I think you?ll agree from these numbers that this coefficient is useless for distinguishing between the four subset x MoSS combinations, and that we can therefore adopt the c on (e,u,u*e) model instead. But if you want to see the 1179 rows of detail data for u^2 underneath the above four roll-ups, let me know and I will send it along.
Also, just out of curiousity, I will run the same analysis for the u*e coefficient of c on (e,u,u*e) and post back the results.
