causal-inferencelisted

# Causal Inference ## When to use this skill Use when the question is "**what is the causal effect of X on Y**" but a clean A/B test isn't possible. Triggers: - "Estimate the impact of <feature/policy/launch> without an A/B test" - "We can't randomize. How do we measure effect?" - "DiD analysis" - "Propensity score matching" - "Instrumental variable" - "Did the launch of X cause Y to change?" If a clean A/B test IS possible, use `ab-test-design` + `ab-test-analysis` — they always beat quasi-experimental methods. If the question is "predict Y" not "what causes Y," use `linear-regression` or `logistic-regression`. ## Decision framework: which method? Answer these questions in order: ``` 1. Is there a clear before/after time and a comparison group not affected by the change? YES → Difference-in-Differences (DiD) NO → continue 2. Did some users opt into a feature/treatment based on observable characteristics? YES → Propensity Score Matching (PSM) or weighting NO → continue 3. Is there a "lever" that affects treatment but doesn't directly affect outcome? YES → Instrumental Variable (IV) NO → none of these will work cleanly; recommend RCT or accept correlational evidence with explicit caveats. ``` ## Required inputs | Input | Why it matters | |---|---| | Treatment definition | Who/what gets the intervention | | Outcome definition | The metric you want to estimate the effect on | | Pre/post time periods (DiD) | When did the change occur? | | Comparis