[R-sig-ME] comparing heritabilities of traits

Jarrod Hadfield j.hadfield at ed.ac.uk
Mon Feb 14 11:40:20 CET 2011

Hi Ardo,

I think I have addressed some of these questions in reply to some of  
your previous posts to the list. In short, testing the 'significance'  
of a heritability estimate is difficult to evaluate from the posterior  
because it has to be positive. I've seen people use DIC to evaluate  
whether a genetic term should be included, but I think DIC is  
'focused' at the wrong level for this sort of question and people  
should be very wary. My personal view is that a significance test for  
h2 is meaningless (everything that varies between individuals in an  
outbred population will have some genetic variance, even if it is  
small) and that people should focus on the width of the credible/ 
confidence intervals rather than the p-values.

With regards to your second question about how to choose the trait  
with greatest h2, you could try this:

h2<-function(x){x["animal"]/sum(x)}  # calculates h2

h2.post<-as.data.frame(lapply(list(m1=m1$VCV, m2=m2$VCV, m3=m3$VCV),  
function(x){apply(x, 1,h2)}))

# create a table of posterior h2 for each response (m1, m2, m3 .....)

prop.table(table(apply(h2.post, 1, which.max)))

# posterior probability for each response that it has the highest h2



On 11 Feb 2011, at 16:02, Ndjido Ardo BAR wrote:

> HI list!
> I'm using MCMCglmm to estimate heritability of candidate traits so  
> as to
> select the most heritable ones for further genome wide association  
> study. My
> problem is to establish the significance of those heritabilities being
> different to zero and to compare them. I do have HPD intervals but i  
> dont
> know how to decide!
> cheers,
> Ardo
> 	[[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