2
$\begingroup$

I want to run robustness tests for my model. For example, by reducing the sample to heavily concentrated groups, running a different regression (probit etc) etc. But, how do I ascertain that my results are robust? Is it sufficient that my key explanatory variables have same sign as the original model, magnitude of coefficients is similar and that they are significant? Is it necessary that the coefficients are exactly same?

Thank you

Improve this question
$\endgroup$
3
  • $\begingroup$ You may look at the standard errors, if they are big... $\endgroup$ Commented Apr 7, 2023 at 11:59
  • $\begingroup$ It’s not an easy question, +1. one way could be to fit the model by some robust estimators and compare the differences, if any. $\endgroup$ Commented Apr 7, 2023 at 12:21
  • $\begingroup$ You seem to use "robust" in a general sense. Thinking about robustness in a broadly defined sense might not be the most fruitful direction. It may be better to decide what quantities you want to estimate/compare and then estimate those robustly. For example, perhaps you are studying different groups (eg males vs females) and you want to compare those group in terms of a robust measure of central tendency. Or alternatively you might do model validation to ascertain that your model fits the data well. This second option seems to be what the question points towards. $\endgroup$ Commented Apr 7, 2023 at 16:24

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.