Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

AW: st: Testing for panel-level heteroskedasticity with xtgls


From   [email protected]
To   [email protected]
Subject   AW: st: Testing for panel-level heteroskedasticity with xtgls
Date   Tue, 1 Feb 2011 16:57:36 +0000 (GMT)

Dear Clive,

I'd like to thank you for your feedback. My problem is that i have unbalanced 
panel dataset, and because of that (I think), I cannot run directly xtpcse.

Moreover, I have a small dataset - N:17 panels and T:24 time periods. I was 
wondering if using xtgls or xtpcse is more appropriate in such cases given I 
also assume heteroscedastic panels and also serial correlation.

I looked at past posts but I saw no discussion on sample size and the use of 
these Stata commands when additional assumptions are imposed. In such cases, 
when datasets are small like in my case, is it ok to use either xtgls or xtpcse 
or it would be better to use other commands such as xtreg. 


Best,
Rado



----- Ursprüngliche Mail ----
Von: Clive Nicholas <[email protected]>
An: [email protected]
Gesendet: Montag, den 31. Januar 2011, 20:30:48 Uhr
Betreff: Re: st: Testing for panel-level heteroskedasticity with xtgls

[email protected] wrote:

> I am a bit confused from the result I got for a LR test. I want to test about
> heteroscedasticity across panels as suggested by the Stata team in their FAQ
> section.
>
> In particular, it is suggested to calculate iterated GLS with only
> heteroskedasticity first and save the likelihood. Then to fit the model but
> without the heteroscedasticity assumption (though it is not specified if this
> second model shall be iterated or it shall be left without this option).
>
> Below is the result I got. i would like also to ask if  my p-value of the test
> now indicates that I have heteroscedasticity or actually not? I am confused
> because they provide no explanation. Moreover, is it a bad sign to get no 
>output
> of estimates for ll(null)?
>
>
> . lrtest hetero . , df(`df') stats
>
> Likelihood-ratio test                                  LR chi2(16) =    493.43
> (Assumption: . nested in hetero)                       Prob > chi2 =    0.0000
>
> -----------------------------------------------------------------------------
>   Model |    Obs    ll(null)   ll(model)     df          AIC         BIC
> -------------+---------------------------------------------------------------
>           . |    367           .   -29.55142     29     117.1028    230.3583
>   hetero |    367           .    217.1648     46    -342.3296   -162.6829
> -----------------------------------------------------------------------------
>               Note:  N=Obs used in calculating BIC; see [R] BIC note

Going simply by the significance of your LR chi-square statistic in
your test, this would appear to support the hypothesis that your IGLS
model does have panel-level heteroscedascity.

Since this is case, maybe you should switch to fitting an OLS model
with panel-corrected standard errors (-xtpcse-) to deal with this
issue?

-- 
Clive Nicholas

[Please DO NOT mail me personally here, but at
<[email protected]>. Please respond to contributions I make in
a list thread here. Thanks!]

"My colleagues in the social sciences talk a great deal about
methodology. I prefer to call it style." -- Freeman J. Dyson.

*
*   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/




*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index