Dear Mark and Stata-listers,
Thank you very much for your advice Mark, and for the reference.
Could you help me with another query ? I would like to estimate a model using a panel household data set over two years. My left hand side variable is a proportion, so it is both censored at 0 and 1. As I want to control for household heterogeneity, and given that the unobserved time-invariant effects are likely to be correlated with the regressors, ideally, I would like to estimate a tobit fixed effects model. I know that there is no command in Stata which would allow for it, Stata has only Tobit random effects. In a previous email, I was recommended the CLAD function.
Is this the same as the Honore (1992) trimmed least absolute deviations that is reported in Baltagi 2001, 2nd edition, p. 212 ? How can I I check for the violations of the different assumptions underlying this estimation ?
If I use xttobit (RE), again, how can I check for the violations of the different assumptions underlying the model (see Wooldridge, 2002, eco analysis of cross section and panel data, p. 541) ?
Your guidance will be much appreciated. Thank you very much in advance.
Thi Minh
------------------------------------------------------------
Date: Tue, 25 Feb 2003 18:09:32 -0000
From: "Mark Schaffer" <[email protected]>
Subject: Re: st: 2SLS with a censored endogenous variable
You can use single-equation IV and it's consistent - no problem.
It's not hard to prove it or see that it's true because the standard
proof of consistency of IV doesn't depend on the functional form of
the equations your not estimating. ... But if you want a reference, I
think there's a paragraph about it in J. Angrist and A. Krueger,
"Instrumental Variables and the Search for Identification," Journal
of Economic Perspectives, Fall 2001. Their reference is to an
endogenous variable that is a 1/0 outcome, but it's the same point.
Hope this helps.
- --Mark
Subject: st: 2SLS with a censored endogenous variable
Date sent: Tue, 25 Feb 2003 13:33:30 -0000
From: "Ngo,PT (pgr)" <[email protected]>
To: <[email protected]>
Send reply to: [email protected]
> Dear Stata-listers,
>
> Could someone give me some guidance on the following estimation issue that I am having ?
>
> I would like to use 2SLS in order to control for the endogeneity
> of one variable using ivreg (or xtivreg since I am estimating a
> first-difference equation, using a two-year household panel data
> set). The credit variable (total amount of formal credit) that I
> am instrumenting for is censored at 0 (about 50% of my households
> do not have credit). Given that the data is censored, the first
> stage would be a tobit. Is it still appropriate to use a linear
> estimaton for the first stage ? What is the impact on the
> efficiency of the 2SLS/IV estimator and on the inference that I
> can make ?
>
> I have looked into a number of books (Greene, Wooldrige 2001 and
> 2002, Davidson MacKinnon) but I could not find a reference to this
> problem. Would someone know of a paper or a reference which
> addressed this estimation issue and could help me make up my mind
> as to what would be the best estimation strategy ?
>
> Your guidance would be much appreciated,
>
> Thanks a lot in advance,
>
> Thi Minh Ngo
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
