[R-sig-ME] Strange results using MuMIn's model.avg() and MCMCglmm model

Kenneth KEUK kenneth@keuk @ending from gm@il@com
Wed May 30 09:57:33 CEST 2018

Hello everyone,

I am modelling parasite loads (counts, Poisson distributed) according
to multiple quantitative behavioural variables in primates. Thus, I am
using the MCMCglmm package to fit Bayesian mixed-models, especially
for later implementation of a pedigree.

I need to average models using the model.avg() function of the MuMIn
package, which is described as supporting MCMCglmm
(https://cran.r-project.org/web/packages/MuMIn/MuMIn.pdf). As
specified, I also had to wrap the MCMCglmm() function with
updateable() for the averaging to "work" without error.

The issue though is, while the function provided averaged estimates
for the variables, the standard errors seem suspicious (Z values and
Pr values also).

Either none were provided (NAs), or they were _exactly_ equal to their estimate.

In other words, I came to the exact same issue as this unanswered
question in StackOverflow, which provide a reproducible example:

Is this behaviour normal or am I missing something in how the average
is done? Have others manage to obtain credible intervals after
averaging MCMCglmm models with MuMIn?

Many thanks for your help!
Kenneth Keuk

More information about the R-sig-mixed-models mailing list