[R-sig-ME] Satterthwaite´s df in lme
Douglas Bates
bates at stat.wisc.edu
Mon Mar 21 21:13:34 CET 2011
On Thu, Feb 24, 2011 at 12:05 PM, Flores-de-Gracia, Eric
<eef201 at exeter.ac.uk> wrote:
> Hi all.
> I am just running a lme model and found some criticism regarding the fact that DF appears just as integers.
> What are the implications of this, in terms of reporting those DF and the reliability of my model when other packages like SPSS that do mixel models reports on DF based on Satterthwait´s correction?
> My data set is unbalanced, so should I expect this to affect the calculation of DF using lme?
There are many formulas for approximation of degrees of freedom in
linear mixed models but they are all approximations. That is, the
ratio of the estimate and the approximate standard error isn't
distributed as a T so the question of which degrees of freedom works
best is a bit vague to begin with.
Typically the exact number of degrees of freedom chosen only makes a
difference when the sample sizes are small or the number of levels for
one of more of the grouping factors for the random effects is small.
In practice the difference between a T distribution with 40 degrees of
freedom and a T distribution with 400 degrees of freedom is
negligible.
If the degrees of freedom provided by the summary of an lme fit is
reasonably large then the particular approximation used is not
important.
More information about the R-sig-mixed-models
mailing list