[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Questions on ivprobit (probit model with an endogenous regressor)

From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   st: RE: Questions on ivprobit (probit model with an endogenous regressor)
Date   Sun, 6 Jan 2008 23:22:42 -0000


> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of 
> [email protected]
> Sent: 06 January 2008 21:45
> To: [email protected]
> Subject: st: Questions on ivprobit (probit model with an 
> endogenous regressor)
> Dear colleagues:
> I need your help with respect to the following questions 
> about ivprobit (command for probit model with an endogenous 
> regressor):
> 1. Is there anything with my syntax below? If yes, how can it 
> be corrected?
> ivprobit vrics mhol acmt iacm bdze mdir dual ihol aud_1 aud_2 
> aud_3 aud_4 size perf levg  ind_1 ind_2 ind_3 ind_4 ind_5 
> ind_6 ind_7 ind_8  rev_gwth  fcash optns (bdin= dual ihol 
> mhol fcash size levg ind_1 ind_2 ind_3 ind_4 ind_5 ind_6 
> ind_7 ind_8 optns vrics_1), first

It's impossible to tell without seeing the actual call to -ivprobit- and
what Stata makes of it.  You should post this.

> 2. The dependent variable, vrics, is a dummy coded 1/0. 
> Hence, my use of ivprobit. However, the endogenous regressor, 
> bdin, is not. Isn't ivprobit reading the data on bdin as 
> dummy?

No.  It's because you're using the default ML estimator.  If you use the
two-step estimator, you'll see that the first-step estimates for bdin
are exactly the same estimates you get if you use -regress-.

> My question is based on the fact that OLS estimates of 
> the bdin equation is different from those returned by 
> ivprobit for the first-stage regression. Also, Stata 10 
> reports iteration for "Fitting exogenous probit model".
> 3. Stata increases the number of the instrumented variables 
> in its results than I originally specified in the syntax? Why 
> that? Is it because my model is under-identified?

Again, it's impossible to tell unless you show us the call to -ivprobit-
and the results.

> 4. For the fitting of the full model (i.e., probit model with 
> the endogenous regressor), Stata goes through iteration from 
> 1 to 1070, reporting that the intervening iterations (i.e., 1 
> to 1068) are not concave. Why this long iteration? Does it 
> suggest that the model is mis-specified? Or implies that the 
> results are not correct?

You're probably asking a lot of the data, maybe too much.  Perhaps you
have some multicollinearity problems.  Are many of the coefficients

> 5. Stata 10 does not report model summary statistics for the 
> first-stage regression with bdin as the dependent variable. 
> Is there any way of getting these statistics?

With the ML estimator, the "first-stage regression" isn't really a first
stage, since it's estimated simultaneously with the main equation.  I
think this means that you just have to get the stats you want from the
main -ivprobit- results with the ML estimator.  You could switch to the
two-step estimator so that the first-stage results are reproducible with
a simple call to -regress-, but this doesn't seem like a good reason to
do this.

Hope this helps.


Prof. Mark Schaffer
Director, CERT
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS
tel +44-131-451-3494 / fax +44-131-451-3296
email: [email protected]

> I look forward to hearing from you. Thanks for your cooperation.
> Regards,
> Stephen
> --
> Stephen Owusu-Ansah, PhD, CIA, CBM
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

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