OLS robustness tests

[Esraa Alharasis]

Hi all,

I am using OLS regression and I need to add robustness tests to ensure the validity of my findings? what are your suggestions regarding the robustness tests to the OLS regression?
I was thinking about the bootstrap standard error, what else I can use?

Thanks in advance

[Marcos Almeida]

If I understood right, you can use the - robust - option to get, well, robust SEs. But the validity of a study is related to several issues, not necessarily a test.

[Esraa Alharasis]

Thanks, Marcos, I meant the robustness standard error. There are some options that can robust the linear regression standard error, such as bootstrap. What do you think about the bootstrap option or the cluster option?

I used the "robust" option for each regression model.

Regards,
Esra'a

[Carlo Lazzaro]

Esra'a:
clustered standard errors are invoked when you detect autocorrelation (and, possibly, along with heteroskedasticity) in your residual distribution. In order to work properly, you should have a relevant number of clusters (otherwise, the resulting standard errors might be misleading), though. Hence, I do not see clustered standard errors as a way to the test the robustness (whatever that may mean) of their default counterparts.
Boostrap can be the way to go (200 replications are usually suggested for standard errors bootstrap estimates - see https://www.routledge.com/An-Introdu.../9780412042317, page 47).

That said, I would assess first the possible misspecification of the functional form of the OLS regressand via -linktest-.

[Esraa Alharasis]

Thanks, Carlo, that is useful.

Regards,
Esra'a