[R-sig-ME] number of levels of grouping factor == number ofobservations

[David Duffy]

On Tue, 9 Mar 2010, Douglas Bates wrote:

> For what families in a generalized linear mixed model should a
> grouping factor be allowed to have as many levels as there are
> observations?  Just the binomial and the poisson or everything except
> the gaussian family or ...?

It has to be all families.  As someone pointed out earlier, pedigreemm 
won't fit a simple univariate animal LMM unless this is allowed, and if 
one wishes to also estimate biometrical dominance, epistatic terms, 
assortment etc, then one has identified models with 2,3... random effects 
per individual, providing the data is adequate (that is the appropriate 
types of relationships are observed in the pedigrees).  For example, if 
one has pairs of identical twins, and ordinary siblings pairs in the data, 
there are two intraclass correlations which can be used for the 
decomposition into additive and dominance genetic variances rMZ=a+d, 
rSS=0.5*a+0.25*d (obviously the range of allowable r's is bounded).

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