Discover ways to make the most of DML for estimating particular person degree remedy results to allow data-driven focusing on
This text is the 2nd in a 2 half collection on simplifying and democratizing Double Machine Studying. Within the 1st part, we coated the basics of Double Machine Studying, together with two primary causal inference purposes. Now, in pt. 2, we’ll lengthen this information to show our causal inference drawback right into a prediction process, whereby we predict particular person degree remedy results to help in choice making and data-driven focusing on.
Double Machine Studying, as we realized in part 1 of this series, is a extremely versatile partially-linear causal inference methodology for estimating the typical remedy impact (ATE) of a remedy. Particularly, it may be utilized to mannequin extremely non-linear confounding relationships in observational information and/or to scale back the variation in our key consequence in experimental settings. Estimating the ATE is especially helpful in understanding the typical impression of a particular remedy, which could be extraordinarily helpful for future choice making. Nonetheless, extrapolating this remedy impact assumes a level homogeneity within the impact; that’s, whatever the inhabitants we roll remedy out to, we anticipate the impact to be much like the ATE. What if we’re restricted within the variety of people who we will goal for future rollout and thus need to perceive amongst which subpopulations the remedy was best to drive extremely efficient rollout?
This challenge described above issues estimating remedy impact heterogeneity. That’s, how does our remedy impact impression completely different subsets of the inhabitants? Fortunately for us, DML offers a robust framework to do precisely this. Particularly, we will make use of DML to estimate the Conditional Common Therapy Impact (CATE). First, let’s revisit our definition of the ATE:
Now with the CATE, we estimate the ATE conditional on a set of values for our covariates, X: