[R-sig-ME] glmer, p-values, and mcmcsamp

Sandra Hamel sandra.hamel at bio.ulaval.ca
Thu Feb 10 11:17:51 CET 2011


I'm aware of the problems related to p-values in linear mixed models. 
I'm working on a glmer model and was a bit surprise to get p-values 
because I was thinking there were also problems for glmer. From recent 
posts in December on the the validity of p-values in glmer, the 
conclusion was that p-values might actually suffer even worst problems 
for glmer than for lmer. One first (maybe naive) question is why are 
p-values provided in glmer if they suffer from similar issues than 
p-values in lmer (which are not provided for that reason)?

In linear models, I use mcmcsamp to get CI or p-values, but I understood 
that this function is not yet implemented for glmer models. I was 
wondering then what is the best option to get either CI or p-values that 
would be "valid" for glmer models? I have seen an old post (2008) from 
Ben Bolker where he mentioned a "glmersim.R" function he had written 
(and a following post from Douglas Bates where he mentioned that this 
function could be included in lme4). Would "glmersim.R" allow me to get 
something similar to mcmcsamp? Are there still plans to include the 
possibility to allow glmer to run in mcmcsamp?

Thanks for your help,

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