Causal Inference Framework for incrementality
summary
Estimates the incremental effect of app installs on one‑year booking value using propensity‑score matching, inverse‑propensity weighting, doubly robust / double machine learning and an instrumental variable (get the app banners randomized exposure on M.dot) design; overlaps get trimmed where no controls exist; refutation tests and sensitivity analyses report robustness to substantial unmeasured confounding; raw Booking.com numbers stay masked but the pipeline and limitations (LATE vs ATE, extreme weights, limited generalizability) are explicit.
AI-written summary · about
chapters (20)
- 0:00Causal inference framework
- 1:08App install impact for marketing
- 3:37Observational causal inference challenges
- 5:04Unbiased estimates assumptions
- 6:32Variables for causal analysis
- 7:58Propensity scores for matching
- 9:48Propensity model validation
- 11:11Overlap in propensity scores
- 11:33Alternative approaches overview
- 13:15Nearest neighbor matching
- 13:55Regression with PS adjustment
- 15:19Inverse propensity weighting
- 16:21Doubly robust estimators
- 17:03Double ML approach
- 18:25Instrumental variable method
- 20:11Statistical vs design approaches
- 21:36Sensitivity analysis for confounders
- 22:40Refutation tests for robustness
- 23:45Key takeaways
- 25:12Q&A
related talks
30mWinning (Some) Churners Back
35mGeo Experiments and Causal Impact in Incrementality Testing
39mCausal Effect Estimation in Practice: Lessons Learned from E-commerce & Banking
41mOn the Path to Causal Inference
46mRNNs for data-driven multi-touch attribution at Booking.com
41mKeynote: Personalization, Bandits, and Causal inference
31mPython and the Holy Grail of Causal Inference
32mCausal inference and scenario generation within Just Eat Takeaway.com