Skip to content
Prev 17057 / 20628 Next
Original Raw

MCMCglmm and multimembership analysis

Roslyn Dakin 
Dear List,

I have two questions about MCMCglmm and multimembership models (aka
multiple membership models).

I am using these models to study social influence. In my scenario, the
focal individuals express a phenotype (the dependent variable) depending on
their social environment. The social environment is made up of other
individuals. For example:
MCMCglmm(phenotype ~ 1, random = ~focalID + mm(partnerID1 + partnerID2 +
partnerID2), data=data)
...but the partners are also weighted (i.e., some interact more closely
with the focal than others).

Q1: Can MCMCglmm specify weights for the multimembership structure?

I have been able to do this in brms, but I haven?t seen an example for how
the weights could be done in MCMCglmm.

The second part of my question stems from the fact that a given individual
can act both as a focal and a partner, and I am specifically interested in
the covariance between these two sources of variation. (i.e., I would like
to be able to evaluate whether IDs that score highly on the phenotype also
elicit higher levels of the phenotype in others, using the same model.) In
a simpler, non-multimembership scenario, I would model this covariance as:
MCMCglmm(phenotype ~ 1, random = ~str(focalID + partnerID), rcov = ~units,
data=data)

Q2: Can MCMCglmm accommodate the covariance between a single
multimembership structure and focalID, in the same model?
?e.g., something like this:
MCMCglmm(phenotype ~ 1, random = ~str(focalID + mm(partnerID1 + partnerID2
+ partnerID3)), rock = ~units, data=data)
I have verified that brms doesn't currently have the functionality to do
this.

Any guidance is greatly appreciated.
Thank you!