Re: st: panel date - test for heteroscedasticity within groups

[Austin Nichols <austinnichols@gmail.com>]

Katharina (gmx) <K.Raatz@gmx.de> :
I would not trust any SE with 12 or 14 obs in a regression, nor would I
trust a test for heteroscedasticity with that many obs.  Why not use
xtreg y x1 x2, fe i(fundid) robust
which is robust to both het and clustering (incl serial corr) instead?
How many panels do you have?
Is it true that no panel is more than 5% of the total sample?

You might need a test for clustering and/or het which considers
only a subset of coefs at a time, which is possible but not yet available
in a published command.

On Wed, Sep 14, 2011 at 10:14 AM, Katharina (gmx) <K.Raatz@gmx.de> wrote:
> Dear all,
>
> I am working on panel data which is uniquely characterized by an ID =
> (panelvar) and a date (timevar in the form of e.g. 1995-07). E.g. for ID =
> 1 there is data from 1995-07 until 1999-12, for ID 2 there is data from =
> 1992-01 until 2000-12 and so on. Now, I would like to test for =
> heteroscedasticity within these groups having the same ID. The general =
> "estat hettest" tests heteroscedasticity for the overall file containing =
> several groups, doesn't it? So how can I test for heteroscedasticity =
> within groups? (Question 1)
>
> Does STATA yield unreliable results if I use "by fundid: regress y x1 =
> x2, vce(robust)" for all groups irrespective of the fact whether their =
> variances are heterogenous or not? (Question 2) If it does yield =
> unreliable results, I would have to distinguish between regressions =
> including the robust feature and excluding it, right? (Question 3) Or is =
> there a more convenient way to test for heteroscedasticity within =
> groups?
>
> Thanks a lot! I'd really appreciate any help.
>
> Best, Kristin
*
*   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/