📰 Summary (use your own words)

• Frequentist approach to statistical tests significance can be prone to large sample size

✍️ Notes

• Don’t fall prey to “repeated significance testing errors”
• When a A/B test says there is a “95% of chance of beating the original” or “90% probability of statistical significance”
  • It is asking the question “what is the chance that we observe a difference like what we see in data randomly?”
  • And that is the significance threshold (0.05 or 0.01)
• However, this is precedent on a key decision -> the sample size was fixed in advance
  • Because might be tempted to take action once we see significant results and if that becomes insignificant later we wouldn’t have known
• So the key is to determine a sample size before the experiement
  • One rule of thumb is $N=16*\frac{\sigma^2}{\delta^2}$ where $\delta$ is the minimum effect you want to observe
• For Bayesian A/B testing, this is less of a problem compared to the frequentist approach. But not completely fool-proof -> Is Bayesian AB testing immune to peeking