R help - Feb 2012 - General question about GLMM and heterogeneity of variance

My data have heterogeneity of variance (in a categorical variable), do I need
to specify a variance structure accounting for this in my model or do GLMMs
by their nature account for such heterogeneity (as a result of using
deviances rather than variances)? And if I do need to do this, how do I do
it (e.g. using something like the VarIdent function in nlme) and in what
package?

This is my first post so apologies for any breach of etiquette!


--
View this message in context:
http://r.789695.n4.nabble.com/General-question-about-GLMM-and-heterogeneity-of-variance-tp4424429p4424429.html
Sent from the R help mailing list archive at Nabble.com.

GibsonR <rachel.gibson <at> bristol.ac.uk> writes:
> 
> My data have heterogeneity of variance (in a categorical variable), do I
need
> to specify a variance structure accounting for this in my model or do GLMMs
> by their nature account for such heterogeneity (as a result of using
> deviances rather than variances)? And if I do need to do this, how do I do
> it (e.g. using something like the VarIdent function in nlme) and in what
> package?
  We need a little more information.

  Also, it might be better to send follow-ups to r-sig-mixed-models <at>
r-project.org .

  Is your a categorical variable a predictor ("independent") or
response ("dependent") variable?  If it's a predictor, then the
details of its distribution are not important for the validity
of the analysis.  It it's a response, then you need to be doing
a multinomial or ordinal response model.  GLMs and GLMMs do account
for some forms of heterogeneity in variance.

  You probably need to tell us more about what you tried to do
and how you concluded that heteroscedasticity was a problem.

Sorry, I was not particularly clear.

 I ran my data through a GLM (the response variable is a proportion, and I
ignored the random effects for the purposes of data exploration), and
plotted the residuals against each of my predictor variables (some of which
are continuous, some categorical). The heterogeneity showed up in the
residuals of the response variable plotted against a categorical predictor
variable (Insect functional group).

Do I need to use something other than the GLMM in this case?

Thank you very much for your help.



--
View this message in context:
http://r.789695.n4.nabble.com/General-question-about-GLMM-and-heterogeneity-of-variance-tp4424429p4430970.html
Sent from the R help mailing list archive at Nabble.com.

I will also post the above to the sig mixed models forum now. 

Thanks.


--
View this message in context:
http://r.789695.n4.nabble.com/General-question-about-GLMM-and-heterogeneity-of-variance-tp4424429p4431107.html
Sent from the R help mailing list archive at Nabble.com.