Negative Hausman test

[Zainul Muhtarom]

Hello Stata-fan

When I use the command "hausman fe re" the chi-square result is negative (-810.83), when I change the command to "hausman re fe" the result is positive (810.83). Are the Hausman test results still valid after I change the command?

Thank you
Zainul Muhtarom

[Jeff Wooldridge]

The traditional form of the Hausman statistic requires "hausman fe re" (because FE is less efficient than RE under the ideal assumptions). So a negative test statistic is a problem. If you have time dummies this happens a lot, and the degrees-of-freedom might be computed incorrectly. The test also suffers in that it is not robust to serial correlation or heteroskedasticity. You should use xtoverid after using

xtreg y x1 ... xK d2 d3 ... dT re vce(robust)
xtoverid

xtoverid is a user-written command that has to be installed. It doesn't work with factor notation, so things like time dummies have to be created by hand (d2 ... dT above).

Actually, I just did an example, and xtoverid uses the incorrect df with time dummies. You should implement the correlated random effects approach by including the time averages of all time-varying variables.

[Zainul Muhtarom]

Thank you very much, sir. that was very helpful 😊