[R] estfun & df

[Douglas Bates]

On 10/8/07, Achim Zeileis <Achim.Zeileis at wu-wien.ac.at> wrote:
> On Mon, 8 Oct 2007, Abdus Sattar wrote:
>
> > Hello EVERYONE,
> >
> > I need an URGENT help from you please!
>
> This type of requests is not considered to be very polite, please have a
> look at the posting guide.
>
> > How can I see the "estfun" (empirical estimating function)
>
> I guess (because you are not telling us) that you would like to have an
> estfun() method for "lmer" objects. I don't provide one in my sandwich
> package (where the estfun generic is taken from) and AFAIK there is
> nothing analagous readily available in "lme4"... But I guess that you
> should be able to extract/compute the empirical estimating functions from
> the fitted "lmer" object.

> > and "df" (degree of freedom) from the following mixed-model please?

> > (fm1 <- lmer2(Reaction ~ Days + (Days|Subject), sleepstudy))

> When you compute
>    logLik(fm1)
> it returns a "logLik" object that has a "df" attribute:
>    attr(logLik(fm1), "df")

That degrees of freedom is a count of the number of parameters
estimated in the model.  That may be what Abdus Sattar wants but I'm
not sure.  Many times when people refer to the degrees of freedom for
linear mixed models they are thinking about the degrees of freedom for
a t-statistic or the denominator degrees of freedom for an F statistic
related to hypothesis tests on the fixed-effects parameters.  That was
what I was referring to in my response when I, in effect, said "I
don't want to talk about it any more."