st: two stage ordered probit

[TEDLA HAILE <kebede1963@gmail.com>]

Dear Statalist,

 I am trying to run a simultaneous equation model which is modeled as;



P*=D*+1X1+e

D*= zP*+ 2X2+e2



Where each of the above two equations are ordered probit equations and
P* and D* are ordered latent endogenous variables which will have
ordinal values between 1up to 4.

 The reason the model is formulated in this manner is because the
objective of the study is to determine if the existence of each of
these endogenous variables have cause and effect relationship, like
the existence of P causes D and vice versa. Literature supports
applying 2-stage least square technique for ordered probit models and
thus I am trying to fit the model in this way;

Given the original simultaneous equation,

P*=D*+1X1+e

D*= P*+ 2X2+e2

I will first regress each equation via ordered probit by Stata 11
software and have predicted values for P* and D* i.e  P^ and D^

Next stage I will undertake ordered probit regression by replacing D*
and P* in the right side by predicted values P^ and D^ that is,

 P*=bD^+a1X1+z1

D*= cP^+ a2X2+z2

 Now I would like Statlist community suggest on the following point;

 Which predicted value should I use i.e predicted probabilities
(catagories) or the linear prediction generated by Stata where in my
case two options

 . oprobitP <varlist>

. predict P1 P2 P3 P4

. gen Phat = 1 if P1 > .50

. gen Phat = 2 if P2 > .50

. gen Phat = 3 if P3 > .50

. gen Phat = 4 if P4 > .50



OR



. oprobit P D X

. predict Phat, xb

. olprobit D Dhat X



I really thank you for any assistance which you might be able to provide.



With kind regards,



Kebede



PhD Student

*
*   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/