is imputation done for average treatment effect (ATE) estimation?

[aczura2024]

Hi,

I have read the missing values section of the grf https://grf-labs.github.io/grf/REFERENCE.html#missing-values and understand how causal forests handle missing values for individual treatment effect estimation (by sending missing observations to the left node) but I am unclear on how the missing values are handled when it comes to ATE estimation (like when using the average_treatment_effect() function in grf). Is this discussed in procedure 3 of Mayer et al (2019)? Please let me know.

[erikcs]

Hi @aczura2024, the average_treatment_effect function computes an ATE using AIPW. When missing values are present, the nuisance components used to form these AIPW scores will by default use forests with MIA splitting.