Marginal Effects of a Binary Endogenous Regressor in Ordered Probit Model

[Srajal Nayak]

Dear Members,

First of all, apologies if I fail to follow any general rules of the forum in my post. I am new to the forum and require some urgent help.

I am running an ordered probit regression with a binary endogenous regressor. My dependent variable is a learning outcome of children (call it LEARN, which ranges from 0-4) and the endogenous regressor is whether the household is electrified (ELEC = 0 or 1). Since household electrification is endogenous, I am using IVs (call them IV1, IV2, IV3 and IV4) for that. I am running the "eoprobit" command in Stata15, where the first stage is estimated using a probit model.

So my regression looks something like this:

"eoprobit LEARN ELEC X1 X2... , endogenous (ELEC = IV1 IV2 IV3 IV4 X1 X2...., probit) "

The regression runs fine and I am getting a positive and significant coefficient on the variable ELEC which is what I expected.

Next, I am trying to estimate the marginal effect of ELEC using the following command: "mchange ELEC". I keep getting the error: "Variable IV1 not found in the list of covariates".
I am unable to understand how to solve this problem. Why should I be including the IVs in the list of covariates?

Alternatively, if I run "margins ELEC", then I get margins for each level of outcome learn and ELEC = 0/1. I assume if I take the difference, then I can interpret it as household electrification increases the probability of the outcome LEARN=4 by the amount Pr(LEARN=4 | ELEC=1) - Pr(LEARN=4 | ELEC=0)?

However, I can't understand why do I keep running into trouble with the mchange command. Any help would be highly appreciated!

Thank you.

Regards,
Srajal

[Richard Williams]

mchange is a user-written command, part of Long and Feese's spost13 package. If no one else answers you, you may want to contact the authors. mchange is a shell for margins though, so whatever you want to do with mchange you could probably do with margins (if it can be done).

Also, a reproducible example might improve your chances of getting an answer. See the FAQ, especially point #12, which talks about dataex and code tags.

[Srajal Nayak]

Dear Richard,

Thank you for your suggestions. Yes, I was surprised as well that I am able to get the desired result with margins but not with mchange. I feel it could have something to do with mchange working slightly differently with the ERM models. I think I will contact the authors. Also, many thanks for directing me to the FAQ. I will have a look at the dataex and code tags as you recommended.

Best,
Srajal