Bootstrap after two stage probit

[Abdullah Selim]

Dear all,
I was using ivprobit but it seems there is a problem with the margins command when I want to get average marginal effects. It gives the same value as the coefficient from ivprobit. Also, it takes hours to calculate the AME.
Therefore, I decided to do it manually. In the first stage, I run a linear regression for my instrument, then calculate predict.
In the second stage, I run probit with the predicted values.
I have two questions:
1-) Is this procedure exactly the same with ivprobit command?
2-) I know I need to bootstrap the standard errors for the margins command. But I don't know how to do that. I also have to use sample weights. Please let me know the correct syntax for this.

Could you please advise on me this matter?

[FernandoRios]

Hi Abdullah
Before you go and try to do this manually using Bootstrap, would you share how were you trying to estimate the marginal effects after ivprobit?
In older versions of Stata (I m currently using version 15), you needed to specify that what you needed to estimate were marginal effects on the predicted probability. Perhaps you are omitting that step.
In addition, if you want to try to obtain marginal effects "faster", you can try using the "nose" option. This reports the point estimates only, without trying to obtain the standard error calculations.
HTH
Fernando

[Abdullah Selim]

Hi Fernando, Thank you for your answer. Here is my code: I am using 14.2 MP

For iv probit:

ivprobit Y i.gender#i.education#i.age i.region i.year (endo= instrument) [w= weight], vce(cluster nuts2)
margins, dydx( instrument)

for manual one:

- regress endo instrument i.gender#i.education#i.age i.region i.year [w= weight], vce(cluster region)
- predict ahat
- probit Y ahat i.gender#i.education#i.age i.region i.year [w= weight], vce(cluster nuts2)
- margins, dydx( ahat )

margins with nose after ivprobit does the job but gives the same probability coefficient with ivprobit estimation

What is the missing part?

[FernandoRios]

It is as i suspected. if you look at the helpfiles for ivprobit / postestimation/margins it tells you the default is to obtain marginal effect on the latent index (xb). That is the same as the coefficients (unless interactions are used).
To obtain marginal effects on the probability, you need to use the "pr" option

margins, dydx(*) predict(pr)

HTH

[Abdullah Selim]

I knew this, but I am trying to calculate the AME, not the probability, like in usual logit and probit. Is there a way for that?