[R] Denominator Degrees of Freedom in lme() -- Adjusting and Understanding Them

[Spencer Graves]

Prof. Bates may be able to give us more recent references on this, but 
the best literature I know on this is Pinhiero and Bates (2000) 
Mixed-Effects Models in S and S-Plus (Springer, sec. 2.4).  This 
includes description of a "simulate.lme" function, which you can use to 
generate random numbers according to a given assumed model and then 
compare some results with a reference distribution.  Something like this 
could be used to answer your question of what is the correct number of 
degrees of freedom to use for any particular model. 

hope this helps.  spencer graves

Douglas Bates wrote:

>Contributions of code to provide alternative calculations of
>denominator degrees of freedom are welcome :-)
>
>I think it would be good to bear in mind that the use of the t and F
>distributions for models with mixed effects is already an
>approximation.  If your design is such that you end up with a very few
>denominator degrees of freedom then the whole question of whether you
>should be using F or t distributions in the first place becomes
>problematic.   If the number of denominator degrees of freedom is
>moderate than the distinction between alternative methods becomes
>unimportant.
>
>  
>