# [R-sig-ME] cross-sex genetic correlation

Simona Kralj Fiser simonakf at gmail.com
Tue Jan 16 13:22:47 CET 2018

```Hi!

I have some further questions on the cross-sex genetic correlations.

It seems that many studies fail to obtain sample sizes that is big enough
to get the realistic 95% CI. One of the solutions I noticed is comparing
the models (ASReml) where rmf is set to 1 or 0. How could that be done when
using MCMCglmm models? I was thinking of comparing

1. model 1, rmf=1

prior1<-list(G=list(G1=list(V=matrix(p.var*0.5),n=1)),R=list(V=matrix(p.var*0.5),n=1))

model1 <- MCMCglmm(trait ~ sex, random = ~animal, pedigree = pedigree,data
= data, nitt = 100000, thin = 100, burnin = 15000, prior = prior1,verbose =
FALSE)

2. model 2 allowing sexes to have different variance and covariance

prior2 <- list(R=list(V=diag(2), nu=0.02), G=list(G1=list(V=diag(2), nu=2,
alpha.mu=c(0,0),alpha.V=diag(2)*1000)))

model1 <- MCMCglmm(trait~sex, random=~us(sex):animal, rcov=~idh(sex):units,
prior=prior2, pedigree=Ped, data=Data1, nitt=100000, burnin=10000, thin=10)

3. model 3 allowing sexes to have different variance, but covariance = 0
--> rmf = 0

prior2 <- list(R=list(V=diag(2), nu=0.02), G=list(G1=list(V=diag(2), nu=2,
alpha.mu=c(0,0),alpha.V=diag(2)*1000)))

model3 <- MCMCglmm(trait~sex, random=~idh(sex):animal,
rcov=~idh(sex):units, prior=prior2, pedigree=Ped, data=Data1, nitt=100000,
burnin=10000, thin=10)

My reasoning may be stupid, I hope that's not forbidden :-)

Is there any other way? Fixing variances with prior?

Thank you. Best wishes

Simona

On 26 July 2017 at 14:42, Jarrod Hadfield <j.hadfield at ed.ac.uk> wrote:

> Hi,
>
> The second way is a *much* better way of doing it but should give the same
> answer. However, in both cases the residual covariance is not identifiable
> (no individual is both male and female) and so you should use idh rather
> than us.
>
> The "subscript out of bounds" error is to do with your code that
> post-processes the model output not an issue with MCMCglmm. Probably you
> have used the wrong names for the (co)variance components.
>
> Also, you haven't passed the prior to MCMCglmm, nor is the prior a valid
> one for the problem as it specifies scalar variances rather than 2x2
> covariance matrices. You could try
>
> prior2 <- list(R=list(V=diag(2), nu=0.02), G=list(G1=list(V=diag(2), nu=2,
> alpha.mu=c(0,0),alpha.V=diag(2)*1000)))
>
> Cheers,
>
> Jarrod
>
>
>
>
> On 26/07/2017 13:33, Simona Kralj Fiser wrote:
>
>> model <- MCMCglmm(W~sex, random=~us(sex):animal, rcov=~us(sex):units,
>> prior=prior2, pedigree=Ped, data=Data1, nitt=100000, burnin=10000,
>> thin=10)
>>
>
>
> --
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
>

[[alternative HTML version deleted]]

```