[R] Lmer, heteroscedasticity and permutation

[Alan Juilland]

Hi everybody,

I'm trying to analyse a set of data with a non-normal response, 2 fixed 
effects and 1 nested random effect with strong heteroscedasticity in the 
model.

I planned to use the function lmer : lmer(resp~var1*var2 + (1|rand)) and 
then use permutations based on the t-statistic given by lmer to get 
p-values.

1/ Is it a correct way to obtain p-values for my variables ?

2/ I read somewhere that lme is more adequate when heteroscedasticity is 
strong. Do I have to use lme instead of lmer ?

3/ It is possible to fit a glm in lmer using family="...". Is it 
possible to use it in spite of hard heteroscedasticity ?

4/ A last question concerning SAS. My model appears to not converge in 
SAS, indicating a "structure" in the variance. Is it implying something 
in lmer or lme ?

Many Thanks

-- 
Alan Juilland

-- 
Alan Juilland – PhD Student
Department of Ecology and Evolution
Biophore, University of Lausanne
1015 Dorigny
Switzerland
Tel : ++41 21 692 41 74
Fax :  +41 21 692 41 65