Average treatment effect for lm forest

[fresskylin]

I am now using the lm_forest() function to estimate multiple continuous treatment effects. I am interested in estimating the average treatment effect using this function. Is there any way to help compute the average treatment effect for lm_forest() based on the grf package?

[erikcs]

Hi @fresskylin, not currently, it would depend on exactly how you are using lm_forest. In the future we could consider making a tutorial that covers some estimands you could target with lm_forest and how to construct doubly robust averages of these.

[JFLmkt]

Hi @fresskylin, not currently, it would depend on exactly how you are using lm_forest. In the future we could consider making a tutorial that covers some estimands you could target with lm_forest and how to construct doubly robust averages of these.

Hi @erikcs, we are using a similar approach that needs average partial effect estimates from lm_forest. Let's say it is an experiment with multiple continuous/binary treatments. I think this is also linked to #1330

We understand that it takes some time to implement the function in the package, but before that would you mind sharing some math regarding doubly robust scores construction so that we can write customized codes for our own? My sense is that they should be close to what you did for continuous treatments in Causal Forest, and many required information has been shown in the output of lm_forest.

Many thanks for your help!

[erikcs]

Hi @JFLmkt, Graham & Pinto (2022) consider doubly robust constructions for many average partial effects, and the answer should be in there somewhere.