hard20 pts · first answer countsab-testingcausal-inference
Adjusting for a variable on the causal path
A test shows a new recommendation widget raises revenue. An analyst adds 'number of items viewed' as a control and the effect vanishes. The widget works by getting users to view more items. What went wrong?
Discussion 0
Join free to post in the discussion.
Related questions
- Unit of randomisation for a new menuData Scientist
- Why run an A/A testData Scientist
- What MDE meansData Scientist
- The counterfactualData Scientist
- Why random assignment worksData Scientist
- Same user, same variantData Scientist
- When a bandit beats an A/B testData Scientist
- Keeping a long-term holdoutData Scientist