genmod not printing corrw

[psh23]

I would like to calculate an ICC from my genmod model. When I request 'corrw' the model runs for hours and then eventually SAS freezes. I have tried specifying all correlation structures and this doesn't change anything. The model runs quickly without 'corrw'. I can't figure out what the reason might be for this?

proc genmod data=all descending;

where site='temp';

class msex race marst urbrur registry;

model er = msex race marst urbrur / link=log dist=bin;

repeated subject=registry / corrw;

run;

Thank you!

[StatDave]

Since you haven't specified a correlation matrix structure, the independence structure is used and there are no correlation estimates to report. The correlation matrix is just the identity matrix. See the "Details: Generalized Estimating Equations:Working Correlation Matrix" section of the GENMOD documentation for details on the structures and the correlations that are estimated for each.

[psh23]

It does not change if I specify a correlation structure - which is why it is not included in the code above.

[StatDave · Accepted solution]

Check your data to see how many observations there are in the largest cluster (level of REGISTRY). Or, if you get partial output, see if the GEE Model Information table is there which will show the Correlation Matrix Dimension. The appearance of SAS freezing is probably actually due to it having to take a lot of time computationally because of a large cluster size. The time needed to estimate and display all of the correlations in a large matrix could be considerable. If you can reduce the cluster size, then that will reduce the time needed. If you intend to use only the independence or exchangeable structure, then you don't need the CORRW option. The single correlation estimate for the exchangeable structure is printed by default without the CORRW option. You could also try fitting the same model using the newer PROC GEE instead of GENMOD.