Unbalanced panel data - test for heteroskadisticity

[Kate Mill]

Dear all,

I have unbalanced panel data and want to use the Breusch Pagan test and the White test to test for heteroskadisticity. I can't seem to find the right commands.

Can someone explain to me if these tests can be used with panel data, so after xtreg, and what the commands are?

Thanks in advance,

Best regards

[Jeff Wooldridge]

It's not easy to test for heteroskedasticity after random effects or fixed effects estimation. It's possible, and I can tell you how to do it if you give more details about your estimation. But why do you want to do it? Serial correlation is typically much more of a worry. And usually I wouldn't test for that, either, because I can cluster my standard errors (which accounts for serial correlation and heteroskedasticity). What are your specific data dimensions?

[Kate Mill]

Dear Mr. Wooldridge,

Thank you for your reply. The reason I want to test for it is that the instructions tell me to. I have to include them for the completion of my work. I need to show why I have to take the log of some variables for example.

Do you mean with "cluster my standard errors", robust standard errors?

What dimensions do you mean? My dependent variable is FDI per GDP outflows, from host to home country, so bilateral. My ID are country pairs, my independent variables are posssible determinants of FDI.

[Kate Mill]

Dear Prof. Wooldridge,

I have read more about it in the mean time.

Is it correct that I can solve for heteroskedasticity by simply using robust standard errors? So the White test is not even necssary.

Also I understand that serial correlation is autocorrelation and this can be solved with clustered standard errors. Could you explain to me how I can do this, and can I also combine clustered and robust standard errors?

I'm new at STATA and doing these tests for the first time. I hope you can help me with this.

Maybe I also have to mention I have orthogonalized some of my independent variables.

Thank you in advance

[Jeff Wooldridge]

Kate: It would be much easier if you showed something about your data and commands. Generally, you want something like the:

xtreg y x1 ... xK, fe cluster(id)

The standard errors are robust to any kind of heteroskedasticity and serial correlation. Of course, so are all test statistics and confidence intervals. I'm not sure why you would orthogonalize your independent variables. That will make their coefficients hard to interpret (even though it artificially eliminates multicollinearity). JW

[Attaullah Shah]

User written program xttest2 is available, its ssc description provides the following information
DESCRIPTION/AUTHOR(S)

xttest2 calculates the Breusch-Pagan statistic for
cross-sectional independence in the residuals of a fixed effect
regression model or a GLS model estimated from cross-section
time-series data. xtreg, fe estimates this model assuming
independence of the errors. A likely deviation from independent
errors in the context of pooled cross-section time-series data
(or panel data) is likely to be contemporaneous correlations
across cross-sectional units. These correlations are those
exploited by Zellner's seemingly unrelated regression (SUR)
estimator, and this test is provided by sureg, corr in that
context. This version of the routine is updated from that
published in SJ 1-1 to properly handle unbalanced panels, and
will work after ivreg2 on panel data.

KW: contemporaneous correlation
KW: panel data
KW: cross-section time-series

Requires: Stata version 6.0


Author: Christopher F Baum, Boston College
Support: email [email protected]

Distribution-Date: 20110815


INSTALLATION FILES (type net install xttest2)
xttest2.ado
xttest2.hlp

[Cláudio Carvalho]

Ok, I already know how to correct heteroskedasticity, but I'm still not sure how to formally test it (i.e. diagnose) in random effects model. Is there an equivalent method to wald test (which is applicable to fixed effects) for random effects?

[Carlo Lazzaro]

Kate:
There's no need to combine cluster and robust standard errors under a panel framework. In fact you can use -cluster- and -robust- option interchangeably.
It is worth noting that this feature does not work for cross-sectional regression commands, such as -regress-.