I want to estimate a model for the number of years of school completed as a
function of individual, household, community, and school characteristics.
It's in a setting where many kids don't go to school at all, so I only
observe the number of years of schooling for a (non-random) subset of the
age group that ever attends school.
At first I estimated it as a standard -heckman- model, until someone pointed
out to me that because the sample includes both children who are still in
school and those who have already dropped out, the number of years completed
is a censored variable, and that -cnreg- would be more appropriate for the
second stage.
This seems reasonable to me, but my problem is this: I don't know how to do
a selection model with -cnreg-. I haven't tried working through the algebra
yet, but I strongly suspect that one can't just calculate the IMR (or
non-selection hazard) from the selection probit and slap it on the
right-hand side of the -cnreg-, the way one can do when the second stage is
OLS (ie, Heckman's two-step). Joint -ml- estimation would probably work, but
I'm not sure where to start.
Does anyone have any suggestions for this?
Thanks.
-Ken
*
* 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/
