Interview Query

Decisions based on Confidence Intervals

peaceful-amber-geckoJune 12, 2022

Here’s a screenshot from a course offered by Google on A/B testing. Basically, these are the scenarios possible for confidence intervals. The dotted lines are minimum effect for launch (we can focus on the right dotted line assuming we’re measuring some engagement metric). And the 3 possibilities are Launch, No Launch, and Additional data points.

My question is: for the very last CI in this screenshot, the CI does not contain 0, and most of it is past the minimum effect threshold, however, part of it is not. The suggestion here is to increase sample size, however this isn’t possible in many cases - whether it’s due to limited resources or other reasons. What would be the recommendation in practice if we see the very last CI?


for some reason clicking this link doesn’t work, so to open just copy and paste the link

16 days ago

Related posts

Data Analytics Live Course -> Starting June 29th
I am going to get interviewed for Google cloud machine learning consultant role can someone help me with peer interview
Have a Google Interview coming up? Practice with these Sample Questions
Airtable Data Science Interview