st: RE: Heckman with variables that perfectly predict selection

[Nick Cox <n.j.cox@durham.ac.uk>]

If you look at the top of the webpage referred to you will see that it is explained and explicitly dated: 

"ADO-files What's New 
Stata Ado-files - November 1998" 

Unless your version of Stata is really, really old, it does not refer to an innovation as far as you are concerned.
 
Nick 
n.j.cox@durham.ac.uk 

Maria Alva

I recently came across  the following statement in
http://www.stata.com/support/updates/ado/whatsnew.html

"(STB-43) heckman heckman now has the capability to estimate models
with variables that perfectly predict selection. Previously heckman
would simply drop such variables from the selection equation, which is
inappropriate in most cases."

Puzzled, I tried estimating a heckman selection correction model in a
data set where death is observed, and where deaths perfectly predict
non-responses to questionnaires.

heckman y $x, select(selected= death $x) // selected=1 if a person
completes a questionnaire

this gives me a negative single digit and statistically significant
coefficient. If instead I use

probit selected death $x

as expected, death drops out

My question is twofold: what is the innovation in the Stata command
that causes death not to be dropped out of the selection equation as
it perfectly predicts the selection indicator? and most importantly,
what would be a valid instance for which it would be appropriate to
include a variable in the selection equation when this variable
perfectly predicts selection?


*
*   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/