Skip to content
Prev 5499 / 20628 Next

lmer formula syntax?

On Tue, 2011-02-01 at 14:14 -0800, Colin Wahl wrote:
Hi Colin,

I have very little to add to the discussion thus far (from which I have
learned a lot by the way!), except to question the use of family =
gaussian for this response? %EPT will be bounded at zero (there will be
samples for locations that are so polluted as to not have any of these
critters living there), and potentially bounded at 100% (theoretically
at least, although I find it hard to see how one would get 100% of this
group of inverts in a kick sample). The errors may depart from the
assumptions of the LMM.

If you know the total counts then perhaps a binomial GLMM would be
appropriate. If not, beta regression via the betareg package might be an
alternative - the sandwich package can be used to  fix up the covariance
matrix to account for the clustering in the data.

Just my tuppence,

G