I am running a regression analysis on a panel data set. The Hausman test and the logical setup of the research question indicate that a fixed effects model would be best for running the regression.

But I am unsure now what kind of assumptions I need to test for ? The "normal" assumptions of heteroscedasticity, autocorrelation, exogeneity, ...? Or is it enough to run the Hausman test as it tells me the most important assumption for fixed effects is fulfilled (namely: the individual specific effect is correlated with the independent variables)?


Correlation of individual-specific effects with the regressors is not assumption of FE, FE merely is robust to this, unlike RE.

The fixed-effects transformation will wipe out any variable without time-variation (think gender), so if such a variable is the focus of your analysis, FE will not be suitable.

Also, exogeneity is indeed a major concern, as things like reverse causality or measurement error do not cease to be a concern once you employ panel data. (Indeed, measurement error is argued to have more serious consequences when using panel data.)

  • $\begingroup$ Thank you very much! Is there a statistical test for exogeneity? I remember reading somewhere the Hausman test can be used to test for this as well? If so, how do I interpret the output in this context? $\endgroup$ – Lena Feb 9 '16 at 10:36
  • $\begingroup$ There are tests for endogeneity, but these require you to have access to external instruments - essentially, you would need to have another estimator that would be consistent even in the presence of endogeneity, and then you could compare if the consistent estimator and FE differ significantly according to the idea of the Hausman principle. $\endgroup$ – Christoph Hanck Feb 9 '16 at 10:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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