Skip to content
Prev 10613 / 20628 Next

lmer: parametric bootstrap versus mcmcsamp

Thanks for the comments Ben. Additional simulations with 2-3 
hierarchical levels give stronger evidence that parametric bootstrap is 
doing a good job. Distributions are often identical to the posterior 
obtained by MCMCglmm. mcmcsamp() is, however, in general very instable 
in my opinion. It sometimes surprises and matches the other results. I 
havent figured out under which conditions this is happening though.

I understand that few levels are problematic, but for some 
applications/questions you may be able to get some sensible answers 
(e.g. bias may be small compared to other uncertainties, etc).

Sometimes the pm bootstrap runs fail (convergence problems). I just read 
that there is going to be a slot in the release of lme4 (I guess it is 
already implemented in the development version) where warnings messages 
are stored. This will make it easier to sort out pm bootstrap runs that 
encountered problems.

Gustaf