[R-sig-ME] Inflated heritability estimate for animal model with binary response using MCMCglmm

Jarrod Hadfield j.hadfield at ed.ac.uk
Sun Aug 30 11:57:14 CEST 2015


Is this the h2 on the latent or data scale? If the latent scale do  
families/relatives have correlated sexes/ages? If they do, this may  
explain a drop in the latent scale h2.

Also, if you have phenotypes for parents it is very important that you  
give them dummy fam variables not NA. They did belong to a family,  
even if you did not record it.

I would also use family="threshold" rather than family="ordinal". They  
are equivalent but you do not need to add 1 to the denominator  
variance of the h2 with family="threshold". If you set V_units=1,  
family="threshold" is the standard probit model.

Also, V=1 and nu=1 is a pretty strong prior, and so you may want to  
change this.



<kanixwang at uchicago.edu> on Wed, 26 Aug 2015 15:11:17 -0500:

> Dear list,
> I'm using MCMglmm to estimate heritability for binary traits with the model
> below. I have seen a inflated heritbility estimate when I use sex and age
> as fixed effects vs only using intercept. Both estimates passed the
> convergence test and half-width test.
> Can anyone help me identify the likely causes of this inflation? Is the
> model mis-specified? Also, I'm wondering if using age as a random effect
> make any sense.
> priorA <- list(R = list(V = 1, fix = 1), G = list(G1 =list(V = 1, nu =
> 1000, alpha.mu = 0, alpha.V = 1), G2 =list(V = 1, n = 1)))
> modelbin <- MCMCglmm(pheno ~ sex + age, random = ~animal +fam , family =
> "ordinal", pedigree = pedigree, prior = priorA, data = databin, nitt =
> nitt, burnin = burnin, thin = 500, slice=TRUE, pl=TRUE)
> Sex and age are factor variables. The fam variable is for the common
> environment effects in the descendants. All parents have NA for fam.
> Any help would be greatly appreciated.
> Best,
> Kanix Wang
> 	[[alternative HTML version deleted]]
> _______________________________________________
> 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