Is there any command to implement LATE (local average treatment effect)?

[Kyu Kim]

Recently, I am working on the panel data, NLSY.

The problem is that I can not find command to implement LATE estimator in the panel setting, even not in a cross-sectional data setting.
I know that STATA provide some command implementing ATE or else. I cannot find any document, command, or program to estimate LATE, although I have spent hours..

Is there any command to implement LATE???
If there is, please help me out.

Thank you for help in advance.

[Jorge Eduardo Perez Perez]

I think you're confused. LATE is a definition, not a method. A local average treatment effect is what you estimate when you estimate a treatment effect through an instrumental variable method. The treatment effect estimate you get is local, because it only applies to the subset of individuals who are encouraged to take the treatment because of variation in the instrument.

[Kyu Kim]

Thanks for reply, sir.

As far as I know of, LATE applies to the problem when randomized control trials broke down since there are some compliers and defiers.
In this situation, how can I modify usual instrumental variable regression?

I have variables such as ln_wage, tenure, tenure^2, X which is variable of interest, treatment that is instrumental variable indicating whether a person is assigned to which group (dummy variables) to control endogeneity of X, and compliance indicating whether a person is complier or defier.

Microeconometrics by Cameron suggests regression such as:

1st stage: reg compliance treatment

reduced form: reg ln_wage tenure, tenure^2 compliance_hat

I am quite confused because there is not the variable of my interest.
I am not sure whether I understand the method properly or not. Is it all right?

[Mark Schaffer]

Another way to make Jorge's point is that it's about how you interpret the results of an IV estimation. In the usual IV model, there is a homogenous treatment effect - beta is the same for everybody. Now say you change the model so that the treatment effect is heterogeneous - everybody in the sample has their own beta_i - but you use still the IV estimator to do your estimation. The estimator/method/coefficient estimates reported by Stata/etc.are all the same as before, but the interpretation/model is now LATE.