Hausman test after fixed effect with robust standard error

[Faisal Abdullah]

Dear all,
I am trying to do the Hausman test but it does not work with robust standard errors. Are there any way to do it or it does not matter to adjust for the standard error when doing the Hausman test?

many thanks and regards
Faisal

[Clyde Schechter]

When you use the robust variance estimator in the FE and RE models, the Hausman statistic no longer has a chi square distribution. So -hausman- complains and refuses..

You can do a generalized Hausman test, however, by running both the FE and RE models without the robust VCE, storing the results after each (as with the -hausman- command, and then running -suest-. After -suest- then -test- for equality of the coefficients of each variable across models. When -suest- runs, it calculates robust variance estimates for the combined model, so you will get the test you are looking for. Using -suest- is a bit complicated, so I suggest you read the help file, and probably the manual section as well, before you proceed.

[Carlo Lazzaro]

Faisal:
you may also want to look at the following thread: http://www.stata.com/statalist/archi.../msg01069.html

[Mark Schaffer]

xtoverid, available from SSC in the usual way, will do a robust Hausman-like FE-vs-RE test after RE estimation.