I have a panel of firm data and my supervisor recommended vce(cluster firmID) for clustering the standard errors. However, the vce(robust) command yields higher significance for relevant predictor. Actually, this model has the best fit out of the OLS models. Of course, my supervisor could not predict these results beforehand, but I wonder what may explain the result and how I can argue for using robust instead, or should I include both?

  • 1
    $\begingroup$ In Stata terms these are options, not commands. The fit, meaning parameter estimates, is identical across these choices. vce() just controls standard error calculation. Which standard errors should be calculated depends on your ideas about the generating process, not a search for best buy models driven by numerical criteria. As your question has a statistical core, I am not voting to close, but its title would seem to imply a request for coding advice, so should be rewritten. I have edited only trivially: a more drastic rewriting is up to you. $\endgroup$ – Nick Cox Mar 4 '19 at 12:38
  • $\begingroup$ I'd ask this on Statalist where the issues will be familiar to more people, but make any cross-posting explicit. $\endgroup$ – Nick Cox Mar 4 '19 at 12:40
  • $\begingroup$ Take a look here. $\endgroup$ – Dimitriy V. Masterov Mar 4 '19 at 18:34
  • $\begingroup$ Thank you for these suggestions. Forgive me if I am naive, my Interclass Correlation Coefficient for y, ID is 0,87 suggesting that ids can be clustered? Again, this option yields insignificant coefficients. How do I justify a different option in this case? $\endgroup$ – Kristian Pal Mar 5 '19 at 16:53

Your Answer

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

Browse other questions tagged or ask your own question.