[R-sig-ME] Mixed model and negative binomial distribution

[Seth W. Bigelow]

Dave, that's a very generous offer to modify admb, but I don't know whether
or how estimating separate overdispersion parameters before and after
treatment would affect the test of fixed effects. I will wait a bit to see
if any statisticians chime in on this before sending you the *.dat & *.pin
files.
--Seth

-----Original Message-----
From: dave fournier [mailto:davef at otter-rsch.com] 
Sent: Monday, October 22, 2012 8:37 AM
To: Seth W. Bigelow
Subject: Re: [R-sig-ME] Mixed model and negative binomial distribution

On 12-10-21 04:58 PM, Seth W. Bigelow wrote:

It is not difficult to parameterize the model with an overdispersion
parameter which depends on a number of covariates, but I odn't think
it has been done (you could ask Ben).  glmmadmb is written
using ad model builder which I created. I can easily modify the
program to do that if you want.  Creating an r interface is
the difficult part so I prefer to work just with admb.  If you want what
I need to test this out are the files  *.dat and *.pin which you can 
save using
the savedir option of glmmadmb.

       Dave

> Ah, ok, a light bulb just went on. It must be that the overdispersion
> parameter is equivalent to the 'k' parameter of the negative binomial, and
> that rather than estimating a different k for each cell mean, a single k
for
> the whole experiment is estimated in glmmadmb.
>
> Would it be possible, or logical, to estimate different overdispersion
> parameters before and after treatment?
>
> -Seth
>
> -----Original Message-----
> From: dave fournier [mailto:davef at otter-rsch.com]
> Sent: Sunday, October 21, 2012 7:10 PM
> To: seth at swbigelow.net
> Subject: Re: [R-sig-ME] Mixed model and negative binomial distribution
>
> Hi,
>
> Do you mean that you want to use different parameters for the
overdispersion
> before and after treatment?
>
>        Dave
>