Testing for cross-sectional dependence in long panel T>N (if xttest2 does not work)

[Friedrich Osterhoff]

Dear Statalist,

I am currently working on an unbalanced panel with N=8 and observations per panel ranging from 73 to 2572, that is, T>N.

I need to test for cross-sectional dependence between my 8 Panels, in order to decide whether to use xtreg , fe with vce(cluster) or whether I should use xtscc , fe with Driscoll-Kraay standard errors, in order to account for correlation between groups. My understanding is that xtreg, fe would generate more reliable standard error estimates if there is no cross-sectional dependence (see “Robust Standard Errors for Panel Regressions with Cross-Sectional Dependence” by Daniel Hoechle, http://fmwww.bc.edu/repec/bocode/x/xtscc_paper.pdf).

So far, I have tried the Breusch-Pagan LM test for cross-sectional correlation in fixed effects models - xttest2. But I get the error message: "Correlation matrix of residuals is singular. not possible with test r(131)"

In the forum I have fond numerous posts on this subject, including posts with the same error message. However, the answers and hints are always based on datasets with T<N. Pesaran's xtcsd for example would not be valid for my T>N set, as far as I understand.

I would be grateful for any hint, how to overcome this problem and successfully test for cross-sectional correlation, either with an adjusted ttest2 or with alternative methods.

Thanks and best regards,
Friedrich

[Zahrah Rafique]

Hi,

I am facing this same issue. How did you fix yours?

Thanks!

[Carlo Lazzaro]

Friedrich:
since you have a large T, small N panel dataset, have you laready taken a look at -xtgls- and -xtregar-?