R help - Jul 2006 - denominator degrees of freedom and F-values in nlme

Hello,

I am struggling to understand how denominator degrees of freedom and 
subsequent significance testing based upon them works in nlme models.

I have a data set of 736 measurements (weight), taken within 3 
different age groups, on 497 individuals who fall into two 
morphological catagories (horn types).

My model is:  Y ~ weight + horn type / age group, random=~1|individual

I am modeling this using glmm.PQL function with family=neg.bin 
(negative binomial distribution, estimating theta based upon a glm 
without individual as a random effect). My data set will not be 
balanced, with varying numbers of measurements taken on different 
individuals and some individuals have no weight measures just a 
morphological type.

My output:
              denDF    numberdf
Intercept     495
weight        232       1
horn type     495       1
horn type:age 232       4

So my question is where do these denDF come from and how are they 
calculated? I wish to then test significane of these fixed effects and 
can get F-ratio's and P-values but are these appropriate?

Thank-you for your time.
Kind regards
Matthew

*********************************
Matt Robinson

Institute of Evolutionary Biology
Room 413, Ashworth Labs,
King's Buildings,
University of Edinburgh
EH9 3JT, UK

Tel: 0131 650 5990

On 7/8/06, M R Robinson <matthew.r.robinson at ed.ac.uk>
wrote:
> Hello,
>
> I am struggling to understand how denominator degrees of freedom and
> subsequent significance testing based upon them works in nlme models.
>
> I have a data set of 736 measurements (weight), taken within 3
> different age groups, on 497 individuals who fall into two
> morphological catagories (horn types).
>
> My model is:  Y ~ weight + horn type / age group, random=~1|individual
>
> I am modeling this using glmm.PQL function with family=neg.bin
> (negative binomial distribution, estimating theta based upon a glm
> without individual as a random effect). My data set will not be
> balanced, with varying numbers of measurements taken on different
> individuals and some individuals have no weight measures just a
> morphological type.
>
> My output:
>               denDF    numberdf
> Intercept     495
> weight        232       1
> horn type     495       1
> horn type:age 232       4
>
> So my question is where do these denDF come from and how are they
> calculated? I wish to then test significane of these fixed effects and
> can get F-ratio's and P-values but are these appropriate?
The algorithm for calculating those denominator degrees of freedom is
given in Chapter 2 of Pinheiro and Bates (2000), Mixed-effects Models
in S and S-PLUS, Springer.  It was designed to reproduce the results
of the BETWEENWITHIN option in SAS PROC MIXED.  On looking at that
algorithm recently I no longer feel that it is a good way of doing the
calculation but I don't have a better alternative at present.

Also, that algorithm and the use of the F test is suggested for linear
mixed models.  I'm not sure that it would apply "out of the box"
to a
generalized liner mixed model, which is what you are fitting here.
However, for practical purposes you could assume a "worst case" of 232
denominator degrees of freedom for all terms because there is so
little difference between an F statistic with 232 denominator degrees
of freedom and one with  495 denominator degrees of freedom.
> Thank-you for your time.
> Kind regards
> Matthew
>
> *********************************
> Matt Robinson
>
> Institute of Evolutionary Biology
> Room 413, Ashworth Labs,
> King's Buildings,
> University of Edinburgh
> EH9 3JT, UK
>
> Tel: 0131 650 5990
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
>