[R-sig-ME] F test in lmer quasipoisson

[Ben Bolker]

  Please don't cross-post to r-help and r-sig-mixed-models : it's
considered bad form (most of the experts on this subject are reading
r-sig-mixed-models anyway ...)

  Because this capability is not built in.
  If you can send or post your data, I can try to show an example.
More to the point, it's not clear that this will completely solve
your problem -- as the running discussion on this list indicates,
extrapolating from GLMs or LMMs to procedures for GLMMs is a little
dangerous (although it's often the best we can do).  In this case,
the hard part is probably picking the degrees of freedom for
the denominator of the F test ...

  cheers
   Ben Bolker


Julien Vezilier wrote:
> Hello !!
> 
> II'm trying to test for my fixed effects using an lmer with quasipoisson
> errors.
> 
> Since my lmer model is corrected for overdispersion using this kind of
> errors, I should use during model simplification in my Anovas *F test *and
> not *Chi square test* to compare two models. So I write:
> 
>> anova(model,model2,test="F")
> 
> but R keeps performing a Chi square instead of the F test I want...
> 
> Any ideas why ?
> 
> Thank you very much !!!
> Julien.
> 
> 
> 
> 
> 
> 
> 
> 


-- 
Ben Bolker
Associate professor, Biology Dep't, Univ. of Florida
bolker at ufl.edu / www.zoology.ufl.edu/bolker
GPG key: www.zoology.ufl.edu/bolker/benbolker-publickey.asc