st: Marginal effects after censored regression model

["Andreas Drichoutis" <adrihout@cc.uoi.gr>]

Dear all,

I'm comparing marginal effects from a censored regression model between
Stata and Limdep. Limdep reports ∂ E[y | X]/ ∂ X as marginal effects. It
is funny that Limdep's estimates coincide with "mfx,
predict(ystar(lower,upper))" which is ∂ E(y*|a<y<b) / ∂ X. However, these
conditional means are not the same since:
E[y|x]=Lower*Prob(<=lower) + Upper*Prob(>=upper)+
[Prob(<=upper)-Prob(<=lower)]*E[y*|lower<=y<=upper]
Therefore, the conditional mean derivatives should not coincide. Stata
should report the last term of the above expression which is different from
the left hand side. I've replicated the estimate of Stata "by hand" using
nlcom and it actually reports ∂ E[y | X]/ ∂ X while it says it reports ∂
E(y*|a<y<b) / ∂ X.

Am I missing something here? Where is the catch?

Best,
Andreas Drichoutis



*
*   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/