10:41 a.m.
Dear all,
I tried to estimate a random effects probit model using the --robust option. After this, I
re-estimated the model without it.
I got the same results (i.e. standard errors) in both cases and looking at the output it
became clear why, because Gretl said that estimation was "...based on inverse
Hessian".
However, I found this post from 2013, where Sven Schreiber already commented on this
issue.
http://lists.wfu.edu/pipermail/gretl-devel/2013-December/004835.html
However, I did not find an answer to this post, which brings me to my question, is it
possible use robust standard errors when one estimates a random effects probit model in
Gretl.
By "robust" I mean robust to some form of unspecified misspecification, like one
checks the robust option, when using ols.
Kind regards,
Jan
10:54 a.m.
Am 22.04.2016 um 12:41 schrieb Jan Tille:
Dear all,
I tried to estimate a random effects probit model using the --robust
option. After this, I re-estimated the model without it.
I got the same results (i.e. standard errors) in both cases and looking
at the output it became clear why, because Gretl said that estimation
was “…based on inverse Hessian”.
However, I found this post from 2013, where Sven Schreiber already
commented on this issue.
http://lists.wfu.edu/pipermail/gretl-devel/2013-December/004835.html
As time goes by....
However, I did not find an answer to this post, which brings me to my
question, is it possible use robust standard errors when one estimates a
random effects probit model in Gretl.
I don't think so. In the meantime I would repeat my Dec 2013 statement
that gretl should issue a warning, so that users don't think they have
actually estimated with robust standard errors.
Thanks for reminding.
By “robust” I mean robust to some form of unspecified misspecification,
like one checks the robust option, when using ols.
That is neither true for probit (with or without panel structure) nor
for OLS. If you're relying on robust standard errors as a cure for
unspecified misspec then you're cheating yourself (well, and your readers).
thanks,
sven
10:57 a.m.
Sven,
thank you for your prompt answer.
Meanwhile, I found the note in the user guide in chapter 18:
Robust standard errors
For most estimators, gretl offers the option of computing an estimate of the covariance
matrix
that is robust with respect to heteroskedasticity and/or autocorrelation (and hence also
robust
standard errors). In the case of panel data, robust covariance matrix estimators are
available for
the pooled and fixed effects model but not currently for random effects. Please see
section 17.4
for details.
Jan
-----Ursprüngliche Nachricht-----
Von: Plus.line MailSystem [mailto:cyrus@mailer.plusline.de] Im Auftrag von Sven Schreiber
Gesendet: Freitag, 22. April 2016 12:54
An: gretl-users(a)lists.wfu.edu
Betreff: Re: [Gretl-users] RE Probit --robust
Am 22.04.2016 um 12:41 schrieb Jan Tille:
Dear all,
I tried to estimate a random effects probit model using the --robust
option. After this, I re-estimated the model without it.
I got the same results (i.e. standard errors) in both cases and
looking at the output it became clear why, because Gretl said that
estimation was ".based on inverse Hessian".
However, I found this post from 2013, where Sven Schreiber already
commented on this issue.
http://lists.wfu.edu/pipermail/gretl-devel/2013-December/004835.html
As time goes by....
However, I did not find an answer to this post, which brings me to my
question, is it possible use robust standard errors when one estimates
a random effects probit model in Gretl.
I don't think so. In the meantime I would repeat my Dec 2013 statement that gretl
should issue a warning, so that users don't think they have actually estimated with
robust standard errors.
Thanks for reminding.
By "robust" I mean robust to some form of unspecified
misspecification, like one checks the robust option, when using ols.
That is neither true for probit (with or without panel structure) nor for OLS. If
you're relying on robust standard errors as a cure for unspecified misspec then
you're cheating yourself (well, and your readers).
thanks,
sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
1:09 p.m.
On Fri, 22 Apr 2016, Jan Tille wrote:
Sven,
thank you for your prompt answer.
Meanwhile, I found the note in the user guide in chapter 18:
Robust standard errors
For most estimators, gretl offers the option of computing an estimate of the covariance
matrix
that is robust with respect to heteroskedasticity and/or autocorrelation (and hence also
robust
standard errors). In the case of panel data, robust covariance matrix estimators are
available for
the pooled and fixed effects model but not currently for random effects. Please see
section 17.4
for details.
Panel-robust inference can be obtained rather easily by using the unit id
variable as clustering variable. For example, here's a slight variation
of the "reprobit.inp" sample script:
<hansl>
open union_wooldridge.gdt
# list regressors (including time dummies)
list X = const black married educ dummify(year)
# RE probit
probit union X --random
# simple probit, but panel-robust
probit union X --cluster=nr
</hansl>
If you want robust inference and RE at the same time, I'm not aware of an
available estimator (RE uses quite strict assumption on the distribution
of disturbances and individual effets --- I don't think there is a way to
bypass those).
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
1:20 p.m.
Jack,
thank you for the script. Meanwhile, I have estimated the re probit. As the LR test
(rho=0) could not be rejected, I switched to a pooled specification, where robust standard
errors are available.
Comparing the results, they look quite similar.
Best,
Jan
Random-effects Probit Probit
const -2,016** -2,046**
0,5365 0,5195
X1 0,1285** 0,1288**
0,02188 0,02127
X2 -0,0001056 0,003282
0,05446 0,05352
X3 0,2031** 0,2014**
0,09639 0,09358
X4 -0,06002 -0,06318
0,1124 0,1088
X5 -0,1715** -0,1622**
0,07222 0,06928
X6 0,0002977 0,0003395
0,004089 0,00395
lnsigma2 -3,366**
1,393
n 1450 1450
R2 0,029
lnL -961,4 -961,7
-----Ursprüngliche Nachricht-----
Von: Plus.line MailSystem [mailto:cyrus@mailer.plusline.de] Im Auftrag von Riccardo (Jack)
Lucchetti
Gesendet: Freitag, 22. April 2016 15:09
An: Gretl list <gretl-users(a)lists.wfu.edu>
Betreff: Re: [Gretl-users] RE Probit --robust
On Fri, 22 Apr 2016, Jan Tille wrote:
Sven,
thank you for your prompt answer.
Meanwhile, I found the note in the user guide in chapter 18:
Robust standard errors
For most estimators, gretl offers the option of computing an estimate
of the covariance matrix that is robust with respect to
heteroskedasticity and/or autocorrelation (and hence also robust
standard errors). In the case of panel data, robust covariance matrix
estimators are available for the pooled and fixed effects model but not currently for
random effects. Please see section 17.4 for details.
Panel-robust inference can be obtained rather easily by using the unit id variable as
clustering variable. For example, here's a slight variation of the
"reprobit.inp" sample script:
<hansl>
open union_wooldridge.gdt
# list regressors (including time dummies) list X = const black married educ
dummify(year)
# RE probit
probit union X --random
# simple probit, but panel-robust
probit union X --cluster=nr
</hansl>
If you want robust inference and RE at the same time, I'm not aware of an available
estimator (RE uses quite strict assumption on the distribution of disturbances and
individual effets --- I don't think there is a way to bypass those).
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
1:55 p.m.
Am 22.04.2016 um 15:20 schrieb Jan Tille:
Jack, thank you for the script. Meanwhile, I have estimated the re
probit. As the LR test (rho=0) could not be rejected, I switched to a
pooled specification, where robust standard errors are available.
Comparing the results, they look quite similar.
What you do is probably reasonable. However, the LR test is based on
assumptions that you subsequently discard by insisting on robust
standard errors. So it could be a little bit self-contradictory.
cheers,
sven
3228
days inactive
3228
days old
5 comments
3 participants
participants (3)
-
Jan Tille -
Riccardo (Jack) Lucchetti -
Sven Schreiber