I'm running an ivprobit model, and I'm trying to get marginal effects out of the second stage coefficient.

When I run the ivprobit model, the variable of interest has a coefficient of 2.62 with a p-value of 0.000.

After I run the ivprobit model, I am running the the following code:
margins, dydx(variable_of_interest) predict(pr) atmeans


It's giving me an output that suggests the variable's marginal effect has negative sign and a p-value of ~0.8. On the surface, that does not seem right to me. Am I writing the correct code to get marginal effects out of an ivprobit model? Any thoughts on what's going on here?

As an aside, I feel really good about the exogeneity of the instrument, and it is a very strong predictor of the variable I'm interested in.

Thanks,
Sam