Ivprobit and marginal effects

[Zainab Ali]

I applied IV probit model and then applied marginal effects. The value of coefficient and standard error is same in both estimations however in case of simple probit regression and marginal effects, they are different. Am I doing it right?

[Andrew Musau]

The default is -predict(xb)- for margins after ivprobit. See

help ivprobit_postestimation##margins

If you want to predict the probability of a positive outcome, specify:

margins, dydx(*) predict(pr)

[Zainab Ali]

The default is -predict(xb)- for margins after ivprobit. See

help ivprobit_postestimation##margins

If you want to predict the probability of a positive outcome, specify:

margins, dydx(*) predict(pr)

Thanks Andrew.
I have two more confusions if you can help me out.
1. Should I include fitted value or the instrument variable in my ivprobit command (which I was considering as 2nd step of tsls)?
2. Is there any issue if I just report the values of marginal effects rather than both coefficients and marginal effects in my table?

[Andrew Musau]

1. Should I include fitted value or the instrument variable in my ivprobit command (which I was considering as 2nd step of tsls)?

ivprobit is for a binary outcome with continuous endogenous variables, where the errors are assumed to be normally distributed. See

help ivprobit

ivregress will implement 2SLS, so you don't need to implement the steps separately. In any case, 2SLS applies linear regression in both steps.

2. Is there any issue if I just report the values of marginal effects rather than both coefficients and marginal effects in my table?

Assuming #1 is done correctly, you can report marginal effects alone. There is no issue with that.

[Zainab Ali]

Hey Andrew, I really appreciate your help. My concern is if ivprobit is performed in 1step or 2 steps? As IV is used in 2 stages as per my knowledge so does it mean I should include the fitted value in ivprobit command which I considered the 2nd step? Please enlighten me in this regard. Thanks

[Zainab Ali]

Andrew Musau also I came to know that ivprobit should be applied to continuous endogenous variable. My endogenous variable is age at the time of marriage which is given in complete years like 15, 16, 17,... 25. Should I consider it a continuous variable or discrete? And if discrete how to fix my regression?

[Andrew Musau]

Read the documentation.

help ivprobit

There is the maximum likelihood estimator that does the estimation in a single step and there is Newey's two step estimator implemented via the option -twostep-. Either way, the command does all the steps for you, so there is no manual coding involved. #6: Age can be treated as a continuous variable.