[R-sig-ME] Wald tests GLMM with glmer

[Malcolm Fairbrother]

Dear Thomas,

There are others on this list who are more knowledgeable than me, but I can suggest a couple things:


> I couldn't use mcmcsamp function neither mcsamp function. What are the other possibilities ?

With the new lme4, try:
? bootMer


> Finally, I tried to use MCMCglmm package, despite my poor familiarity 
> with these techniques (maybe a mistake). When specifying 
> family="categorical" (as my response variable is binary) the plot of the 
> traces were really bad , while there were really better with 
> family="gaussian". Does it make any sense ? Not really to me ...
> Moreover, I also have a problem with one of the 2 random effects (traces 
> stucked at zero). It has only 5 levels and I think I read that there 
> might some problems with random effects with few levels.

Five levels is not enough. In most instances, I think, you need at least 20, and ideally more than that. Though it depends on the context, and to some extent who you ask.


> How to deal with this kind of problem ?

Could you turn this random classification into a series of fixed effects, using dummy variables?

Hope that's useful.
- Malcolm