1:27 p.m.
When I run the code in the stata, the results are matching with gretl only when I specify
the robust standard errors. Why is this case?
xtreg ANS FDIENT FDICAP FDIAST FDITUR FDIGDP FDIROTC FDIROFA FDIROS FDIEMP FDICOM FDIWAG
GDPgrowth Size, re
Random-effects GLS regression Number of obs = 434
Group variable: Id Number of groups = 62
R-squared: Obs per group:
Within = 0.7754 min = 7
Between = 0.0372 avg = 7.0
Overall = 0.1280 max = 7
Wald chi2(13) = 525.11
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000
------------------------------------------------------------------------------
ANS | Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
FDIENT | .3538051 .1119847 3.16 0.002 .1343191 .5732911
FDICAP | .0560138 .0341982 1.64 0.101 -.0110135 .123041
FDIAST | -.0488365 .027242 -1.79 0.073 -.1022297 .0045567
FDITUR | -.0249275 .0195643 -1.27 0.203 -.0632728 .0134179
FDIGDP | .0761888 .0221277 3.44 0.001 .0328193 .1195583
FDIROTC | -.0140346 .0176176 -0.80 0.426 -.0485645 .0204953
FDIROFA | .0029775 .0053075 0.56 0.575 -.007425 .01338
FDIROS | -.0061958 .0053613 -1.16 0.248 -.0167037 .004312
FDIEMP | .0093037 .0294696 0.32 0.752 -.0484557 .0670631
FDICOM | -.0078266 .0252738 -0.31 0.757 -.0573623 .041709
FDIWAG | -.2533413 .3531677 -0.72 0.473 -.9455373 .4388548
GDPgrowth | .0565587 .0193329 2.93 0.003 .0186669 .0944505
Size | -12.14111 .6405544 -18.95 0.000 -13.39658 -10.88565
_cons | 69.71459 2.813264 24.78 0.000 64.2007 75.22849
-------------+----------------------------------------------------------------
sigma_u | 1.6051608
sigma_e | 1.0781735
rho | .68909913 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Hausman test results are also different.
Hausman test -
Null hypothesis: GLS estimates are consistent
Asymptotic test statistic: Chi-square(13) = 341.943
with p-value = 3.83701e-65
Stata:
Test of H0: Difference in coefficients not systematic
chi2(13) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= -918.16
Warning: chi2 < 0 ==> model fitted on these data
fails to meet the asymptotic assumptions
of the Hausman test; see suest for a
generalized test.
With sigmamore:
Test of H0: Difference in coefficients not systematic
chi2(13) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 237.83
Prob > chi2 = 0.0000
data:
https://www.dropbox.com/scl/fi/o9copqsun64if9n7vyu4r/2017035-data.dta?rlk...
6:42 p.m.
Am 10.04.2026 um 15:27 schrieb Ramki S:
When I run the code in the stata, the results are matching with gretl
only when I specify the robust standard errors. Why is this case?
xtreg ANS FDIENT FDICAP FDIAST FDITUR FDIGDP FDIROTC FDIROFA FDIROS FDIEMP FDICOM FDIWAG
GDPgrowth Size, re
I don't understand where you were specifying the robust
covariance
estimation, in Stata or in gretl? Gretl would not automatically employ
the robust estimates, but I don't know your Stata settings.
Hausman test results are also different.
Hausman test -
Null hypothesis: GLS estimates are consistent
Asymptotic test statistic: Chi-square(13) = 341.943
with p-value = 3.83701e-65
Gretl by default uses a regression-based form of the
Hausman test, in
the spirit of Davidson / MacKinnon. For comparison purposes you could
try the --matrix-diff option.
Stata:
Test of H0: Difference in coefficients not systematic
chi2(13) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= -918.16
Warning: chi2 < 0 ==> model fitted on these data
fails to meet the asymptotic assumptions
of the Hausman test; see suest for a
generalized test.
This is a nice example of the widespread problem of
negative test
statistics. It cannot happen with the auxiliary regression approach.
With sigmamore:
Test of H0: Difference in coefficients not systematic
chi2(13) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 237.83
Prob > chi2 = 0.0000
And this is also a nice example of my old argument that negative test
statistics typically mean that H0 is wrong. (Shameless plug, but it's
quite old:
https://www.degruyterbrill.com/document/doi/10.1515/jbnst-2008-0407/html)
cheers
sven
8:21 p.m.
On Fri, 10 Apr 2026, Ramki S wrote:
When I run the code in the stata, the results are matching with
gretl only when I specify the robust standard errors. Why is this
case?
I don't find that to be the case. If I run the same random-effects
regression in gretl and Stata 12.1 I get the same coefficients and
standard errors without using the --robust option.
I have the data file downloaded as ramki_panel.dta and here's my gretl
script:
open ramki_panel.dta -q
setobs Id Year --panel
panel ANS const FDIENT FDICAP FDIAST FDITUR FDIGDP FDIROTC \
FDIROFA FDIROS FDIEMP FDICOM FDIWAG GDPgrowth Size --random
foreign language=stata --send-data
xtset id year
xtreg ans fdient fdicap fdiast fditur fdigdp fdirotc fdirofa fdiros fdiemp fdicom fdiwag
gdpgrowth size, re
end foreign
(On Linux at least, Stata coerces all variable names to lower-case on
CSV importation.)
Allin
5:48 p.m.
On 4/10/26 22:21, Allin Cottrell wrote:
On Fri, 10 Apr 2026, Ramki S wrote:
> When I run the code in the stata, the results are matching with
> gretl only when I specify the robust standard errors. Why is this
> case?
I don't find that to be the case. If I run the same random-effects
regression in gretl and Stata 12.1 I get the same coefficients and
standard errors without using the --robust option.
In attachment, there is a gretl4py script for exact replication of
random effects example from Principles of Econometrics 4th Edition
(chapter 15.4, pages 556 and 560). As far as I remember, Hill, Griffiths
and Lim used Eviews when writing the textbook.
Cheers,
Marcin
12
days inactive
13
days old
3 comments
4 participants
participants (4)
-
Allin Cottrell -
Marcin Błażejowski -
Ramki S -
Sven Schreiber