IVPROBIT ISSUE- Different directions of the coefficients commands IVPROBIT and Margins

[CRISTINA CRUPI]

Dear all,

I am having an issue with my dissertation; my STATA version is 14.2.

When I run the IVPROBIT command I get a positive partial effect of the linear index for my variable of interest, sugPane_serving.
Then, when I estimate the average partial effect i get a negative one.

As per my knowledge, the direction should be the same, since the density function by which is multiplied the coefficient of my regressor (sugPane_serving) is always positive.

Could you please give me some insight on this point?
How can I justify this situation? or better can I fix it somehow?

Thanks a lot in advance,

best regards,

Cristina

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Screen shots and attached files are discouraged. For example, your screen shot is extremely hard to read.

I'm not sure, but generally dydx for an individual variable in such models depends on the values of all the variables - the dydx is done holding all other other variables at their means. A sign switch seems very odd however. This might be causing your problem. You might look at dydx(sugPane_serving) at different values of the other variables.

[John Mullahy]

Christina: I suggest that you take a look at the full documentation for margins after ivprobit. The computation of the marginal effect for the predicted outcome is not just the "standard" density times coefficient formula, but rather is based on a computation that accommodates the endogenous nature of the RHS variable. The description is on p. 1177 of the pdf documentation for v.15.1 (I'm not sure about other versions).