Hi Pierre -
My full-sib families of larvae are thousands strong. I typically put 50 or more individuals into a single measurement, so I would consider them reasonably close to family-wise averages. Also, I can cross as many sires with as many dams as I want (actually, as many as I can handle, bucket-wise). In my current experiment, two dams were crossed with the same ten sires each, so I am looking at 20 full-sib families. This should give me reasonably good estimate of the size of sire effects, don't you think?
M
On Dec 29, 2011, at 4:31 AM, Pierre B. de Villemereuil wrote:
Hi !
I'm wondering if you noticed my e-mail on the R list, so I'm transferring it to you, just in case !
I'm still puzzled by your design : could you explain it further to me ? You say you have a full-sibs design and estimating sire effect : is that to say you have the offspring of one male (sire) and one female (dam) for each sire ? Or only one dam, and several sire ?
And also, if I understand correctly, you don't have individual measurement for the offspring, but only the average sibling phenotype, is that right ? And you don't have parents phenotype ?
Cheers,
Pierre.
-------- Message original --------
Sujet: Re: [R-sig-ME] animal model: calculating heritability and evolvability from sire effects
Date : Wed, 21 Dec 2011 21:56:21 +0100
De : Pierre B. de Villemereuil<bonamy at horus.ens.fr>
R?pondre ? : bonamy at horus.ens.fr
Pour : r-sig-mixed-models at r-project.org
Hi !
Concerning the 4 coefficient : you say the additive variance is
estimated within the full-sibs of the same sire. Is that to say
offspring of the same sire descent as well from the same dam ? In that
case (same father and same mother), then the relationship coefficient is
of 1/2 (2 * coefficient of coancestry of 1/4). So, I think (but I can be
wrong) the coefficient should be 2.
If each individual descent from a different mother, then kinship is 1/4.
So the coefficient of 4 is correct.
When estimating the heritability of a binomial trait, you have to keep
the residual variance in the total variance. Just add the 'link
variance' (say Vlink) such as :
h? = Va / (Va + Vr + Vlink) (Vlink is pi?/3 for logit link and 1 for
probit link)
Cheers,
Pierre.
Le 21/12/2011 12:18, Szymek Drobniak a ?crit :
Hi,
both your code and the way you calculate VA using sire variance seems
fine. In lmer residual variance is fixed as it assumes fixed
relationship between variance and mean in binomial data so I'm not
sure if simply putting this variance in your formula solves the
problem. In MCMCglmm residual variance quantifies overdispersion so as
far as I know here it's just a matter of substituting gaussian to
multinomial2.
Cheers,
sz.
--
Szymon Drobniak || Population Ecology Group
Institute of Environmental Sciences, Jagiellonian University
ul. Gronostajowa 7, 30-387 Krak?w, POLAND
tel.: +48 12 664 51 79 fax: +48 12 664 69 12
www.eko.uj.edu.pl/drobniak