[ADMB Users] generalizing REML

[H. Skaug]

Yes, but one could say that this practice needs an argument. I think
(I have seen a paper once.) The argument I like myself is that
integration with respect to nuisance parameters (fixed effects
in the context of estimating variances) is a good thing:

Title: Integrated likelihood methods for eliminating nuisance parameters
Author(s): Berger JO, Liseo B, Wolpert RL
Source: STATISTICAL SCIENCE   Volume: 14   Issue: 1   Pages: 1-22
Published: FEB 1999

The question then becomes whether you thrust the Laplace approximation,
but in a GLMM context you already to that for the random effects, so why
not for the fixed effects.

Hans

On Tue, Jun 14, 2011 at 9:22 PM, dave fournier <otter at otter-rsch.com> wrote:
>
> See
>
> https://stat.ethz.ch/pipermail/r-sig-mixed-models/2011q2/006218.html
>
>
> Isn't this just what you get if declare some of the the fixed effects to be
> random effects
> in an ADNB RE model? Somone please enlighten me.
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users
>