[R] multiple nesting levels in GEE

[Thomas Lumley]

On Sat, 29 May 2004 Aurelie.Cohas at univ-lyon1.fr wrote:

>  Hello,
>
> I'm actually trying to fit a gee model with 2 nesting levels since I expect a
> correlation between all members of a litter at a first level and between all
> individuals sharing a mother at a second superior level with an exchangeable
> matrix. I order my dataframe by both mother and litter

The gee() function does not compute this correlation structure (because
no-one has implemented it).  I don't think other software implements it
either, although people keep pointing out that it would be useful.

  This need not matter -- the point of GEE is that the correlation
structure is not important for validity of inference (although it may
affect efficiency).  If you are fitting a linear model I would recommend
lme(), otherwise you should just be able to use an exchangeable model at
the largest clustering size.

	-thomas


> I try several syntaxes:
> id= mother*litter which give the same correlation matrix as id= litter*mother
> and id=litter with a matrix size of 7*7 which correspond to the maximum number of
> young per litter
>
> id=litter|mother which give the same correlation matrix as id=mother|litter with
> a correlation matrix size of 235*235 which correspond to the number of
> observations
>
> id=litter/mother and id=mother/litter give different result with a strange matrix
> of 11*11 with no 1 on the diagonal
>
> If anybody know if it is possible to construct nesting levels in R and what is
> the good syntax for this kind of model?
>
> Thanks in advance
>
>
> Aurélie Cohas
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle