Causal Analysis
Go beyond correlation to causation.
Randomized Experiments
Gold standard: random assignment. Difference in means estimates causal effect.
A/B testing is randomized experiment.
Observational Methods
Matching: match treated to control. Propensity score: probability of treatment. Instrumental variables: uses randomness.
Difference-in-differences: before/after, treatment/control.
Challenges
Unmeasured confounders. Selection bias. SUTVA violations.
Key Takeaways
- Randomization is gold standard
- Matching and propensity score for observational
- Difference-in-differences for panel data