Check James Hardin's paper on how a simultaneous variance estimator
can be constructed:
http://www.stata-journal.com/article.html?article=st0018. I imagine
you could extended in Newey-West fashion, although you'd face a lot of
specialized coding.
On Wed, Oct 21, 2009 at 11:09 AM, margherita Comola
<[email protected]> wrote:
Dear Statalist,
I am estimating an probit model where the observations are by
construction non-independent. In order to get the corrected standard
errors, I apply a version of the Newey and West (1987) robust
covariance matrix [specifically, the version proposed by Fafchamps and
Gubert (2007) in the context of dyadic observation: I correlate the
error of the observation ij with all the other observations which
include either i or j: ik, jk, ki, kj].
Aside, I also need to instrument a (continuous) regressor. For
practical convenience, I proceed with a two-step estimation adding to
the probit model the fitted values of the first step regression. This
boils down to a generated regressor problem, that I know it is usually
corrected with Murphy-Topel technique.
Now my question is the following:
How can I incorporate the Murphy-Topel (1985) correction into a Newey
and West type VCV matrix? are the two correction compatibles? how can
I do it, in practice?
Thanks alot,
Margherita
--
Margherita Comola
Paris School of Economics
Email: [email protected]
http://www.pse.ens.fr/comola/index.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/
I believe that Hardin's paper presents an alternative to the Murphy-Topel correction. The paper by Hole in the link below is what you are looking for, albeit you would have to purchase it.
http://www.stata-journal.com/article.html?article=st0114
