Method
Doubly robust estimation
Combining outcome and propensity models so one correct model suffices.
11 cases from 7 companies apply Doubly robust estimation. 2 were checked in an online A/B test; 6 report results in production.
Typical tasks: Measure impact without an A/B test, Experiment when users affect each other, Evaluate models, Get experiment results faster and 7 more. Companies: Lyft, Netflix, Sber, Instacart, LinkedIn and 2 more. Often combined with: Inverse propensity weighting, Instrumental variables, Bootstrap, Inverse propensity scoring, Multi-armed bandit and 27 more.
