How to obtain partial derivatives of marginal effects (estat mfx) wrt model parameters?

[Wilson Law]

I am using estat mfx to compute marginal effects after asmprobit. After that I need to compute the following partials:

dP/dx = (dP1/dx)s1 + (dP2/dx)s2 + (dP3/dx)s3

and I need to compute the standard error of dP/dx. Stata computes dP1/dx, dP2/dx, and dP3/dx for me. s1, s2, and s3 are constants. It’s difficult to manually calculate them because asmprobit is simulation-based and slow. Program the algebraic solution to use predictnl might be feasible but really messy. Since Stata uses delta method to compute standard errors, somewhere in the code, the partial derivatives of dP1/dx must be available, right? Margins returns the Jacobian matrix. Can we get the same matrix from "estat mfx"? If so, I can compute my own standard error by computing (d^2P/dBdX). Or, is there another way to accomplish this?

Thank you so much!

Wilson

[Phil Bromiley]

I'm not sure it makes sense to talk about dP/dx - when Stata doesn't give you something, it often means that something doesn't make sense. With non-ordinal discrete outcomes, I don't see what dP/dx could mean.