st: Average Marginal Effects on Recursive Bivariate Probit

["Damien Jourdain AIT" <djourdain@ait.ac.th>]

Dear all, 

I am new to stata, and I am trying to run a recursive bivariate probit as
follow:

biprobit (qgregistration = labha perceptionimpact education exp1 exp2
govplus knowgap1 gapchannel ///
	own1 own2 afilliations expcostreduction offfarm ricetraining ) ///
	 (ricetraining = labha perceptionimpact exp2 afilliations,
noconstant), vce(robust)

It runs ok, and I am rejecting the H0: rho=0;

I am now  interested in calculating the marginal effects... and run into a
problem

When I run the command (I actually want to have the average marginal effect
of labha on prob(qregistration=1):
margins, predict(pmarg1) dydx(labha own1 own2)

I have the following error message:
prediction is a function of possibly stochastic quantities other than e(b)
r(498);

After reading the docs, I found that it apparently cannot calculate the
standard errors with the data available in e(b)
So when I run the command 
margins, predict(pmarg1) dydx(labha own1 own2) nose

it actually works, but obviously I do not have the se (usually required!)

So I have two questions:
1. Am I really doing things properly to calculate the average marginal
effect of labha on prob(qregistration=1), knowing there is a direct effect
on the first equation and an indirect effect via the second equation... (I
guess the question is whether margins takes care behind the scene of this
aspect!)
2. If I do properly thing, then how can I calculate se by another mean?

Apologies if this questions appears obvious to long time practitionners, but
I am stuck after quite some effort to find equivalent problems on this list.

Thank you for any help.

Damien





*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/