st: use of vwls against heteroskedasticity

["Marion Collewet" <collewet@ese.eur.nl>]

Dear Stata users,

I am estimating a model where I regress productivity in economic sectors on training and education level of the workforce in these sectors. The data I have are from different countries, and therefore of variable quality / reliability.
Tests indicate that there is a heteroskedasticity problem when I do a simple OLS regression.

In order to correct for the different response rates and reliability across countries, I want to estimate the model by feasible weighted least squares, using the vwls command. As an estimate of the conditional standard deviation of the dependent variable, I use the standard deviation of the residuals within each country (or alternative similar measures).

My main question is: I would like to check whether using this model indeed corrects for the heteroskedasticity problem. But after vwls, the postestimation function does not allow for hettest, and does not give residuals either. I suppose there is a good reason for that, and I would like to know which one.

I have computed the residuals by substracting the fitted values from the observed dependent variables, and conducted a 'manual' Breusch-Pagan test. According to this test, heteroskedasticity is still present in the model. However, if I simply weigh my data by the inverse of my estimated variance (using the aweight function), I obtain results very similar to the vwls estimation (identical coefficients and slightly different standard errors, as one would expect), and hettest then indicates that hetereoskedasticity has disappeared.

Can anybody help me understand what is happening there?
Thank you very much in advance.

Kind regards,

Marion Collewet


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