Thanks for the suggestions Nick. It is true that there is not much juice in this data. The correlation matrix shows a weak positive correlation.
t1 t2 t3
t1 1.0
t2 0.19 1.0
t3 0.01 0.4 1.0
I tired three methods: regress with cluster option, xtreg with re and the GEE. All the three give very similar results. The log transformation that I used in regress went a long way in eliminating hetroscedasticity and gave good results but couldn't give prediction intervals.
I used GEE thinking that I could explicitly model the correlation (doesn't do a very good job) and I could avoid the log transformation.
I "predicted" the residuals like you suggested and the residual vs. fitted plot tells me that hetroscedasticity is still very much a problem.
That leaves me with regular regression with robust estimators.
Thanks again for your comments.
Leny
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
