[R-sig-ME] df numbers in lmer
Ben Bolker
bbolker at gmail.com
Tue May 10 15:13:17 CEST 2011
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 05/10/2011 08:14 AM, Zofia Prokop wrote:
> Hello,
>
> I'm wondering how to obtain df-s associated with t values for fixed factors
> in lmer. I need them in order to calculate the effect sizes (Pearson's r)
> for those fixed factors.
>
> One method of calculating df number I came across is to take the total
> number of observation minus the number(s) of levels of random factor(s). But
> I have my doubts, such as
>
> (1) what about the number of fixed factors in the model - shouldn't it be
> subtracted as well?
>
> (2) I can see how this method can be applied to models where random factors
> are crossed with fixed factors but what about when they are nested? Or in
> case of more comlex designs? The model I'm struggling with at the moment is
> such:
> I have 10 females and 10 males crossed in a full factorial design, resulting
> in 100 offspring trait means (one per each pairing). Fixed factors are:
> female category and male category (2 levels each). I'm fitting a model:
>
> siz <- lmer(sz~femcat+malecat+(1|femid)+(1|maleid))
>
> What about those df-s for femcat and malecat?
>
> I'd be very grateful for advice as well as literature suggestions (I'm very
> new to mixed modelling and trying to get my head around it).
>
> -Zofia
>
As you may well have seen, this is a giant can of worms. See for
example <http://glmm.wikidot.com/faq>. I agree that if you are going to
do this by subtracting 'parameters' from the total number of
observations, you should subtract the number of fixed effect parameters
as well.
I would guess at 100-20-3 for your df, keeping in mind that this is
going to be approximate.
Ben Bolker
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk3JOe0ACgkQc5UpGjwzenN6tACcDVSQErDnA0HODI2MW7nLojM4
86kAnR41SgAsO1bEYlLyBpCvUTY6sufr
=b1Ik
-----END PGP SIGNATURE-----
More information about the R-sig-mixed-models
mailing list