All--
I have received a number of suggestions for extensions and/or
improvements to -ivpois- on SSC (-ivpois- is a GMM estimator of
Poisson regression allowing instrumental variables, suitable for any
model with E(y)=exp(Xb) when you want to include cases where y=0 in
the estimation, so you might be tempted to run -ivregress gmm lny x-
or -poisson y x- but want something with traits of both). Many of
these proposals were similar to changes I had made in various versions
of the program on my own system, but the version of -ivpois- on SSC
did not include. Various bugs that no one reported are also fixed.
Thanks to Kit Baum, a new version of -ivpois- is now available that is
much faster and has more options. The -from- option allows specifying
a matrix of initial guesses to feed -optimize-. If no -from- matrix
is specified, an estimate from -poisson- is used, and if the initial
guess fails to properly initialize -optimize- a matrix of zeros is
used (Mark Schaffer reported that -optimize- would fail in certain
situations when begun far from the optimum, so a matrix of zeros is
not guaranteed to produce an estimate). The -exposure- and -offset-
options work as in -poisson- (Henry Schneider requested this feature).
The gradient is now used together with the "d1 evaluatortype" of
Mata's -optimize- routine to speed up the search for the optimum (John
Zedlewski suggested this), and standard errors justified by the
asymptotic properties of GMM estimators are now reported, rather than
using -bootstrap- by default.
The new version runs about 50 times faster, and should issue fewer
errors. Let me know if any other changes are in order.
To install, type
ssc inst ivpois, replace
or if you want to read the help file before installing, type in Stata:
view "http://fmwww.bc.edu/RePEc/bocode/i/ivpois.sthlp";
As before, Stata 10 is required.
*
* 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/
