Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: ivreg2 and xtoverid error


From   John Antonakis <john.antonakis@unil.ch>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: ivreg2 and xtoverid error
Date   Sat, 03 Apr 2010 18:45:05 +0200

Thanks Mark.

I realize that the test of endogeneity is not one of instrument validity. My thinking was that if OLS seems like it is not consistent then I am better off with IV, which is less efficient and given what evidence I currently have, probably consistent. The instruments are fixed-effects (of leaders); thus, these individual differences due to leaders (stemming from ability, personality, are constant and theoretically exogenous as they are mostly genetically determined).

Unfortunately, -ivregress- does not give me the first-stage critical values: it says "(not available)".

Thanks for your help.

Best regards,
John.

____________________________________________________

Prof. John Antonakis, Associate Dean 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

Faculty page:
http://www.hec.unil.ch/people/jantonakis

Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________



On 03.04.2010 16:56, Schaffer, Mark E wrote:
John,

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of John Antonakis
Sent: 03 April 2010 14:47
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: ivreg2 and xtoverid error

Hi Mark

Thanks for the note.

I get exactly the same estimates and standard errors with -ivreg- and -ivregress-, with the cluster robust variance estimator. When using -ivreg2- with the -noid- option it works and I get the same estimates; more importantly, I also get the Hansen J-test,

Looks like it was -ranktest- that was also causing invsym() to choke,
probably for the same reason - the reduced form first-stage matrix is
probably too big and badly behaved for it to cope.

which is what interests me most (the -ivregress- estimator does not report an overid for cluster-robust vce's):

Hansen J statistic (overidentification test of all instruments): 402.476, Chi-sq(404) P-val = 0.5121

Note, my estimates make sense and the regressors I expected to be significant are mostly significant. I guess I can assume that my estimates are consistent--and this because the endogeneity test is significant (from -ivregres-), right?

You can get -ivreg2- to give you an endogeneity test statistic with the
-endog- option.  I think it will agree with the -estat- test.

But -estat endogenous- and -ivreg2,endog()- are testing whether you need
to treat your endogenous regressors x1-x13 as endogenous (and the answer
seems to be yes, you do).  They aren't testing whether your equation is
underidentified or weakly identified, which is what you'd also like to
know.  This is the test that -ranktest- and -ivreg2- are choking on
because the matrix is so big.

I'd be curious to know if Stata's -estat firststage- is able to generate
the Cragg-Donald ("minimum eigenvalue") under/weak-identification
statistic in your case.  You'll need to use the -forcenonrobust- option
to get it if you're using cluster-robust.

Cheers,
Mark

. estat endogenous

  Tests of endogeneity
  Ho: variables are exogenous

  Robust regression F(13,417)     =  119.753  (p = 0.0000)
    (Adjusted for 418 clusters in lead_number)

Best,
J.

____________________________________________________

Prof. John Antonakis, Associate Dean
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

Faculty page:
http://www.hec.unil.ch/people/jantonakis

Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________



On 03.04.2010 15:16, Schaffer, Mark E wrote:
John,

You've got a very large number of instruments (almost 500),
but also
13 endogenous regressors. The under- and weak-id statistics are calculated by -ranktest-, and it's running of memory
because it works
with the reduced form in matrix form (13 first-stage
regressions, each
with almost 500 regressors).

If you use the -noid- option, it might address that problem problem (but you won't get the weak id stats).

The

invsym(): matrix has missing values

is different. -ivreg2- is having problems inverting this huge Z'Z matrix and -ivregress- isn't (I think possibly because the latter uses a different Mata matrix inversion function). I do
wonder whether
the results you're getting actually make sense.

Have you tried using the old -ivreg-, which uses -regress-? It is numerically very stable and accurate. Does it give the
same results
as -ivregress-?

Cheers,
Mark

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of John Antonakis
Sent: 03 April 2010 01:20
To: statalist@hsphsun2.harvard.edu
Subject: st: ivreg2 and xtoverid error

Hi:

I am running Stata version 11 and have everything up to date.

When running a two-stage model with ivreg2, I get the
following error:
. xi: ivreg2 sat (x1-x13=i.lead_n) , cluster(lead_n)
i.lead_number _Ilead_numb_1-484 (naturally coded;
_Ilead_numb_1
omitted)
quadcross(): 3900 unable to allocate real <tmp>[5421,5421]
                rkstat():     -  function returned error
                 <istmt>:     -  function returned error
r(3900);

If I remove the vce cluster command I get another error:

. xi: ivreg2 y (x1-x13=i.lead_n) ,
i.lead_number _Ilead_numb_1-484 (naturally coded;
_Ilead_numb_1
omitted)

invsym(): matrix has missing values

This model is estimated fine with the official -ivregress-
command,
whether I use a cluster-robust or normal variance estimator, e.g.,

. xi: ivregress 2sls y (x1-x13=i.lead_n) ,cluster(lead_n)
i.lead_number _Ilead_numb_1-484 (naturally coded;
_Ilead_numb_1
omitted)

Instrumental variables (2SLS) regression
Number of obs
=     832
Wald chi2(13)
=
1020.01
Prob > chi2 = 0.0000 R-squared = 0.7020 Root MSE = .2955

                          (Std. Err. adjusted for 418 clusters in
lead_number)
--------------------------------------------------------------
----------------
             |               Robust
y | Coef. Std. Err. z P>|z|
[95% Conf.
Interval]
-------------+------------------------------------------------
----------
-------------+------
x1 | .3849406 .0509433 7.56 0.000
.2850935
.4847877
x2 | -.0452167 .0541093 -0.84 0.403
-.1512689
.0608355
x3 | -.0214062 .0392473 -0.55 0.585
-.0983295
.0555171
x4 | .0743079 .0528296 1.41 0.160 -.0292363 .177852 x5 | .1559398 .056997 2.74 0.006
.0442276
.2676519
x6 | .168241 .0577832 2.91 0.004 .0549879 .281494 x7 | -.1359489 .0290323 -4.68 0.000
-.1928512
-.0790465
x8 | .0485811 .0358857 1.35 0.176
-.0217535
.1189157
x9 | -.1772587 .0512706 -3.46 0.001
-.2777472
-.0767701
x10 | .1785753 .0570718 3.13 0.002
.0667166
.2904339
x11 | .0309138 .0533183 0.58 0.562
-.0735883
.1354158
x12 | .2282491 .0554658 4.12 0.000 .1195381 .33696 x13 | -.0723148 .0486346 -1.49 0.137
-.1676369
.0230072
_cons | .4183937 .1616004 2.59 0.010
.1016627
.7351247
--------------------------------------------------------------
----------------
Instrumented:  x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13
Instruments: _Ilead_numb_2 _Ilead_numb_3 _Ilead_numb_4
_Ilead_numb_5
[output snipped]


Interestingly, when I run it with -xtivreg-, the model is
estimated
fine; however, -xtoverid- gives me the following error:

. xtoverid
invsym(): matrix has missing values
r(504);

This is the same error that follows ivreg2 estimation.

I suspect it might have something to do with the fact that
I have a
large number of instruments (fixed-effects, with 483
dummies) and clustering on those fixed-effects.

Best,
John.

--
____________________________________________________

Prof. John Antonakis, Associate Dean
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

Faculty page:
http://www.hec.unil.ch/people/jantonakis

Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________


*
*   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/



*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index