EasyDeepLearn

Why is peeking at running A/B tests dangerous?

hard

Answer

  • Peeking = repeatedly running a test and stopping when p < 0.05 → hugely inflates false-positive rate (up to ~40% for one-sided at α=0.05).
  • Fixes: (1) commit to sample size ex ante and don't peek.
  • (2) Sequential testing: mSPRT, α-spending (Pocock, O'Brien-Fleming), Bayesian sequential decisions.
  • (3) Always-valid inference (Howard & Ramdas): give up 10-20% efficiency for the freedom to check any time.
  • Optimizely / Statsig / Eppo implement always-valid or sequential correctly.
Check yourself — multiple choice
  • Safe to peek
  • Peeking inflates FPR up to ~40%; fixes: pre-committed n, sequential (mSPRT / α-spending), always-valid inference
  • Only for regression
  • Same as multiple testing

Peeking inflates FPR; use pre-committed n, sequential, or always-valid inference.

#hypothesis-testing#ab-testing

Practise Statistics Fundamentals

215 interview questions in this topic.

Related questions