Propensity Score Matching: A Non-experimental Approach to Causal...
summary
Explains a reproducible recipe: build a propensity model (usually logistic regression), score every unit, match exposed to unexposed using a 0.25 SD caliper on the logit, check balance with standardized mean difference <0.1, then estimate lift with regression adjustment, stratification or IPTW. Shows a marketing example (HelloFresh ad → conversions), warns about limited Python packages (causalinference, pymatch) and common‑support sensitivity and omitted‑variable risk.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:03Propensity score matching overview
- 2:50Need for propensity score matching
- 5:38Four steps in matching
- 6:42Calculating propensity score
- 10:15Selecting model variables
- 13:41Building the model
- 14:02Matching algorithms
- 17:57Checking balance
- 20:43Estimating effects
- 24:15Limitations and cautions
- 26:40Resources and tips
- 28:12Q&A
related talks
24mBeyond the P-Value: A Data Scientist’s Guide to Tier-1 Feature Launches
06mCausal Machine Learning – What’s in it for Data Scientists?
37mAttributing real-world outcomes to sponsorships
30mClimbing the causal ladder for fun and profit
39mExperimentation and the gold standard of data champions
34mOptimizing E-commerce: Data Science for Better CX & Sales
45mCustomizable probabilistic record linkage with Name Match
13mOptimizing Subscription Services with Uplift Modeling