Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: RE: Hausman test (via a Wald test)
From
John Antonakis <[email protected]>
To
[email protected]
Subject
Re: st: RE: Hausman test (via a Wald test)
Date
Sun, 01 May 2011 20:45:17 +0200
Agreed--but this is not what I am getting it.
The first test is the Durbin-Wu-Hausman augmented regression (Davidson,
R., & MacKinnon, J. G. (1993). Estimation and inference in econometrics.
New York ; Oxford: Oxford University Press) ; the second test is using a
procedure discussed by Mundlak, Y. (1978). Pooling of Time-Series and
Cross-Section Data. Econometrica, 46(1), 69-85.
These are Wald tests, but not of the sort I was discussing, i.e,.
testing a coefficient against a specific value. I think that the safest
was out of this is the augmented regression or something similar,
because the Wald test I am talking about treats the specific value as a
population value, which is evidently not right in most situations.
Best,
J.
__________________________________________
Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis
Associate Editor
The Leadership Quarterly
__________________________________________
On 01.05.2011 19:01, Brian P. Poi wrote:
On 05/01/2011 12:01 PM, John Antonakis wrote:
Right, Eric.
But some estimators are more efficient; take the case where you compare
IV estimates (consistent) versus OLS estimates (efficient). The Hausman
test will suggest that an efficient estimator is not consistent in the
case where, for example, a regressor is endogenous, because the
consistent vs efficient estimates will be significantly different, as
tested by the usual Hausman test:
(d_IV - d_OLS)/(SE_d^2_IV - SE_d^2_OLS)^-1.
Thus, I am suggesting that one uses the Wald postestimation test
following the OLS model to determine whether the estimate/s of the OLS
model is/are significantly different from the estimate/s of the IV
model. My question has to do with the fact that the Wald test, in this
case, tests the OLS coefficient/s against a specific value/s (that of
the IV estimator) and ignores the variance in the IV estimate/s--so I
was wondering whether this was econometrically sound to do.
In the case of IV versus OLS, there is a way to use a Wald test after
OLS to see whether the IV and OLS estimates are significantly
different, though it doesn't escape the need to have instruments
available in the first place. In this case, fit the regression
y1 = y2*beta + X1*gamma + V*theta + error
by OLS where y2 is your endogenous regressor(s), X1 are exogenous
variables, and V are the residuals from the regression(s)
y2 = X1*pi1 + x2*pi2 + error
with extra instruments X2. Test whether theta = 0 using a Wald test;
a rejection of H0 suggests that y2 is endogenous and therefore that
OLS is not consistent.
But to reiterate, this presumes you are able to come up with
instruments in the first place so that you can add that V term to your
OLS regression.
>> I was wondering, though, if anyone is aware of literature showing that
>> a Wald test could do too (which would be particularly useful in the
>> case of models where the Hausman test can't be used in Stata); more
>> specifically, the Wald test I am suggesting is to test whether the
>> parameters (of interest) from the efficient estimator are
>> significantly different from those of the consistent estimator. It
There's also a way to do a Wald test equivalent to a Hausman test of
fixed versus random effects. Instead of doing
. xtreg y x1 x2, re
do
. egen x1bar = mean(x1), by(panelvar)
. egen x2bar = mean(x2), by(panelvar)
. xtreg y x1 x2 x1bar x2bar, re
. test x1bar x2bar
Wooldridge derives this test in his cross-sectional and panel textbook.
There are probably other examples of Hausman-like tests that can be
done as Wald tests, too.
-- Brian Poi
-- [email protected]
Best,
J.
__________________________________________
Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis
Associate Editor
The Leadership Quarterly
__________________________________________
On 01.05.2011 17:03, DE SOUZA Eric wrote:
John,
I don't get your point. Consistency refers to the parameter estimate,
efficiency refers to the variance of the estimate.
The Hausman principle compares two estimators, both of which are
consistent under the null but only one under the alternative.
This is quite general.
Eric de Souza
College of Europe
Brugge (Bruges), Belgium
http://www.coleurope.eu
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of John
Antonakis
Sent: 01 May 2011 16:51
To: [email protected]
Subject: st: Hausman test (via a Wald test)
Hi:
Given that sometimes Stata cannot compute the Hausman test (because it
is undefined or because the estimation procedure does not allow it), I
have been thinking about ways to go around this limitation (and find
SUEST to be particularly useful in this regard).
I was wondering, though, if anyone is aware of literature showing that
a Wald test could do too (which would be particularly useful in the
case of models where the Hausman test can't be used in Stata); more
specifically, the Wald test I am suggesting is to test whether the
parameters (of interest) from the efficient estimator are
significantly different from those of the consistent estimator. It
would be very simply to do and could accommodate a large class of
estimators.
However, this test is only a constraint on the coefficient estimate/s
from the efficient estimator and ignores the variance of estimates
from the consistent estimator (thus I don't know to what extent this
test would be useful).
Any thoughts?
Best,
John.
--
__________________________________________
Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis
Associate Editor
The Leadership Quarterly
__________________________________________
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/