[ADMB Users] Getting the likelihood from a Random Effects model
hskaug at gmail.com
Tue Oct 27 21:23:00 PDT 2009
The .par file reports the (negative-log) integrated likelihood which should be
used with a likelihood ratio test or the AIC criterion. I use the former
for nested models and the latter for un-nested models.
Model validation: standard diagnostic tools apply. In addition, the
estimated random effects
(in .par). If they are assumed Gaussian a histogram
or Kolmogorv-Smirnov test can be used. There are some problems with this.
No concensus in the literature, as far as I can see.
I agree that this should be in the user manual.
On Wed, Oct 28, 2009 at 12:43 AM, Mark Payne <mpa at aqua.dtu.dk> wrote:
> Dear ADMBers,
> It's possible I'm asking a stupid question. But here goes anyway.
> As I've mentioned earlier this week, I am doing some random effects
> modelling. There are a few different ways that one can set the model in
> question up and I would like to use a likelihood ratio test to compare,
> for example, two slightly different model configurations. However, as I
> understand things, to do this properly I would need to get the actual
> likelihood value value out of ADMB, which is calculated internally by
> integrating out the random effects using the laplace approximation. Is
> it possible to do this? I couldn't find anything in the ADMB-RE
> manual... And perhaps more importantly, is this a sensible thing to be
> doing... ;-)
> On a similar vain, if anyone has any good suggestions about how to
> validate such a model and show that it does a good job of representing
> the data (other than just eye-balling it), I'd love to hear your ideas!
> Users mailing list
> Users at admb-project.org
More information about the Users