st: predictnl after mprobit estimation

[Gerald Manthalu <gmanthalu@hotmail.co.uk>]

Hi Statalist,
 
  
I am estimating a multinomial probit “difference-in-differences” model as follows:
 
 
 mprobit y x1 x2 x1*x2 x3
 
  
y is the dependent variable and has five outcomes, x1 is a time dummy (0 before, 1 after), x2 is a treatment group dummy (1 treated, 0 non-treated), x1*x2 is an interaction and x3 a vector of other regressors. I am assuming that the treatment affects the probabilities of the five outcomes in y for an individual. I then want to use predictnl to predict the treatment effect along the lines of 

norm(beta1+beta2+beta12+beta3*x3)- norm(beta1+beta2+beta3*x3) in the case of the probit model (Puhani, 2012).
 
  
The problem I have is to derive the cross differences of the mprobit function with respect to x1 and x2 to use with predictnl as in the formula above. I wonder if someone can help with this or give other ideas.
 
 
 Thanks,
 
  
Gerald Manthalu
  		 	   		  
*
*   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/