warning: derivative missing; try rescaling variable

[Jian_Zhang]

Hi all,

I am running a bivariate probit model and trying to get the marginal effect. I use the code: mfx compute, predict(pmarg1) .However, Stata issues the warning message: derivative missing; try rescaling variable. I already check the Stata FAQ to get to know how to rescale the problematic variable. However, my problem is my variable is a dummy variable. I do not know how to rescale a dummy variable.

Any suggestion is welcomed! Thank you!


Best,
Jian

[Maarten Buis]

You normally don't rescale an indicator variable. I would suspect the problem is somewhere else. For example, what is the correlation between the two error terms?

[Jian_Zhang]

Hi Maarten,

Thanks for your reply. Yes, you are right. I find that there is a problem somewhere else. I have two questions.

1) If I dont use the biprobit but write my own code by using the ml model, is there any way for me to calculate the marginal effect after my ml maximize?

2) I am actually estimating a bivariate probit model with partial observability. The "biprobit (eq1)(eq2), partial" always can not achieve convergence. However, my ml maximize can generate a converged result. I suppose that the ml maximize and biprobit with partial should generate the same result. However, it is not. Do you know the reason for the inconsistency?

Thanks!


Best,
Jian

[Maarten Buis]

That is impossible to say with the limited information you have given us.