st: FEGLS from Wooldridge 2002 in operation

[Udaya Raj Wagle <udaya.wagle@wmich.edu>]

I am trying to figure out a way to operationalize the FEGLS approach mentioned in Ch. 10 of Wooldridge (2002) because the usual robust standard errors I am getting from "xtreg ..., fe r" are too large (relative to the usual ones). I understand that I am supposed to use residuals from the first stage regression that excludes one observation (t) as my weighting variable for the second fe regression. But I am not sure if I also need to follow the standard process of WLS to transform this weighting variable by regressing against all of the x's, which when I try gives no estimates at all. If I adopt this simple process of using residuals from the first stage regression as they are, on the other hand, I am not sure if they can effectively address the issues of heteroskedasticity and serial correlation. I would like to know what the process is and how to find out if the very issues for which GLS is needed in the first place are addressed.

Any suggestion will be highly appreciated.

------
Udaya R. Wagle, Ph.D.
Associate Professor
School of Public Affairs and Administration
Western Michigan University
1903 W Michigan Ave
Kalamazoo, MI 49008
Phone (269) 387-8934; Fax (269) 387-8935
Web: http://homepages.wmich.edu/~uwagle

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/