[R-sig-ME] denominator DOF in lme, inner-outer rule

Ben Bolker bbo|ker @end|ng |rom gm@||@com
Thu Jun 6 01:35:32 CEST 2019

  Briefly, this is a "real" issue -- anova uses these "wrong" denDOFs.
(You can try an example and see for yourself!)

  Can you clarify your last paragraph?  As far as I remember (I'm not
looking at the details/code right now), PB define the "inner-outer" rule
fairly generally -- my understanding based on stuff I played around with
a few years ago is that it only *works* for random-intercept models, but
I don't remember seeing this clarified anywhere in the book? Maybe I
missed it ... in any case, IIRC the "inner-outer" rule implemented in
nlme, which is used in the anova method for lme objects, does not work
reliably for random-intercept models.

On 2019-06-05 10:19 a.m., Cristiano Alessandro wrote:
> Dear all,
> I have read in this
> <https://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#why-doesnt-lme4-display-denominator-degrees-of-freedomp-values-what-other-options-do-i-have>
> blog, that the function lme (in the nlme package) provides inaccurate
> denominator dof for random slope models according to the definition given
> by Pinheiro and Bates. Is this only a visualization issue, or such "wrong"
> denDOFs are actually those used if I run an anova on the fitted model. In
> the latter case, I guess the p-values and F-statistics provided by anova
> for random slope models should not be trusted, correct?
> I know that the estimation of the denominator DOF is a controversial topic.
> When I say "wrong" denDF I only mean that they do not reflect the
> estimation based on inner and outer rule defined by Pinheiro and Bates
> (which is used for random intercept only models).
> Thanks a lot
> Cristiano
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

More information about the R-sig-mixed-models mailing list