[R-sig-ME] Prior for a bivariate model

[Malcolm Fairbrother]

Hi Gregory,

I can't answer your question, but it sounds likely that someone on the list
will be able to. However, you'll increase the chances if you can send a
snippet of your R code thus far...

What call to MCMCglmm and what specification of your prior are you using
for a univariate model for each outcome separately, for example? If you can
show people something of what you've been able to do so far, you'll likely
get a response about how to go further.

Best wishes,
Malcolm





> Date: Mon, 20 Feb 2017 18:41:14 +0100
> From: Gr?gory DANIEL <daniel.gregory3 at gmail.com>
> To: r-sig-mixed-models at r-project.org
> Subject: [R-sig-ME] Prior for a bivariate model
> Message-ID: <8960d3bb-9c0c-50e4-38ed-e7763469d76b at gmail.com>
> Content-Type: text/plain; charset=utf-8; format=flowed
>
> Dear list members,
>
> I am working with MCMCglmm package on a bivariate model to estimate the
> genetic covariance between my two traits that are my response variables.
> One is binary and the other is gaussian.  I have about 700 individuals,
> one measure per trait for each individual.
>
> The model is simple : sex as fixed effect, and animal as random effect.
> But I have been struggling for several weeks to find a proper prior for
> which the model converge concerning the "animal" variance of my binary
> variable, and to do not have a big auto-correlation between iterations
> for the same variance.
>
> Concerning the residual variance, I have no problem, only with the prior
> for the additive genetic variance of my binary response variable. I
> think we cannot specify two priors for a same variable...
>
> I hope I was clear enough... Does anyone could help me to find a
> solution please  ?
>
> Thanks a lot,
>
> Gr?gory DANIEL - Ph.D. in evolutionary ecology
> mail : daniel.gregory3 at gmail.com
>

	[[alternative HTML version deleted]]