Clustered standard errors

[Larissa MM]

Hi,

I ran fixed effects regression on my panel data and found that heteroskedasticity and autocorrelation are present.

So I used xtreg- vce(robust). How can I explain the clustered robust standard errors, like how others mentioned using specific names; newey west/ Huber/White

Do I just mention clustered robust standard errors in my paper?

Thanks!

[Carlo Lazzaro]

Larissa:
see -help _robust- and related entry in Stata .pdf manual.
That said, I would consider enough mentioning clustered robust standard errors in your paper.

[Phil Bromiley]

To expand on Carlo's comment, clustered robust is in a sense more conservative than just robust or neither. The norms on these things vary by area, but many areas are comfortable with the more conservative choice automatically.

[Carlo Lazzaro]

Larissa:
see also https://www.stata.com/support/faqs/s...cs/references/.

[Eric de Souza]

With panel data fixed effects, robust and cluster robust will give you the same standard errors because Stata substitutes cluster robust for robust. the reason is that correcting panel data FE estimators just for heteroscedasticity results in inconsistent standard errors.
That said, you can simply write: the standard errors have robustified for heteroscedasticity and autocorrelation.

[Aneta Blazejowska]

Should I use clustered standard errors if I have cross-sectional dependence? I dont know how to model this problem if tests indicate serial correlation and heteroskedastcity because my T is smaller than N (N/T equals to about 5).

[Carlo Lazzaro]

Aneta:
if by cross-sectional dependence you mean within-panel correlation of the idiosyncratic error and you've also detected heteroskedasticity, in short panels (N>T) clustered (or robust) statndard errors take both these features of your data into account.