[R] degrees of freedom in lme

Jean-Baptiste Ferdy Jean-Baptiste.Ferdy at univ-montp2.fr
Tue Jun 26 10:08:32 CEST 2007


>> This is such a common question that it has a an "FAQ-like" response from 
Doug Bates. Google "lmer p-values and all that" to find the response. 

>Isn't this a different question, though, since Jean-Baptiste is using
>nlme. 

>Details on the calculation of DF in nlme can be found in chapter 4 of
>the book by Pinheiro and Bates "Mixed Effects Models in S and S-PLUS.
>Using the formula provided, I get denDF of 10 for level 1 and 32 for
>level 2. I'm not sure why lme is using the denDF estimated at level 2 in
>this example ...

My question was more on the fact that anova.lme seems to behave differently 
when called on a two way ANOVA than when called on an ANCOVA. I am sure that 
DenDF is estimated with the same procedure in the two situations and, I as 
already said, I understand there is a real issue with calculating dfs and 
p-values in this kind of situation. 

But again I thought this issue was serious only in the case of unbalanced data 
sets. When data are balanced my guess was that lme and the traditional strata 
approach would give essentially the same answer. This is indeed what happens 
in the case of an ANOVA (and that's what Pinehiro and Bates say in section 
2.4 of their book) but not in the case of an ANCOVA.

Does all this mean that we should forget the F-test provided by lme and turn 
to the approach used in lme4??
-- 
Jean-Baptiste Ferdy
Institut des Sciences de l'Évolution de Montpellier
CNRS UMR 5554
Université Montpellier 2
34 095 Montpellier cedex 05
tel. +33 (0)4 67 14 42 27
fax  +33 (0)4 67 14 36 22



More information about the R-help mailing list