Dear Statalist users,
I have a model of two equations estimated by cmp command, first equation is ordered probit, the second one is probit. How can I test the validity of my model? I do not have pseudo R2 as the output. I calculated it as Roodman wrote in one of the posts as 1/(1-2*e(ll)/e(chi2)), but it gives me extremly low Pseudo R2, such as 0.0003944. I am suspicous about this value, since the Pseudo R2 for the first equation is 0.1276 and for the second 0.2212. The athanrho is statistically significant and I belive that equations are related.
When I use the other formula PseudoR=1-1/(1+2*(ll1-ll0)/N), I get strange value, extremly high 0.9954901. ll0 is estimated as the model with only constant in both equations.
Also, is it possible to calculate ROC curve after ordered probit, and how can I do it?
I am almost sure that my model is quite good, but I want to show that with some formal criterium/test.
Thank you.
Best,
Aleksandra
I have a model of two equations estimated by cmp command, first equation is ordered probit, the second one is probit. How can I test the validity of my model? I do not have pseudo R2 as the output. I calculated it as Roodman wrote in one of the posts as 1/(1-2*e(ll)/e(chi2)), but it gives me extremly low Pseudo R2, such as 0.0003944. I am suspicous about this value, since the Pseudo R2 for the first equation is 0.1276 and for the second 0.2212. The athanrho is statistically significant and I belive that equations are related.
When I use the other formula PseudoR=1-1/(1+2*(ll1-ll0)/N), I get strange value, extremly high 0.9954901. ll0 is estimated as the model with only constant in both equations.
Also, is it possible to calculate ROC curve after ordered probit, and how can I do it?
I am almost sure that my model is quite good, but I want to show that with some formal criterium/test.
Thank you.
Best,
Aleksandra
