stats

🟢 Stats: A/B Testing Design Checklist

  1. Define hypothesis: "Changing X will increase metric Y"
  2. Pick metrics: Primary, secondary, guardrail
  3. Calculate sample size: n ≈ 16σ²/δ² (α=0.05, power=0.80, δ = min detectable effect)
  4. Set duration: Minimum 2 weeks (capture weekly patterns)
  5. Randomize: By user (not session) — hash user ID for consistency
  6. Analyze at predetermined end date — don't peek
  7. Check statistical AND practical significance

Practice Questions

Q: Why randomize by user ID, not by session?