[R-sig-ME] pedigreemm with one observation per individual

[David Duffy]

On Wed, 10 Feb 2010, Aaron Rendahl wrote:

> I'm trying to fit a mixed model using pedigreemm, but unlike any of
> the examples, I only have one observation per individual.
> I'm getting the following error:
>
>   Error in function (fr, FL, start, REML, verbose)  :
>     Number of levels of a grouping factor for the random effects must
> be less than the number of observations
>
> However, with correlated random effects with a known correlation
> pattern, as with pedigree data, it seems this is a reasonable thing to
> do.  Is there a way to tell lme4 to allow this in this case?  Or is
> this still not a reasonable thing to do?
>

It is reasonable, especially when one wants to estimate breeding values 
(so could be many more REs than observations).

Continuing the example (FWIW)...

library(kinship)
p1 <- data.frame(id=pedCowsR at label, fa=pedCowsR at sire, mo=pedCowsR at dam)
kmat <- kinship(p1$id, p1$fa, p1$mo)
lmekin(sdMilk ~ 1, data=m1, random = ~1|id, varlist=list(kmat))

Linear mixed-effects kinship model fit by maximum likelihood
   Data: m1
   Log-likelihood = -57.32775
   n= 47

Fixed effects: sdMilk ~ 1
             Estimate Std. Error t value     Pr(>|t|)
(Intercept) 4.311032  0.1208144 35.6831 3.558233e-35

Random effects: ~1 | id
  Variance list: list(kmat)
                         id     resid
Standard Dev: 0.0025911150 0.8193825
% Variance:   0.0000099999 0.9999900

(Not a great example, as in herd 89, the half-sib correlation for this
variable, the sole source of information, is -0.04)

-- 
| 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