Statalist


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

Re: st: RE: re: Re: Suppressing the Constant in the First Stage with ivreg2


From   Anson Soderbery <asoderbe@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: re: Re: Suppressing the Constant in the First Stage with ivreg2
Date   Thu, 18 Jun 2009 11:09:58 -0700

That is exactly what I am trying to say (poorly I admit), Mark. It is not really a classic regression framework that leads to interpretable parameters (and as a consequence there is little intuition behind the first stage). I wanted to go to LIML to check if it will improve the small sample performance, and 3sls is not a possibility in practical applications. Thank you all for your help.
On Jun 18, 2009, at 8:31 AM, Schaffer, Mark E wrote:

Anson,

Another way to put it is that the IV estimator is a single-equation,
limited information estimator.  You say that your model is structural,
so in effect you are saying that your "first-stage" regression is not a
standard IV first-stage linear projection at all - it's a structural
equation.  If so, then you are no longer in IV land; you want to use
instead a multiple-equation, full information estimator, e.g., 3SLS.

It's worth emphasising that IV is still giving you consistent estimates,
and the rationale for moving to 3SLS or some other system estimator is
to get more efficient estimates.  Of course, the usual
efficiency-robustness tradeoff also kicks in - 3SLS might be more
efficient than IV, but it also requires more assumptions than IV for
consistency.

Hope this helps.

Cheers,
Mark

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Kit Baum
Sent: 18 June 2009 16:07
To: statalist@hsphsun2.harvard.edu
Subject: st: re: Re: Suppressing the Constant in the First
Stage with ivreg2

<>	
Anson said

Thank you for the response.  You are exactly right about
include a vector of ones and specifying noc, it will be
included.  Is there a way to force ivereg2 not to include an
exogenous var (ie iota) in the
FSRs? All I can say is the model is structural, and there is a heavy
theoretical justification for suppressing the constant in the
first stage.



As I understand it there is no justification in terms of
econometric theory for suppressing the constant in the FSR if
a constant appears in the structural equation to be
estimated. The specification of FSRs is driven by your desire
to use an estimator which produces consistent estimates of
the structural equation when OLS does not. If you use that IV
estimator, you must play by the rules. The Z matrix contains
all instruments, included and excluded, in the model. If you
include iota in the model, it must appear in the Z matrix,
and will then be used in the IV formula.

Kit Baum   |   Boston College Economics & DIW Berlin   |
http://ideas.repec.org/e/pba1.html
                              An Introduction to Stata Programming
|   http://www.stata-press.com/books/isp.html
   An Introduction to Modern Econometrics Using Stata  |
http://www.stata-press.com/books/imeus.html



*
*   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/

*
*   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   |   What's new   |   Site index