Correlation Sucks - Long Live Causality! Mastering Causal Inference in Data Science
summary
Years of R&D or a seasonal ad campaign require causal, not just predictive, answers: claim is that A/B tests give unbiased effects when feasible, otherwise methods like IPTW (weights estimated via logistic regression) or synthetic controls reconstruct counterfactuals. Concrete pitfalls include naive before/after comparisons, using SHAP or big ML feature importances as causal evidence, and conditioning on mediators. Examples: Christmas ad vs seasonality, logistic-regression weights to balance treatment, synthetic control for a state smoking law.
AI-written summary · about
chapters (9)
- 0:00Introduction to causal inference
- 6:55Potential outcomes framework
- 10:50Directed acyclic graphs explained
- 15:10Common causal pitfalls
- 18:05A/B testing best practices
- 20:14Advanced causal methods overview
- 23:29Inverse probability weighting
- 25:42Synthetic controls method
- 26:48Why causal inference matters
related talks
45mCausality and Statistics, a Love Not So Impossible After All
29m🧠🧹 Causality - Mental Hygiene for Data Science
46mFrom correlation to causation: measuring ad effectiveness at scale
27mUsing Causal thinking to make Media Mix Modeling
18mHow Data Scientists Broke A/B Testing (And How We Can Fix It)
47mIncrementality Bidding and Attribution
21mHow to Get Ahead in Advertising
01h 11mStatistical Limits of Causal Inference