[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Re: predicting with probit
I don't think this is a sensible strategy. The subsamples defined by
core/non-core are incidentally truncated based on a number of
factors. The error term in the probit equation for core/non-core is
doubtless correlated with the error term in the condom-use equation.
That correlation will render estimates based on the subsamples
inconsistent. It would seem that an alternative way to model this
would be to write down your condom use regression in the context of a
-heckman- selection model.
Kit Baum, Boston College Economics
On May 19, 2006, at 2:33 AM, Scott wrote:
I have individual level data observed over three years. An
individual is considered a part of a "core" group if they had 5 or
more sex partners in the last twelve months. I am then going to see
how geographic conditions affects core and non-core members' condom
usage, but arguably condom usage and sex partners are jointly
determined. So, instead, I have been advised to estimate a reduced-
form probit of "core" onto various exogenous determinants, and then
based on that value of "core-hat" sort my sample into predicted in/
out categories, followed by my reduced form condom-use regressions
onto each subsample. Is there a post-estimation command that will
allow me to do this?
* For searches and help try: