I am currently doing the analysis using firm-level data of three countries combining together. Also, it is a cross-sectional analysis. Therefore, in the regression, I control for industry dummies to account for differential effects in industry level.

I think I don't need to control for country dummies due to some reasons: (1) The countries that I use in my analysis share similar characteristics including development level, political, and socio-economic conditions. (2) There are only three countries in my analysis.

In this case, I expect that the results will not be affected even though I don't control for country specifics.

However, I am not sure if my idea is appropriate since I cannot find any reference to support it yet. Therefore, I would like to ask everybody here if you have any idea whether or not country dummies is necessary in my case.

Thank you!


1 Answer 1


I would advise you to think deeper about your research question first, as this will guide the decision to use country fixed effects. If you would like to exploit cross country variation, for example by studying how the same industry functions differently across countries, then do not use country dummies because it will absorb the variation you want.

If you want to study a causal relation that is independent of country, then using "within-country" variation (i.e. include country dummies) is the most restrictive approach and will sweep out all fixed country characteristics. Your coefficient will then be identified off deviations from within country means.

Are you sure that countries are similar across most characteristics in order to defend not using country dummies? It may be hard to defend solving omitted variable bias on those grounds. If you really want to go this route, then include several country covariates as controls.

  • $\begingroup$ Thank you for your idea. My objective is to study a causal relation that is independent of country. So, I also think that including country dummies will be the right thing to do. However, I got a problem when I include those country variables, that is the parallel assumption for ordered probit regression is violated. So, I am now trying to figure out any possible solution. I would also appreciate it if you have some suggestions, like to run the regression for each country separately or other possible ways. $\endgroup$
    – Vannaro
    Commented May 21, 2020 at 5:42

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.