[R-sig-ME] pedigreemm and heritability of binary data

[David Duffy]

On Wed, 9 May 2012, Pierre de Villemereuil wrote:

> Ana Vazquez (who is in charge of the development of pedigreemm I shall
> think) responded me that the package was not meant to deal with this
> kind of model.
>
> If I understood correctly, the function is not designed to fit GLMM with
> p=n (i.e. number of effects equal to the number of observations =
> generalized animal model in my case).

If you briefly reconsider, you will realise this is not very likely, given 
the entire purpose of the package.  The problem a while back was that a 
rather coarse check for nonidentified models was added to lmer(): by 
testing for p >= n.  This stopped pedigreemm from working, as well as 
blocking the perfectly valid fitting of overdispersion GLMMs with one p 
per observation.  That is to say, you _should_ be able to get a sensible 
answer.

There are several R packages that can fit an animal model GLMM: 
pedigreemm, AnimalINLA, MCMCglmm (see eg the thread starting from 
https://stat.ethz.ch/pipermail/r-sig-mixed-models/2012q1/014702.html).
There are several non-R packages you can compare numerical results to for 
the probit-normal at least.

Cheers, David Duffy.

-- 
| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v