[R-sig-ME] Predicting values from MCMCglmm model with statistical weight in mev argument

Kamal Atmeh k@m@|@@tmeh @end|ng |rom hotm@||@com
Sat Feb 15 23:57:08 CET 2020

Dear list,

I'm quite new to MCMCglmm and after reading much of Jarrod Hadfield's 
documentation I was able to understand the basic construction and 
interpretation of models. However, I am having a bit of trouble when 
predicting values from my gaussian MCMC model. I am running bayesian 
phylogenetic mixed models to determine factors responsible for the 
variation of my movement parameter. My response variable is continuous 
and each point has a standard error associated to it and which I added 
as a statistical weight in the model using the mev argument in the 
MCMCglmm function. I have 5 random effects: phylogeny (to which I 
attributed the species covariance matrix using the ginverse argument), 
species, population, year and individual. My model is thus as follows:

 >>> model <- MCMCglmm(lD ~ 
                                      , random = 
                                      , family = "gaussian"
                                      , mev = SE^2 # error variance 
associated to each data point
                                      , ginverse = list(sp_phylo = 
inv.phylo$Ainv) # include a custom matrix for argument phylo
                                      , prior = prior1
                                      , data = Data
                                      , nitt = 22e+04
                                      , burnin = 20000
                                      , thin = 100
                                      , pr=TRUE)

When I try to predict values from the model above, I obtain the 
following error:

 >>> pred_DGLOB <- predict.MCMCglmm(model_D_GLOB_ch1
                                    , type = "response"
                                    , interval = "confidence"
                                    , newdata = newdt)

 >>> Error in rep(as.numeric(rcomponents %in% mcomponents), 
object$Random$nrt) :
       invalid 'times' argument

It appears that this error is originating from the following specific 
code in the predict.MCMCglmm function:

 >>> marginalise<-rep(as.numeric(rcomponents%in%mcomponents), 

It appears that object$Random$nrt has one additional element compared to 

I already ran predict.MCMCglmm with models without any statistical 
weight and the predict.MCMCglmm function worked fine. Is this a problem 
caused by the fact that I have the standard error in the mev argument? 
If yes, is there a way to go around it? I may be skipping some steps 
before running the predict function but I did not find any documentation 
that addresses this issue.

I would greatly appreciate your help and happy to provide further 
information if needed!

Thank you in advance!


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