Skip to content

r-sig mixed models mailing list response

1 message · Ben Bolker

#
On 13-01-20 10:22 PM, Belinda Burns wrote:
[cc'ing to r-sig-mixed-models]
OK.
It might be a good idea to look at the books by Zuur and co-authors,
which are fairly ecologist-friendly. While I don't agree with everything
in them, they generally exhibit common sense and give good general
descriptions.
Transformation of this sort doesn't make sense.  NB models
must be run on *count* data.  It sounds like your data act more
like a Beta distribution (used to model proportions), since
"number of seconds" isn't really that likely to act like a count.
Note that the random effects are extremely small, suggesting that
the amount of noise at the observation level (within the individual
counts) is big enough to swamp any observable effects of Group or
Individual within group.

  I would consider trying this with family="beta" and zero-inflation,
although admittedly that's a combination I have never tested.

Make sure to plot your data and make sure that the estimates
make sense!