Hi all,

I am trying to use a probit model to analyze a bunch of interaction effects for a binary dependent variable in a probit model. My dependent variable is callback, taking the value of 1 if a subject receives a job interview, 0 if not. At the moment, I have 4 independent variables, three of which are binary: black (1 if black, 0 if not), woman (1 if woman, 0 if not), parent (1 if parent, 0 if not), and occupation (which can take 6 values).

As is widely known, marginal effects for interaction terms can't be calculated. For this reason, I am exploring other ways of making group comparisons.

As an example, I want to know what the marginal effect of being a woman is on being black with respect to the probability of receiving a callback. I.e. I want to compare black men with black women.

To do this, I generated the following code:

Code:
margins black, dydx(woman)
I have three questions:

1. is this code appropriate for the described purpose?

2. In the following table, how should I interpret the base outcome? Should I read that the effect of being a woman raises the probability of a callback with .0098 for black subjects and .0104 for white subjects?

Array

3. should I include interaction terms in my initial probit regression? I have combined the previous line of code with the following two regressions, yielding different results:

Code:
probit callback i.occupation##i.black##i.woman##i.parent
Code:
probit callback i.occupation i.black i.parent i.woman


Thanks for your time!