Hacker News new | past | comments | ask | show | jobs | submit login
How not to run an A/B test (evanmiller.org)
2 points by arjunnvl on July 11, 2023 | hide | past | favorite | 1 comment



If I understand correctly, the issue is when you track the effect as a rolling value and stop the test as soon as it dips above the significance level.

So for instance, let's say your test is to roll a dice 6 times. You say "if I get 6 100% of the time, then I'll consider that significant". You do the first roll and get a 6. Well that's 100% of the time, so you end the test and claim you had a significant result.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: