[ADMB Users] latest glmmadmb and p values

Ben Bolker bbolker at gmail.com
Tue Oct 26 09:49:09 PDT 2010

On 10-10-25 07:40 AM, dave fournier wrote:
> Thanks Ben. I agree that one probably should not
> use p values. I would use LR tests. However it seems that
> people need to use p values. the problme with p values
> will be there no matter what package one uses.  so if people
> want p vlaues we might as well supply them.

  Well, this is where philosophy comes in.  I fall somewhere between you
("if people want it let's give it to them even if it might be
dangerous") and the cranky R people ("if we don't think it's a good idea
we're not going to implement it and we'll even yell at you for
suggesting it") ...

> Actually I am more concerned with what appear to be commonly
> held beliefs that a simple parameterization of
> overidspersion in the NB takes care of all the
> overdispersion in count models. In my experience using
> LR tests one can generally show that overdispersion
> needs to be modelled more elaborately.

  Fair enough.  Perhaps glmmADMB could be extended to make it easier to
test this; not looking at it at the moment, but I suspect that it only
allows a single parameterization?  I wonder if one could build
exploratory graphical tools that showed residuals/nonparametric
estimates of the mean/variance relationship and superimposed
quasi-Poisson (var=phi*mu), NB (var=mu*(1+mu/k)), etc. ... relationships?

   Ben Bolker

More information about the Users mailing list