StatisticalMetrics and Experimentation
You want to test a change that alters comment sorting in a subreddit, but users participate in multiple subreddits and may share posts across them. How do you choose the unit of randomization and analysis to handle interference on the social graph?
Also asked as "Walk me through your approach to switchback testing versus cluster randomization when evaluating a new comment ranking algorithm."
Was asked at