[R-sig-ME] MCMCglmm Poisson with an offset term and splines

[Jarrod Hadfield]

Hi,

The model looks OK as far as can be assessed without knowing the data. 
For the offset term you need to hold the associated coefficient at 1 by 
placing a strong prior on it. If  you want everything else to have the 
default prior then use:

k<-11 # number of fixed effects

prior<-list(B=list(V=diag(k)*1e8, mu=rep(0,k)),
             R=list(V=1, nu=0),
             G=list(G1=list(V=1, nu=0),
                    G2=list(V=1, nu=0),
                    G3=list(V=1, nu=0)))

prior$mu[k]<-1 # assuming the offset term is last
prior$B[k,k]<-1e-8

The interpretation of the offset is simply the coefficient is assumed to 
be one and that the rate at which events occur is constant.

Cheers,

Jarrod




On 22/09/2017 01:52, dani wrote:
> Hello everyone,
>
> I have a Poisson model with an offset term that involves repeated observations nested into two cross-classified groups.
>
> The model includes
> - four categorical variables
> - 6 continuous variables (for one of them I would like to include a smoother)
> - the offset=log(duration)
>
> I first used the spl2 function to create the fixed ((x6numspline1) and random terms (x6numspline2) for the smoother. I added the random smoother term to the other two random intercepts (for student ID and classroom) that I have (which are cross-classified).
>
> My question is: Do you find my model sound? Before I study the priors, I just wanted to run a default model - is my inclusion of an offset ok? Also, given that the observations are repeated and nested into both Student ID and classroom, I am not sure how to specify the variance structure in MCMCglmm (beginner here:))
>
> mc_spl0 <- MCMCglmm(number_events ~ x1cat+x2cat+x8cat+x9cat+x3num+x4num+x5num+x6numspline1+x7num+x8num+log(duration),
>                      random =~ ID+class+idv(x6numspline2),
>                      data   = newdat,
>                      family = "poisson",
>                      thin   = 100,
>                      burnin = 10000,
>                      nitt   = 150000,
>                      saveX=TRUE, saveZ=TRUE, saveXL=TRUE, pr=TRUE, pl=TRUE)
>
> In addition, I am not sure what to make of the results for the offset term (included as a covariate in the model) in the output - how should I discuss them?
>
> Thank you all so much!
> Best regards,
> N-M.
> <http://aka.ms/weboutlook>
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.