Dear Statalist,
I am confused by the terms "cross-sectional heteroskedasticity" and "groupwise heteroskedasticity". I tried to find concrete definitions but my search made me even more confused.
According to my understanding a test for groupwise heteroskedasticity (in Stata xttest3) indicates whether the error is homoskedastic within groups but heteroskedastic across / between groups. Regarding cross-sectional heteroskedasticity I am unsure what it exactly means. I guess it means exactly the opposite of groupwise heteroskedasticity?
According to the command xttest3 my small panel of T=10 and N=12 suffers from groupwise heteroskedasticity. According to the Pesaran test it does not suffer from cross-sectional dependence and it is not autocorrelated (I don´t know if that´s relevant for my question).
To account for general heteroskedasticity I read that I should use robust standard errors. However, when using the robust together with xtreg Stata clusters (here for the 12 countries). Regarding clustering, the Stata help file says that clustering "produces an estimator of the VCE that is robust to cross-sectional heteroskedasticity".
Question: What does this statement exactly mean? Will clustering account for my groupwise heteroskedasticity? Or is it useless in my case? And is there a test for only cross-sectional heteroskedasticity in Stata (in case this is not the same as groupwise heteroskedassticity)?
Thank you very much for your help in making the differences between those terms more clear.
All the best,
Leon
I am confused by the terms "cross-sectional heteroskedasticity" and "groupwise heteroskedasticity". I tried to find concrete definitions but my search made me even more confused.
According to my understanding a test for groupwise heteroskedasticity (in Stata xttest3) indicates whether the error is homoskedastic within groups but heteroskedastic across / between groups. Regarding cross-sectional heteroskedasticity I am unsure what it exactly means. I guess it means exactly the opposite of groupwise heteroskedasticity?
According to the command xttest3 my small panel of T=10 and N=12 suffers from groupwise heteroskedasticity. According to the Pesaran test it does not suffer from cross-sectional dependence and it is not autocorrelated (I don´t know if that´s relevant for my question).
To account for general heteroskedasticity I read that I should use robust standard errors. However, when using the robust together with xtreg Stata clusters (here for the 12 countries). Regarding clustering, the Stata help file says that clustering "produces an estimator of the VCE that is robust to cross-sectional heteroskedasticity".
Question: What does this statement exactly mean? Will clustering account for my groupwise heteroskedasticity? Or is it useless in my case? And is there a test for only cross-sectional heteroskedasticity in Stata (in case this is not the same as groupwise heteroskedassticity)?
Thank you very much for your help in making the differences between those terms more clear.
All the best,
Leon
Comment