[Limdep Nlogit List] Hausman Test Interpretation: Limdep Output

[William Greene]

Nelson.

I suppose unclear is in the eye of the beholder.  The language of the
diagnostic is

High values of H favor FEM
or
(Low) values of H favor (REM)

both of which are correct.  I think your understanding is not correct,
since as the difference between the RE and FE estimators grows, the
evidence in favor of the FEM grows, not the opposite.  If the REM is
correct, the FE and RE estimators should be similar.

/Bill Greene


----- Original Message -----
From: "Nelson Daou" <ndaou at fontanagroup.com>
To: limdep at limdep.itls.usyd.edu.au
Sent: Tuesday, February 17, 2009 6:25:54 PM GMT -05:00 Columbia
Subject: [Limdep Nlogit List] Hausman Test Interpretation: Limdep Output

The Hausman statistic tests against H0: that the Random Effects coefficients in a panel data model are not different from the Fixed Effects coefficients.  Alternatively, I have seen H0: that Fixed Effects coefficients are appropriate to use.  In terms of Fixed vs. Random model selection, one would use FEM unless the value of H merited the rejection of H0.  My confusion is that the LIMDEP output says that High (Low) values of H favor FEM (REM).  If my understanding of the distribution of the Hausman statistic is correct, the opposite would be true.  In other words, H would grow along with the differences between FEM and REM coefficients.  Is my understanding correct, or is the LIMDEP output unclear?

Thank you,
Nelson
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au