[R-sig-ME] informative priors in MCMCglmm
j.hadfield at ed.ac.uk
Sat Feb 14 09:03:58 CET 2015
Sorry for the delay. You can't place a prior directly on the
heritability in MCMCglmm (you would have to use JAGS/WinBUGS to do
that). You can place priors on the variance components though. You
have the choice of inverse Wishart priors (aka inverse gamma or
inverse scaled chi-square in the univariate case) or parameter
expanded priors (setting nu=1 and alpha.mu=0 you have a scaled F(1,1)
prior). The inverse Wishart prior is the easiest to understand:
placing a prior of V=2, nu=1 on the additive genetic variance is
equivalent to having a prior observation of a single breeding value of
value sqrt(V). A prior of this strength sounds innocuous but it can
have a surprisingly strong effect.
Quoting Diana Caro <dianacaro0918 at hotmail.com> on Mon, 26 Jan 2015
> I want to estimate variance components for four weight traits in a
> multivariate model.If I have prior information of variances or
> heritabilities (from literature), how can I incorporate this
> information in a prior?
> [[alternative HTML version deleted]]
> R-sig-mixed-models at r-project.org mailing list
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
More information about the R-sig-mixed-models