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: RE: xtivreg2 -- Instruments dropped
From
DE SOUZA Eric <[email protected]>
To
"[email protected]" <[email protected]>
Subject
RE: st: RE: RE: xtivreg2 -- Instruments dropped
Date
Mon, 17 Jan 2011 09:50:11 +0100
You are misunderstanding the jargon.
Let y be regressed on X and W, where X is endogenous and W is exogenous.
Let Z be the exogenous variables used as instruments that are not part of the regression equation.
W and Z are all instruments:
W is referred to as the included (in the regression equation) instruments
Z is referred to as the excluded (from the regression equation) instruments
The jargon comes from simultaneous equation models.
In a way one can say that W instruments itself
Eric
Eric de Souza
College of Europe
Brugge (Bruges), Belgium
http://www.coleurope.eu
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Ivan Png
Sent: 17 January 2011 00:18
To: [email protected]
Subject: Re: st: RE: RE: xtivreg2 -- Instruments dropped
Dear Mark:
Many thanks. Sorry, I introduced the space when cutting and pasting from my STATA do file into the email. I could run the xtivreg2.
Thanks for the below: you have replicated my problem exactly. The next to last line of your output reads, " Excluded instruments: L.govr_r "
-- that's precisely my problem. Why did xtivreg2 exclude the instrument that I wanted in (viz, L.govr_r)? This problem did not arise with xtivreg.
I encountered the same with other instruments that I tried for the endogenous variable L.utsa. I checked all the instruments by using OLS to regress utsa on the various instruments -- they were significant in the OLS. So, I'm baffled why xtivreg2 would exclude them.
On Sun, Jan 16, 2011 at 10:29 PM, Schaffer, Mark E <[email protected]> wrote:
> Ivan,
>
> I used your corrected dataset, and made some progress in tracking down the problem, but there are still aspects I don't understand.
>
> 1. Easy part: to identify your versions of xtivreg2, ivreg2, xtivreg,
> etc., you use the -which- command. It's best to use the -all- option
> with it, in case there are multiple versions lurking on your machine.
> For example,
>
> . which xtivreg, all
>
> C:\Program Files\Stata11\ado\updates\x\xtivreg.ado
> *! version 1.5.9 11nov2009
>
> C:\Program Files\Stata11\ado\base\x\xtivreg.ado
> *! version 1.5.7 17jun2009
>
> NB: The dulpicate versions above aren't a problem for an official program like -xtivreg-. When you update your Stata and a do file gets updated, the update goes in the updates folder and the original stays in the base folder.
>
> 2. I can't run your example, using either -xtivreg2- or official -xtivreg-. The reason is a bug, or perhaps "inflexibility", in the parsing done by -xtivreg2- and -xtivreg-. It is very simple: if you use a space between the end of an -if- statement and the comma, both -xtivreg2- and -xtivreg- complain in a very uninformative way:
>
> First, -xtivreg2-:
>
> . xi: xtivreg2 xrdr_ln revtr_ln mktbook_mm ebitdar_ln ///
>> taxcred (L.utsa = L.govr_r) i.year ///
>> if xrdr_count >= 6 , fe cluster(state_sic)
> i.year _Iyear_1976-2006 (naturally coded; _Iyear_1976
> omitted)
> 6 invalid name
> r(198);
>
> end of do-file
>
> r(198);
>
> Now official -xtivreg-:
>
> . xi: xtivreg xrdr_ln revtr_ln mktbook_mm ebitdar_ln ///
>> taxcred (L.utsa = L.govr_r) i.year ///
>> if xrdr_count >= 6 , fe cluster(state_sic)
> i.year _Iyear_1976-2006 (naturally coded; _Iyear_1976
> omitted)
> 6 invalid name
> r(198);
>
> end of do-file
>
> r(198);
>
> But delete the spaces between 6 and the comma, and it works:
>
> . xi: xtivreg2 xrdr_ln revtr_ln mktbook_mm ebitdar_ln ///
>> taxcred (L.utsa = L.govr_r) i.year ///
>> if xrdr_count >= 6, fe cluster(state_sic)
> i.year _Iyear_1976-2006 (naturally coded; _Iyear_1976
> omitted) Warning - singleton groups detected. 295 observation(s) not used.
> Warning - collinearities detected
> Vars dropped: _Iyear_2006
>
> FIXED EFFECTS ESTIMATION
> ------------------------
> Number of groups = 3198 Obs per group: min =
> 2
> avg =
> 8.5
> max =
> 30
>
> IV (2SLS) estimation
> --------------------
>
> Estimates efficient for homoskedasticity only Statistics robust to
> heteroskedasticity and clustering on state_sic
>
> Number of clusters (state_sic) = 1471 Number of obs =
> 27044
> F( 34, 1470) =
> 3.85
> Prob > F =
> 0.0000 Total (centered) SS = 251.0415486 Centered
> R2 = 0.2476 Total (uncentered) SS = 251.0415486
> Uncentered R2 = 0.2476 Residual SS = 188.8924922
> Root MSE = .089
>
> ----------------------------------------------------------------------
> --------
> | Robust
> xrdr_ln | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+--------------------------------------------------------
> -------------+--------
> utsa |
> L1. | .0174583 .0902999 0.19 0.847 -.1595263
> .1944429
> |
> revtr_ln | .1265504 .0205411 6.16 0.000 .0862907
> .1668102
> mktbook_mm | .0149745 .3796314 0.04 0.969 -.7290894
> .7590385
> ebitdar_ln | .0408917 .0159209 2.57 0.010 .0096873
> .0720961
> taxcred | -.1079932 .5174864 -0.21 0.835 -1.122248
> .9062615
> _Iyear_1977 | -.0313348 .0448701 -0.70 0.485 -.1192785
> .056609 ...
> <output edited for brevity>
> ...
> _Iyear_2005 | -.0105457 .0033898 -3.11 0.002 -.0171896
> -.0039017
> ----------------------------------------------------------------------
> -------- Underidentification test (Kleibergen-Paap rk LM statistic):
> 46.831
> Chi-sq(1) P-val =
> 0.0000
> ----------------------------------------------------------------------
> -------- Weak identification test (Cragg-Donald Wald F statistic):
> 202.397
> (Kleibergen-Paap rk Wald F statistic):
> 53.412 Stock-Yogo weak ID test critical values: 10% maximal IV size
> 16.38
> 15% maximal IV size
> 8.96
> 20% maximal IV size
> 6.66
> 25% maximal IV size
> 5.53
> Source: Stock-Yogo (2005). Reproduced by permission.
> NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
> ----------------------------------------------------------------------
> -------- Hansen J statistic (overidentification test of all
> instruments): 0.000
> (equation exactly
> identified)
> ----------------------------------------------------------------------
> --------
> Instrumented: L.utsa
> Included instruments: revtr_ln mktbook_mm ebitdar_ln taxcred
> _Iyear_1977 ...
> <output edited for brevity>
> ... _Iyear_2002 _Iyear_2003 _Iyear_2004
> _Iyear_2005 Excluded instruments: L.govr_r Dropped collinear:
> _Iyear_2006
> ----------------------------------------------------------------------
> --------
>
> 3. I still don't understand what you mean when you say
>
>> IP: I meant instruments -- in the sense that I used different
>> instruments in various specifications, and each one was dropped.
>
> I'm pretty sure you don't mean that -xtivreg2- drops every excluded instrument but still reports a coefficient for an endogenous regressor! But then, what do you mean?
>
> But maybe it doesn't matter if simply eliminating the spaces between 6 and the comma solves your problem.
>
> Best wishes,
> Mark
>
>> -----Original Message-----
>> From: [email protected]
>> [mailto:[email protected]] On Behalf Of Ivan Png
>> Sent: 16 January 2011 03:25
>> To: [email protected]
>> Subject: Re: st: RE: RE: xtivreg2 -- Instruments dropped
>>
>> Sorry, Mark. I've updated the data-set.
>>
>> On Sat, Jan 15, 2011 at 10:29 PM, Schaffer, Mark E
>> <[email protected]> wrote:
>> > Ivan,
>> >
>> > I had the same problem as Eric. And there must be
>> something else wrong, since there's only one instrument listed
>> (L.govr_r) and you refer to "instruments".
>> >
>>
>> IP: I meant instruments -- in the sense that I used different
>> instruments in various specifications, and each one was dropped.
>>
>> > Did xtivreg2 report collinearities among your instruments
>> as a reason for dropping some? Did xtivreg? If so, were the dropped
>> variables the same?
>>
>> IP; Neither reported collinearities.
>>
>> > You might also want to compare the first-stage regressions
>> of xtivreg2 and xtivreg.
>> >
>> > Oh, and tell us which versions of xtivreg2, xtivreg and
>> Stata you are using.
>>
>> IP: Stata/IC 10.1. Sorry, please advise how to identify version of
>> xtivreg2 and xtivreg
>>
>>
>> > --Mark
>> >
>> >> -----Original Message-----
>> >> From: [email protected]
>> >> [mailto:[email protected]] On Behalf Of
>> DE SOUZA
>> >> Eric
>> >> Sent: 15 January 2011 12:02
>> >> To: [email protected]
>> >> Subject: st: RE: xtivreg2 -- Instruments dropped
>> >>
>> >> Check the instruction you have posted:
>> >> govr_r does not exist
>> >>
>> >>
>> >>
>> >> Eric de Souza
>> >> College of Europe
>> >> Brugge (Bruges), Belgium
>> >> http://www.coleurope.eu
>> >>
>> >>
>> >> -----Original Message-----
>> >> From: [email protected]
>> >> [mailto:[email protected]] On Behalf Of Ivan
>> >> Png
>> >> Sent: 15 January 2011 12:43
>> >> To: [email protected]
>> >> Subject: st: xtivreg2 -- Instruments dropped
>> >>
>> >> Dear Stata-list:
>> >>
>> >> I'm having difficulty with xtivreg2
>> >>
>> >> Each time, the system drops the instruments for the
>> variable L.utsa.
>> >> utsa is a state-time variable that varies across states
>> and times. I
>> >> checked the instruments: they are state- and time-varying.
>> I don't
>> >> understand why the system drops the instruments. (I tried xtivreg
>> >> and it works, but I would like to use xtivreg2 so that I can get
>> >> cluster standard errors and the diagnostics for weak instruments).
>> >>
>> >>
>> >> Please help. Many thanks!
>> >>
>> >> Data-set:
>> >> http://www.comp.nus.edu.sg/~ipng/research/company_wrkg.dta
>> >>
>> >>
>> >> Code:
>> >> . xtset gvkey year
>> >>
>> >> . xi: xtivreg2 xrdr_ln revtr_ln mktbook_mm ebitdar_ln taxcred ///
>> >> (L.utsa = L.govr_r) i.year trend_AL - trend_WY ///
>> >> if xrdr_count >= 6 , fe cluster(state_sic)
>> >>
>> >>
>> >>
>> >>
>> >> --
>> >> Ivan Png
>> >> National U of Singapore
>> >>
>> >>
>> >>
>> >>
>> >> --
>> >> Skype: ipng00
>> >> T: +65 6516 6807
>> >>
>> >>
>> >>
>> >> --
>> >> Skype: ipng00
>> >> T: +65 6516 6807
>> >>
>> >> *
>> >> * 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/
>> >>
>> >
>> >
>> > --
>> > Heriot-Watt University is a Scottish charity registered
>> under charity
>> > number SC000278.
>> >
>> >
>> > *
>> > * 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/
>> >
>>
>>
>>
>> --
>> Skype: ipng00
>> T: +65 6516 6807
>>
>> *
>> * 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/
>>
>
>
> --
> Heriot-Watt University is a Scottish charity registered under charity
> number SC000278.
>
>
> *
> * 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/
>
--
Skype: ipng00
T: +65 6516 6807
*
* 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/