Robustness Check

[Mohit Arvind]

Hi,

For my project, I've just been running simple OLS regressions. I was wondering if there was any test I could use to check the robustness of the results?

Edit: Just to add, my data is cross sectional and the tests I want to run are to see the validity of my results

Thanks

[Carlo Lazzaro]

Mohit:
see: -estat hettest-; -estat ovtest- or -linktest-.

[Mohit Arvind]

Hi Carlo,

Thanks for reply.

The Breush-Pagan test is indicating that I do have a heteroscedasticity issue within my model. Does this mean there is an issue with my model specification? And therefore, does this mean that I have to change what variables I use in the model?

[Marcos Almeida]

You didn’t say much about the model, command and output.

That said, you shouldn’t rely only in this test. I recommend to use graphics for that matter as well, such as those with residual versus fit, residuals versus predicted, checking ‘culprit’ variables, etc.

Not to forget, modeling involves centering, scaling, adding new variables, using interaction terms, squared or cubic polinomials, etc.

In the end, shall heteroscedasticy persist, using robust errors is a nice way to curb the problem.

[Carlo Lazzaro]

Mohit:
I do share Marcos' helpful advice.
Heteroskedasticity in itself can be managed by -robust- standard errors, as Marcos suggests.
There are instances in which heteroskedasticity disappears just logging the dependent variable.

[Mohit Arvind]

Oh thanks!

This answers my issue. I can manage Heteroskedasticity if I run my regression command and the just add ', robust' at the end. This is what I've been doing anyways

[Marcos Almeida]

Well, not precisely. As previously remarked, we need to be responsible for the modeling. Relying on robust errors can become the last step, not the first. Anyway, applying robust errors won’t be enough to technically ‘beautify’ a clumsy model.