hausman test interpretation

[KIM SUJIN]

dear statalists,

I got hausman test result



---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
Ban | -.4745226 -.4745226 0 0
Language | .1680624 .1680624 0 0
Log_Distance | -1.377694 -1.377694 0 0
Log_GDP_i | .4833085 .4833085 0 0
Log_GDP_j | .8848895 .8848895 0 0
Log_rem_i | -.0116353 -.0116353 0 0
Log_rem_j | .0108957 .0108957 0 0
------------------------------------------------------------------------------
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(0) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 0.00
Prob>chi2 = .
(V_b-V_B is not positive definite)

I have no idea how to interpret it
.
anyone could let me know?
thank you/

[Carlo Lazzaro]

Kim:
your -fe- and -re- specifications seem to return identical results.
Are you sure that you did not unadvertently store the same estimate twice and then run -hausman-?

[KIM SUJIN]

i did , i used same but different option!
xtreg Log_Value Ban Language Regionalization Log_Distance Log_GDP_i Log_GDP_j Remoteness_i Remoteness_j,fe
xtreg Log_Value Ban Language Regionalization Log_Distance Log_GDP_i Log_GDP_j Remoteness_i Remoteness_j,re


is this problem~?

[Carlo Lazzaro]

Kim:
no it isn't, as this is the way you should have done it.
Just out of curiosity: what if you run:

xtreg Log_Value Ban Language Regionalization Log_Distance Log_GDP_i Log_GDP_j Remoteness_i Remoteness_j,fe
estimates store fe
xtreg Log_Value Ban Language Regionalization Log_Distance Log_GDP_i Log_GDP_j Remoteness_i Remoteness_j,re
estimates store re
hausman fe re

[KIM SUJIN]

yes that code is exactly what i've done!

I dont know why result is unusual

[Carlo Lazzaro]

Kim:
did you investigate the presence of panel-wise effect (eg, via the F-test appearing as a footnote of the -xtreg,fe- outcome table and -xttest0- after -xtreg,re)?

[KIM SUJIN]

no i havent~ what is that for~??

[Carlo Lazzaro]

KIm:
1) when significant, the F-test appearing (automatically, if you use default standard errors) as a footnote of the -xtreg,fe- outcome table proves the evidence of panel-wise effect
2) when significant the -xttest0- proves the evidence of random effect after -xtreg,re-.

[KIM SUJIN]

thank you Carlo!!

I also wish to know if I can use fixed effect after heckman!!

if you or anyone know~~ thank you!!!

[KIM SUJIN]

hi.. i've god two results


Fixed-effects (within) regression Number of obs = 6,936
Group variable: YearPeriod Number of groups = 72

R-sq: Obs per group:
within = 0.3368 min = 17
between = 0.6873 avg = 96.3
overall = 0.3580 max = 258

F(7,6857) = 497.56
corr(u_i, Xb) = 0.0149 Prob > F = 0.0000



-------------------------------------------------------------------------------
Log_Value | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
Language | .7993348 .0665767 12.01 0.000 .6688238 .9298459

sigma_u | .29905682
sigma_e | 2.0607748
rho | .02062507 (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------------

F test that all u_i=0: F(71, 6857) = 1.05 Prob > F = 0.3605


.

f result is two!
i dont know what should i use..
THankyou///

[Carlo Lazzaro]

Kim:
-xtreg,fe. entry in Stata .pdf manual explains the mistery.
The seconf F-test (taht assume causes your concern), tests whether a panel-effect does exist in your dataset.
As it fails to reach statistical significance, the aswer is: no. Hence, your -xtreg,fe- model can be safely replaced with an OLS.
As an aside, it sounds really strange that you have one predcitor only in the right-hand side of your regression equation.

[KIM SUJIN]

thank you sooooo much ~!!!!!!

i removed below results

haha

thank you again!!!

[KIM SUJIN]

sorry again but i also got
Prob > F 1.000 result in random effect model, means that is doenst fit to random effect model.
is that possible that i dont use fe, re both?

thank you

[Carlo Lazzaro]

Kim:
please post via CODE delimiters what you typed and what you got from Stata. Thanks.