Skip to content
Prev 13429 / 20628 Next

MCMCglmm priors and random effects for phylogenetic mixed model

Hi Jarrod,

this response was incredibly helpful. I decided ultimately that I didn't
need the random slopes in the model, which simplified things.

I have a follow-up question. I read the Phillimore et al. PNAS paper you
linked to with great interest. It seems in that paper you fitted a
multivariate (i.e., multiple response) mixed model as an alternative way
(from the van der Pol & Wright method) of decomposing between- and
within-species slopes (your Equation 4). The advantage, if I understand
correctly, is that the species-mean effect and individual-level deviations
from it are estimated, rather than calculated from the sample data. I'm
having a hard time figuring out how to alter the model I specified to fit
this framework (or if this is even possible).

My question is, how would I reconfigure the formula in this model to fit a
multivariate model that provides an estimate of the between species effect
(that's all i'm interested in)?

fit <- MCMCglmm(
    fixed = I1.M1 ~ I2.M1.species.mean + I2.M1.within.species,
    rcov = ~ units,
    random = ~ phylo + species,
    data = incisor.dat,
    family = "gaussian",
    ginverse = list(phylo = inv_phylo_mat$Ainv),
    prior = priors1,
    nitt = 1.1e+6, thin = 10, burnin = 1e+5
    )

Alberto


On Fri, Dec 26, 2014 at 1:50 AM, Jarrod Hadfield <j.hadfield at ed.ac.uk>
wrote:

  
  
Message-ID: <CAO+b4j8G1BBtv-19ZJ=09p6R+kL4ca3NJPm8LPnFM1HYxYi4nA@mail.gmail.com>
In-Reply-To: <20141226065015.124611gwbjfox0kg@www.staffmail.ed.ac.uk>