just came across this package recently, with a fairly simple idea:
After a certain A/B test, the data scientist is often asked about which subset of users is most affected by the new feature. One can actually answer this by building a decision tree (splitting the dataset) to maximize the treatment effect within each node.
In a way, this is kinda like a post-hoc subgroup analysis, but might be more useful. So much for food for thought;-)
gbordyugov
commented
7 years ago
just came across this package recently, with a fairly simple idea:
After a certain A/B test, the data scientist is often asked about which subset of users is most affected by the new feature. One can actually answer this by building a decision tree (splitting the dataset) to maximize the treatment effect within each node.
In a way, this is kinda like a post-hoc subgroup analysis, but might be more useful. So much for food for thought;-)