[R-sig-ME] MCMCglmm : Difference in additive genetic variance estimated in univariate vs bivariate models

Jarrod Hadfield j.hadfield at ed.ac.uk
Thu Jul 19 12:16:38 CEST 2012


Bivaraite analyses can increase the precision of Va estimates for  
traits with low heritability if the other trait has high heritability  
and a genetic correlation exists.

However, I would be very surprised  if this was the cause. More likely  
(if you only have so few individuals) your posterior is just  
reflecting your prior. If you have kept V and nu the same in the  
univariate and bivariate analyses then the marginal prior on the  
variances in the bivariate analysis (if V is diagonal) has nu-1 and  
V*nu/(nu-1) - if nu=1 then V*nu/(nu-1) is ....



Quoting Stephane Chantepie <chantepie at mnhn.fr> on Thu, 19 Jul 2012  
11:56:52 +0200:

> Dear all,
> I have been running animal models to estimate whether Va of a trait was
> changing with age, and the correlation between this trait expressed in
> different age classes. The problem is I have a difference in additive genetic
> variance (Va) estimated in univariate vs bivariate models.
> To be clearer: I have run an animal model on “Spz_9” (spz trait for  
> age 9) and
> obtained a Va = 0.15 (lower = 0.04; upper = 0.88). The intervals are pretty
> large because I do not have a lot animals in the pedigree (171  
> animals) but it
> seems that the model succeeds in estimating Va. When I run a bivariate model
> c(Spz_9,Spz_5), the Va estimations of spz_9 dramatically increases with Va =
> 1.02 (lower = 0.35; upper = 1.71). The Va posterior distributions of  
> the trait
> is well shaped, so I do not know whether there is a problem or not. Va
> increases when I run a bivariate model between spz_9 and some other traits
> (spz_2 or spz_3) and the Va values for spz_9 are also closed to 1.
> I have used Gaussian distributions and informative prior (V =(Phenotypic
> variance/2) , nu =1 ), no fixed effect and (birth age +animal)  
> random effects.
> I am wondering what these differences mean? Could the problem come  
> from a lack
> of information? Can the covariance between spz_9 and spz_5 be well estimated?
> Many thanks in advance for your help,
> All the best
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

More information about the R-sig-mixed-models mailing list