Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Instrumental Variable Method and Probit

From   Roger Harbord <[email protected]>
To   [email protected]
Subject   Re: st: Instrumental Variable Method and Probit
Date   Thu, 20 Jan 2005 17:49:08 -0000

--On 20 January 2005 08:56 -0500 April Knill <[email protected]> wrote:

I have a cross-sectional time-series sample from which I am trying to
discern (generically) the following:

P(y=1) = a + Bx + Cy + e

B is endogenous so I have performed manually a 2SLS regressing first B onto
a vector of independent variables, predicting B* and then including this
predicted value in the equation above.  This of course is flawed in that
the standard errors are not adjusted as would be the case if an IV
methodology were used.

I recently became aware of a wonderful command -ivprob- (or divprob),
which allows for an instrumental variable methodology and that will
correct the standard errors.  The problem is that this command does not
allow for clustering.  The data specifically involves firms (which is my
unit) across time in different countries so thus far I have used xtprobit
to account for correlation at the unit level.  To my knowledge there is no
command for xtivprob and no capability to cluster using ivprob.  Any idea
how to get around this?  If there is, does anyone know how to obtain a
Sargan's Test overidentification test statistic for this?  -Overid- does
not work unfortunately.  Can it be done manually perhaps?
Allowing for clustering should be possible with -qvf- (see -findit qvf-).
This does IV for generalized linear models (including probit) and has a cluster() option.

No idea about how you'd do an overidentification test with/after it though. Maybe someone else here has? Or try the related articles in Stata Journal 3(4), available at


* For searches and help try:

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