[R] Degrees of freedom in repeated measures glmmPQL

[Charlotte Burn]

Apologies, I made a mistake with my maths. The degrees of freedom look
correct, assuming they are the denominator and that glmms work this
way. I just under-estimated the number of data points I had. Sorry.

Charlotte


On 02/05/07, Charlotte Burn <charlotteburn at googlemail.com> wrote:
> Hello,
>
> I've just carried out my first good-looking model using glmmPQL, and
> the output makes perfect sense in terms of how it fits with our
> hypothesis and the graphical representation of the data. However,
> please could you clarify whether my degrees of freedom are
> appropriate?
>
> I had 106 subjects,
> each of them was observed about 9 times, creating 882 data points.
> The subjects were in 3 treatment groups, so I have told the model to
> include subject as a random factor nested within treatment.
> There are two other variables and I'm interested in their two-way
> interactions with Treatment.
> I'm happy with the model structure, and the output generally looks right, but...
>
> In the 'DF' column of the output table, it has 882 as the degrees of
> freedom for all the variables (except Treatment itself, which has 0
> degrees of freedom). At the bottom of the output, it says Groups:
> Subjects = 106, Treatment = 3.
>
> Should I be worried or is this what to expect?!
>
> I was expecting it to be more like an ANOVA table, where the error
> degrees of freedom should reflect the number of subjects, not all the
> data points.
>
> I can't see the usual differentiation between the numerater and
> denominator/error degrees of freedom, so am I right in thinking the DF
> column shows the error degrees of freedom? Or do glmms not work like
> this?
>
> Thank you very much in advance,
> Charlotte
>


-- 
------------------
Dr Charlotte C. Burn
Department of Animal Welfare and
Behaviour
School of Clinical Veterinary Science
University of Bristol
Langford House
Bristol BS40 5DU
Tel: 0117 9219134
http://seis.bristol.ac.uk/~frccb/charlotteburn.html