[R] MCMCglmm model set-up and interpretation

Theofania-Sotiria Patsiou, Φαίη Πάτσιου t@@@p@t@|ou @end|ng |rom gm@||@com
Fri May 31 03:44:24 CEST 2019

Dear list,

I am new to MCMCglmm and I am trying to test in my data whether there is a
significant plot effect (and of which plot) on each treatment per group of
species while accounting for phylogenetic relatedness.

My data is structured as follows:

Species: 100 species

Species type: 3 levels

Treatments: 3 levels

Plots: 14 levels

Response variable Y: continuous

What I have done is to fit the following model (after testing for various
priors I came up with the following expanded one)

prior.exp <- list(G = list(G1 = list(V = 1,  nu = 1e+06, alpha.mu=0, alpha.
V=1000)), R = list(V =1, nu = 1e+06))

fit.mod = MCMCglmm(log(Y) ~ -1 + Treatment + Plot + Group + Group*
Treatment * Plot,

            random = ~Species ,  ginverse = list(Species = ainv01), data =

                family = "gaussian", nitt = 5e+06, burnin =6000, thin = 150,

                prior = prior.exp ,verbose=F)

1) Is this syntax correct to extract the effect of the plot on the
treatment per group of species or should I use Group:Treatment:Plot as they
are more like nested effects? Is it correct removing the intercept here?

2) The model summary comes up with CI per treatment for all treatment types
but for the Group and Treatment, one of the levels is kept for comparison
and is missing. This is confusing for more than 2 factors with more than 3
levels interaction, as I cannot figure out which factor's level is kept as
the reference for the given CI.

3) For the overall interaction of the 3 categorical variables, the summary
comes up with specific factor levels and not the overall effect of each
variable or their interactions.

Many thanks in advance!

	[[alternative HTML version deleted]]

More information about the R-help mailing list