2
$\begingroup$

I have panel data. When I use the Hausman test to decide on a fixed vs random model, I get an note saying that:

"the rank of the differenced variance matrix (8) does not equal the number of coefficients being tested (11)".

I am unsure what this means. I repeated this and converted my some of my factor and catagorical independent variables to simpler variables and scaled one variable to produce coefficients more similar to others which increased the rank to 9. Still, the note appeared as the rank, though improved, was unequal to the number of coefficients being tested. In all cases, the significant p value of the Hausman test indicated a fixed model.

I moved on to the xtoverid test. After running a random model (with and without the vce cluster option), there was an error message saying "saved RE estimates are degenerate (sigma_u=0) and equivalent to pooled OLS" implying I should use standard OLS regression.

Then I attempted the Mundlak approach - just for fun! The p value was not significant implying I should use a random model.

Any thoughts to which I should use and where I have gone would be very much appreciated!

$\endgroup$
  • $\begingroup$ Do you get coefficients for all your variables? Did you try sigmamore/sigmaless in the Hausman? $\endgroup$ – VCG Aug 22 '16 at 14:49
  • $\begingroup$ Thanks for your response @frage_man! Do you mean coefficients in the hausman test? If so, I got coefficients for my 11 time-variant variables for both the fixed and random models and the difference. However, in the 4th column, I only got standard errors for 7 of the variables. This is accompanied by the note, "the rank of the differenced variance matrix (8) does not equal the number of coefficients being tested (11)". When I try sigmamore and sigmaless, I still get the note, "the rank of the differenced variance matrix (2) does not equal the number of coefficients being tested (11)". $\endgroup$ – Carol Aug 22 '16 at 15:39
  • $\begingroup$ Oh I meant in your FE model. Look into no reported SEs for Stata as I recall that not being a good thing, meaning don't trust Hausmans if your FE doesn't report right. $\endgroup$ – VCG Aug 22 '16 at 15:44
  • $\begingroup$ Oh yeah - in the FE model I get coefficients and SEs for all my time-varying variables. Do you think the note that crops up in the Hausman test invalidates the result of the test? $\endgroup$ – Carol Aug 22 '16 at 15:48
  • $\begingroup$ Have you googled your error for info from the Stata forum? I just wouldn't trust the Hausman if it's reporting an error. $\endgroup$ – VCG Aug 22 '16 at 16:03

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.