IV Models: CATE or LATE?

[evan-magnusson]

Just wanted to clarify something - for the IV models (such as OrthoIV or ForestDRIV), is the ate() method computing the local average treatment effect for compliers (using the ITT and the compliance rate), or just the ITT?

Thank you!

[yujias424]

I think it might just be taking an average of conditional LATE; check #753. As they mentioned, that ate_ attribute applies a double-robustness correction to the computation of the ATE itself (on the training data); the ate() method allows you to compute the ATE for any population by averaging the computed CATE values for each individual.