[R-sig-ME] Factor level contrasts and associated fixed effect priors in MCMCglmm
Mohammed Armani
moarmani at outlook.com
Sat Apr 7 08:02:44 CEST 2018
Deal all,
I am running a phylogenetic binary model in the package MCMCglmm (The response variable being Resp (0, 1)). One of my predictors is a categorical variable (with 3 levels: T1, T2 and T3) and I would like to set-up the model to explicitly test for differences between the levels of this factor (i.e. T1 vs T2; T1 vs T3 and T2 vs T3). I am following the GLMM worked example of https://ms.mcmaster.ca/~bolker/R/misc/foxchapter/bolker_chap.html and proceeded as follows:
contrs<-matrix(c(0,1,0,
-1,0,1,
0,-1,1),
byrow=TRUE, ncol=3,
dimnames=list(c("T1 vs T2",
" T1 vs T3 ",
" T2 vs T3"),
levels(data$T)))
sol_ctr<-solve(contrs)
newmat1<-model.matrix(~T,data=data,
contrasts=list(T=sol_ctr1))
datar <- with(data,data.frame(Resp,T,phylo,
newmat1[,-1]))
mod_1<-MCMCglmm(Resp~ T1 vs T2+ T1 vs T3 + T2 vs T3-1,random=~phylo,family="categorical",
ginverse=list(phylo=inv.phylo1$Ainv),prior=prior,trunc=TRUE,
data=datar,nitt=3e6,burnin=3000,thin=1500)
Question:
1. First, I would be grateful to receive some feedback if I am doing the right thing with setting the contrasts!
2. Secondly, one of the factor levels (T2) has only success in the response variable (all=1). I understand from the MCMCglmm course note that in such instances a fixed effect prior such as B=list (mu=c (0, 3), V=diag (3)*(1+pi^2/3) can be applied. However, in the instance where an explicit contrasts is specified as I have done, I am unsure of how to specify the contrast for the fixed effect terms
Thanks.
Armani M
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list