Calcaulating ATT and CATT from causal survival forest!

[fmhr12]

Hello, is there any way to calculate ATT and CATT from the causal survival forest? In causal forest, the average_treatment_effect function accepts the argument "treated" for the target.sample, but it is not available for the causal survival forest. Is averaging CATEs over the treated subgroup appropriate, or are there more direct or robust methods available?

Also, is there a way to calculate BLP and TOC if the target sample is treated individuals?

[swager]

You should be able to run average_treatment_effect, rank_average_treatment_effect, etc., on a trained causal survival forest also. This will give you estimates of the ATT, etc.

[fmhr12]

You should be able to run average_treatment_effect, rank_average_treatment_effect, etc., on a trained causal survival forest also. This will give you estimates of the ATT, etc.

Thanks for your reply. When I try something like this (att_res <- average_treatment_effect(cs_full, target.sample = "treated") on the trained causal survival forest, I get this error:

"Error in average_treatment_effect(cs_full, target.sample = "treated") :
For any forest type other than causal_forest, the only implemented option is method=AIPW and target.sample=all"

[swager]

Good point, thanks for highlighting this. The reason we added this filter is that considering ATT for some forest types (e.g., IV forests) raises extra challenges.

But allowing ATT (and other estimands) for causal survival forests shouldn't be an issue; we should be able to add this flexibility. @erikcs -- do you see any potential issues with this?

[erikcs]

Yes, this is something we should add in a future release.

In the meantime, if you'd like to compute an ATT, you should be able to do so using the outputs from a trained causal survival forest, following the same logic as in the average treatment effect function.

[fmhr12]

Thanks a lot for your response and for providing the resources.