Hausman Test fe vs re

[Eliana]

Hello all,
we ran this command and we need help with the interpretation of the result.

. hausman fixed random

---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fixed random Difference S.E.
-------------+----------------------------------------------------------------
average_pl~t | -.1184973 -.1187574 .00026 .0109007
average_pl~e | -.050654 -.0053956 -.0452584 .1504227
average_age | .4468794 .3989634 .0479161 .166521
average_dr~d | -1.072317 -.9654437 -.1068731 .1911281
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test: Ho: difference in coefficients not systematic

chi2(4) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 0.33
Prob>chi2 = 0.9877


how can we know if fixed effects model is more appropriate than random effects model?

[Carlo Lazzaro]

Eliana (please read FAQ #6 about the preference on this forum for full name and surname identifier. Thanks):
the lack of statistical significance of the Hausman specification test points you to random effect linear panel data model.
For more on this topic, please take a look at -help hausman- and related entry (and references) in Stata 13.1 .pdf manual.

[Marcos Almeida]

Hello, Eliana (please re-register with your name + family name (you can do this by clicking on the "contact us" link below to the right),

Unfortunately, you didn't show your models, so we have less information to talk about.

As stated in the output, it tests if there is a systematic difference in the coefficients (within versus between) of two models.

In short, a statistically significant p-value indicates there may be a systematic difference in the coefficients, and we should prefer the fixed-effects ("within") model rather than include the random-effects ("between") model.

And vice versa.

There has been a recent discussion over the Hausman test on this thread:http://www.statalist.org/forums/foru...d-with-haumsna

If we type "help hausman" we learn that the first estimator is taken as the consistent one, whereas the second shall be efficient and consistent "under the hypothesis that you are testing, but inconsistent otherwise".


Best,

Marcos