Tests of robustness?

[Larissa Sheer]

Hey all,

I am new in Statalist and I am looking for help concerning tests of robustness. I shall do "Robustheitstests" but I do not know what exactly that means and how I do it. Do I just have to put some controle variables in my regression, vary the controle variables several times and then I have to look if there are statistical significant differences in the coefficients of my dependent variables? And if there are big differences, does that mean that my regression model is not robust? And when do I know that the differences are big enough to be statistical significant? I really do not know how to do it in the right way and Google did not help me well. I hope you could please help me with this issue.

[Marcos Almeida]

Welcome to the Stata Forum / Statalist,

First, it would be interesting to know which test you are using. Also, the outputs of both regressions.

That said, by "robust" I understand the application of robust standard errors (Huber-White), such as when we add - vce(robust) to the command line, not to be taken as the robust regression (rreg), seldomly used (I guess), but theoretically applied when one wishes to perform (instead of the OLS regression) a regression analysis under too many outliers.

The differences will happen (if so) according to violations of the OLS assumptions, such as homoscedasticity.