minimum AIC mixed model selection

SILVIA DIAZ FERNANDEZ <Silvia.Diaz <at> uclm.es> writes:

Hi!

I am trying to know which habitat variables most affect 
bird counts in a radius of 100m. I obtained bird
counts in 2751 spatial points, and measured percentage of 
21 habitat variables in these points.

I applied a mixed model using the "lmer" function to these data, 
but I do not know how to select the best model
using AIC here. Is there a way to do this automatically with R?

See the MuMIn package, and the "dredge" function.

e.g.:

library(MuMIn)
library(lme4)
example(lmer)
dredge(fm1)

 **however**: if you have 21 habitat variables and want
to consider all subsets of main effects only, you will have
something like 2^21 approx. 2 million models to consider.
Probably a bad idea.  I would strongly recommend that you
consider some sort of dimension reduction (e.g. take the
first few PCAs of the habitat predictors) first.