Heterogeneity using interaction, with ivprobit

[Marry Lee]

Dear all,
I was running the following model, where I wanted to test for any heterogeneity by gender:

ivprobit Y (c.X#i.child_male = c.Z#i.child_male),  cluster(prov_birth_child_code) first

where Y is the binary DV, X is an endogenous variable and Z is an IV.
When trying to run this I got the following error:

could not find initial values

However, when I run the following models, I have no errors:

ivprobit Y (X= Z) if child_male ==1,  cluster(prov_birth_child_code) first
ivprobit Y (X= Z) if child_male ==0,  cluster(prov_birth_child_code) first

Am I doing something wrong with the first model, using the interactions?
I appreciate your help.

[Marry Lee]

So after some thought, I solved the problem with cmp as follows:

cmp ( Y = c.X#ib1.child_male   , iia) (X= c.Z#ib1.child_male  , iia), vce(cluster prov_birth_child_code) ind($cmp_probit $cmp_cont) nolr nonrtolerance

Then, for marginal effects, I have the same problem as above, using

margins child_male, dydx(X) noestimcheck predict(equation(Y) pr)

or using

margins if child_male ==1, dydx(X) noestimcheck predict(equation(Y) pr)
margins if child_male ==0, dydx(X) noestimcheck predict(equation(Y) pr)

These two codes for marginal effects give different results.
Even though it is following a cmp where the whole sample size is used, in the second code of marginal effects subsamples are used.

Do you please have any thoughts on this?


I am also wondering if this code:

margins child_male, dydx(X) noestimcheck predict(equation(Y) pr)

is giving me the marginal effects for X for each value of the male variable, or the marginal effect for the male variable for a one unit change in X, would these two be different?

Thank you.