heteroscedasticity in panels

[Gianni Spolverato]

Dear all,

I have a panel dataset and runned a regression in the form of :

xtreg, dep ind , re

I can`t find a STATA command to test for heteroscedasticity , since the normal command of the Breush-Pagan test does not work here.
I tried:
estat hettest
The output I get is "estat hettest not valid".

Could anyone tell me what the proper command for a hettest would be here?

Best regards.

[Carlo Lazzaro]

Gianni:
if you susoect heteroskedasticity from visual inspection, the best fix is to impose clustered/robust SEs as they deal with both heteroskeadsticity and/or autocorrelation.
However, with many observations (as in a panel dataset), heteroskedasticity usually does not bite that hard.
As an aside, the code used in your analysis is surely different from the one you posted (which is apparently illegal).

[Phil Bromiley]

To expand on Carlo's suggestion, you'll get more useful assistance if you provide Stata code using code delimiters, Stata output, and sample data (using dataex) - see the FAQ on asking questions.

Some of the ancillary statistics are available for regress but not xtreg, fe. However, you can always run the same model in regress as a fixed effects xtreg by including i.panel as one of the regressors in the regression. It may take awhile or require you increase the matrix size. This doesn't work for random effects.

Depending on your discipline and substantive problem, it is likely that you'll need a Hausman test for fixed versus random effects.

[Attaullah Shah]

There is an excellent FAQ written by Vince Wiggins, which discusses how to find hetro in panel data, see http://www.stata.com/support/faqs/st...tocorrelation/