Message-ID: <5339E13F.9010704@lists.r-forge.r-project.org>
Date: 2014-03-31T21:42:23Z
From: Ben Bolker
Subject: [Lme4-authors] Underdispersion in GLMM
In-Reply-To: <CF49F8D43312F648A7CE460833970E414C7B609C@mbx05.adf.bham.ac.uk>
[I'm taking the liberty of forwarding this to r-sig-mixed-models,
where I think it is more appropriate.]
A few thoughts about underdispersion:
* underdispersed _count_ data (e.g. litter sizes for large mammals) are
often better captured by an ordinal model (e.g. ordinal::clmm)
* you _might_ be able to fit underdispersed binomial data by allowing
negative correlation structure within groups: this could be tricky,
though, depending on the size and complexity of your data set. (The
standard approach of nested random effects is essentially a compound
symmetry structure, but only allows positive/non-negative correlations.
The nlme package enabled fitting models with a more general CS
structure, but that's not implemented in lme4, and slightly harder to do
in GLMMs in general).
* the standard 'quasi-likelihood' approach, i.e. taking the estimated
level of underdispersion and shrinking all the confidence intervals
accordingly, might be a reasonable first/hackish approach. The thing to
be careful about there is that I have yet to read any good treatment (or
figure out for myself) how quasi-likelihood estimates of 'residual'
variance interact with the estimates of the random effects variances ...
good luck,
Ben Bolker
On 14-03-31 03:51 PM, Eva Reindl wrote:
> Dear Mr Bates, Mr Maechler, Mr Bolker, Mr Walker, Mr Christensen, and
> Mr Singmann,
>
> My name is Eva Reindl and I am a Phd student at the University of
> Birmingham working with your lme4 package to analyse my data. I am
> running GLMMs with a binomial error structure and found that my full
> model is heavily underdispersed. I am writing you because I would
> like to ask whether your package offers any strategy to deal with
> underdispersion. Any help or reference to resources would be greatly
> appreciated.
>
> I would like to thank you very much in advance for your efforts.
>
> Yours sincerely,
>
> Eva Reindl
>
> Eva Reindl MSc Doctoral Researcher at School of Psychology University
> of Birmingham Edgbaston Birmingham B15 2TT United Kingdom
>
> phone: 0121 414 7209 E-mail: EMR328 at bham.ac.uk
> _______________________________________________ Lme4-authors mailing
> list Lme4-authors at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/lme4-authors
>