Can anyone tell me how one would be able to determine the best correlation matrix to use when fitting a GEE model ? Is there an AIC parameter (or equivalent) to determine the best model? I have a situation whereby the parameter estimate have different signs depending on the correlation matrix I use eg parameter estimate is negative if I use an AR(1) structure whereas it's positive if I use an Exchangeable structure. Both are significant hence it makes interpretation very difficult. If I can determine the best model then I will be comfortable with the interpretation. You can see my dilema. I have already posted this message to the SAS users group but have had no response to date. In fact I got a lot of emails asking me to post any solutions. I will collate and post all responses. Thanks in advance Regards Max