Difference between COVB in genmod's model statement versus repeated statement?

[mlamias]

I noticed that that you can place the "COVB" option after the model statement or after the repated statement when building a GEE model using genmod.  The resulting matrices are totally different.  Does anyone know what the difference is between the covariance matrices that are produced?  I realize that the COVB after the reapeated statement produces both a model-based covariance matrix as well as the empiracal-based one, but I would have thought the COVB option on the model statement would have been the same.  The documentation doesn't make it clear how these are different.

Thanks.

[Ksharp]

Just guess Maybe the type of correlation matrix is different .
Check option type=
 type=exch (exchangeable) OR Independent .

[Rick_SAS · Accepted solution]

The doc does make it clear, but you might be looking in the worng place.

On the model statment, the COVB option estimates the usual covariance of the parameters for the fixed effects. This is explained in the section on "Generalized Models Theory".  It is related to the Hessian (second derivative matrix) of the likelihood function.

For GEEs, you can also model the covariance between repeated measures for each subject. That matrix is explained in the section for "Generalized Estimating Equations."

[mlamias]

Perfect.  Thanks so much!  If I couldn't find this information under the COVB option in the genmod documenation, then no, the documenation didn't make it clear.  I'd ask that this information be added or at least hyperlinked to in the genmode documentation under the COVB options.