Thu, Mar 16, 2017 at 5:15 PM, Marcos Monasterolo
> <mmonasterolo at agro.uba.ar> wrote:
> > Dear all. I am working with a GLMM in the lme4 package using 6 fixed
> > factors and 1 random factor (plot). The syntax of my model is as follows:
> >
> >  M1 <- glmer(riquinsec~  adjacent field+ exph200+db500+exph500+db200+
> width+
> > (1|plot), data = comuni1, family = poisson)
> >
> > I need to make a selection of models to get rid of non relevant
> variables,
> > but the "step" function in lme4 is not appropriate for GLMM models.
> > Which function can I use for a selection of models in GLMM?
> >
> > Thanks in advance for you kind help.
> >
> > ----
> > Biól. Marcos Monasterolo
> > Becario doctoral - Cátedra de Botánica General, Facultad de Agronomía,
> UBA
> >
> >         [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > R-sig-mixed-models at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]