Statalist


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

RE: st: difference between cdsimeq, ivprobit and ivreg2


From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   RE: st: difference between cdsimeq, ivprobit and ivreg2
Date   Tue, 7 Jul 2009 00:13:02 +0100

Anne-Sophie,

> -----Original Message-----
> From: Anne-Sophie Bergerès [mailto:[email protected]] 
> Sent: 06 July 2009 23:46
> To: [email protected]
> Cc: Schaffer, Mark E
> Subject: RE: st: difference between cdsimeq, ivprobit and ivreg2
> 
> 
> 
> Thank you for your answer mark.
>  
> With ivreg2, i am referring to the regression of my 
> continuous variable on my binary endogeneous variable.(vs 
> ivporbit, regression of binary variale on continous 
> endogeneous variable)
>  
> Therefore, it should do a probit in the first stage (my 
> binary endogeneous variable in all exogenous), save the 
> fitted probabilities ans use that as instrument in the ols 
> regression.

"Should" isn't the right word.  -ivreg2- implements single-equation linear IV.  It doesn't do what you want, but that doesn't mean it should!

> Ivreg2 does not allow me to do a probit in the first stage 
> but an ols. thats why my results are different.( i run an ols 
> on my binary and i had the same results as the first stage of 
> ivreg2). I saw in a previous post that i could use treatreg 
> but stata does not allow me to that because some of my 
> exogeneous variable are dichotomeous.
>  
> Is there a command which allow me to do a probit in the first 
> stage and an ols in the second ? How can i implement a sargan 
> s test then ? Or can i rely on the results of overid after my ivreg2 ?

It sounds like you would ideally like an overidentification test for the system estimated by -cdsimeq-, testing all the identifying restrictions used in the 2 equations.

Maybe you can do that by hand, but I'm not aware of any user-written commands that will do it.

The second-best alternative is overid tests for the two equations separately.

-overid-, downloadable from ssc-ideas in the usual way, will report an overidentification statistic after an -ivprobit- estimation.  This gives you an overid test for dependent variable D and continuous endogenous regressor C.

The equation with the continuous variable C as the dependent variable, and D as the endogenous regressor, is trickier.  You can estimate this with -ivreg2-, as you suggest, and it will give you a Sargan-Hansen statistic for this estimation.  But the test applies to the -ivreg2- estimation, and not to the what -cdsimeq- does.  My understanding is that -cdsimeq- uses an additional identifying restriction, namely the nonlinearity (dichotomous nature of D) in the first stage.  Maybe you can program an overid test statistic that uses this additional identifying restriction.  But if you want to refer to the Sargan-Hansen test that -ivreg2- reports,  I don't see alternative to saying "well, estimating using straight IV gives me a low overid test statistic, so imposing the additional identifying restriction of a dichotomous D ought to be OK" - which is just handwaving.

I hope this is of some help at least.

Best wishes,
Mark

>  
> I thank you in advance for your answer.
> 
> 
> Anne-Sophie 
> 
> 
> 
> 
> 
> ----------------------------------------
> > From: [email protected]
> > To: [email protected]; [email protected]
> > Subject: st: difference between cdsimeq, ivprobit and ivreg2
> > Date: Mon, 6 Jul 2009 21:14:35 +0200
> >
> >
> > I think, I have problems with posting.....
> >
> > Good afternoon,
> >
> > I am using the cdsimeq command. In order to test my 
> endogeneity and overidentification conditions, i use ivprobit 
> and ivreg2. However, I have a problem. The results from 
> cdsimeq and the ivprobit/ivreg2 are really different and i do 
> not understand why.
> >
> > The first stage of cdsimeq and ivprobit are the same but 
> not the second stage.
> > The first stage of cdsimeq and ivreg2 are different and the 
> second stage as well.
> >
> > I dont know if i have made a mistake in my syntax or if it 
> is because different estimator are used.
> >
> > Syntax :
> >
> > cdsimeq (C x1 x2 x3 x4 x5 x6) (D x1 x2 x3 x4 x7 x8 x9)
> >
> > ivprobit D x1 x2 x3 x4 x7 x8 x9 (C = x5 x6), twostep
> >
> > ivreg2 C x1 x2 x3 x4 x5 x6 (D = x7 x8 x9)
> >
> >
> > Does anyone can help me with this issue ?
> >
> > I thank you in advance,
> >
> > Anne-Sophie
> > _________________________________________________________________
> > Téléphonez gratuitement à tous vos proches avec Windows 
> Live Messenger ! Téléchargez-le maintenant !
> > http://www.windowslive.fr/messenger/1.asp
> > *
> > * 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/
> _________________________________________________________________
> Inédit ! Des Emoticônes Déjantées! Installez les dans votre 
> Messenger !
> http://www.ilovemessenger.fr/Emoticones/EmoticonesDejantees.aspx
> 


-- 
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/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index