[R-sig-ME] Function Likelyhood for GEE

[Ken Beath]

On 17/04/2008, at 5:04 AM, marcioestat at pop.com.br wrote:

>
>
> Hi listers,
>
> I am analyzing an anova with repeated measures with the function
> GEE and I would like to verify the best structure for the matrix of
> variance and covariance... So I need to obtain the function of
> likelyhood to make the test of comparation between two models (full  
> and
> reduced)... Does anybody could tell me how do I obtain the
> likelyhood...
> (-2log(likelyhood M0/likelyhood M1))
>

Likelihood doesn't exist for GEE models.

If the robust SE are similar to the working then the covariance  
structure is probably close enough. You-Gan Wang (http://www.cmis.csiro.au/You-Gan.Wang/ 
) and others have developed some more technical methods, but they  
don't seem to be widely applied.

Ken