search

vdorie / bartCause

Causal Inference using Bayesian Additive Regression Trees
71 stars 10 forks source link

Continuous treatment? #4

Open ariakhademi opened 4 years ago

ariakhademi commented 4 years ago

I'm exploring the function bartc of the bartCause package and am interested in estimating the effects of continuous treatments. I fed in some data to the function with continuous treatment and got some number as ATE, but didn't find enough info about the continuous treatment case in the documentation. So, I just wanted to make sure, if I feed continuous treatments to bartc, will I get a valid ATE, or is bartc limited to binary treatments?

Thanks a lot! Aria

vdorie commented 4 years ago

Continuous treatments aren't supported at the time. This is largely an interface issue, since (in my understanding) the ATE for continuous treatments would require being specific about what levels of treatment are being compared. The package should probably emit a warning in when the treatment is not binary.

bartCause is basically just a fancy wrapper around the bart function, so if there's a specific estimand you're looking to extract I can code something up.