Continuing the thread, Alan Feiveson <[email protected]> posits:
> So suppose y(select)=1 if a customer enters the store and y(probit)=1 if the
> customer buys one of the store's widgets. Then p10 is the (in general,
> nonzero) probability that the customer buys one of the store's widgets
> without entering the store. Are you then suggesting that the heckprob model
> is not appropriate for this situation?
Unless you are willing to change your definition of y(probit) to:
y(probit) = 1, if customer enters store and buys a widget, OR
customer does not enter store but would have bought
a widget had he entered.
then, no, the heckprob model is not appropriate for your situation.
If you do decide to change your definition of y(probit), then the
interpretation of p10 is the percentage of the population who do not
enter the store but would have bought a widget had they.
The very essence of the heckprob model is to model and estimate the
probability of an event whether that event is actually observed. Regardless
of whether you see it, the event will either happen or it won't.
Otherwise, if all you are interested in are things like P(buy a widget after
entering the store), then just fit a probit model on the complete data.
--Bobby
[email protected]
*
* 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/
