hard20 pts · first answer countsab-testingcausal-inference

Treated drivers take control rides

In a ride-hailing test, treated drivers get a bonus that makes them accept more rides. Treated and control drivers serve the same riders in the same city. Why is a simple driver-level comparison misleading?

A.Driver tests always have too little powerB.The bonus has no effect on control drivers, so the comparison is fineC.Treated drivers take rides that control drivers would have served, so the difference overstates the true effectD.Randomising drivers guarantees an unbiased estimate

← NDCG at 3 for a ranking Benjamini-Hochberg discoveries →