Hello all,
I'm trying to report Hausman test results in my paper to support why I'm using a fixed effects model instead of a random effects one.
I have the following results directly from Stata:
I think I'll be fine in reporting these results, I'm just wondering what the last column of the Hausman test is reporting. I don't really understand these numbers nor can I seem to compute them myself to see how they were produced. I don't really know what the sqrt(diag(V_b-V_B)) means.. from this video https://www.youtube.com/watch?v=54o4-bN9By4 I understand the Hausman statistic for one variable is:
I figured the last column might represent this, but it doesn't.
Also, I understand how the Hausman test works (at least to some extent). There's probably more to it, but I understand that when the difference in Beta's from the fixed and random effects is large enough, the p-value will become smaller and thus is more likely to be significant. Is there anything I can state about which variables cause my Hausman test to be significant? In other words, which variables would have the greatest difference between fixed and random effects coefficients? I would say in the table above that's age, 1.civstatus, all of the educat variables and the potexp one. However, I don't know if there's some lower level from which point upwards the difference is seen as 'large'. I'm just comparing it to the rest of my variables, for which the difference seems small (agesq, 1.status and potexpsq).
Many thanks in advance.
Best,
Arne
I'm trying to report Hausman test results in my paper to support why I'm using a fixed effects model instead of a random effects one.
I have the following results directly from Stata:
Code:
---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fixed random Difference S.E.
-------------+----------------------------------------------------------------
age | -924.7776 2337.739 -3262.516 1017.576
agesq | -17.10693 -.8481488 -16.25878 4.485318
1.status | -5350.406 -5495.106 144.7001 215.3709
1.civstatus | 3084.459 4565.355 -1480.895 439.4448
1bn.educat | 12736.08 11630.62 1105.466 1255.357
2.educat | 9869.011 16590 -6720.989 1639.005
3.educat | 15512.34 24160.17 -8647.82 1851.586
4.educat | 25918.98 41191.66 -15272.68 2271.385
potexp | 7554.487 2962.601 4591.887 1006.716
potexpsq | -68.79827 -77.82987 9.031605 4.145204
------------------------------------------------------------------------------
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(9) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 110.29
Prob>chi2 = 0.0000
Code:
W = (BFE - BRE) / ( Var. of BFE - Var. of BRE) where FE = fixed effects and RE = random effects
Also, I understand how the Hausman test works (at least to some extent). There's probably more to it, but I understand that when the difference in Beta's from the fixed and random effects is large enough, the p-value will become smaller and thus is more likely to be significant. Is there anything I can state about which variables cause my Hausman test to be significant? In other words, which variables would have the greatest difference between fixed and random effects coefficients? I would say in the table above that's age, 1.civstatus, all of the educat variables and the potexp one. However, I don't know if there's some lower level from which point upwards the difference is seen as 'large'. I'm just comparing it to the rest of my variables, for which the difference seems small (agesq, 1.status and potexpsq).
Many thanks in advance.
Best,
Arne
