System Design · Meta · Medium
Experiment Design: Inserting One Extra Ad per 8 Organic Posts (a) Unit of Randomization User-level randomization is the correct unit. The treatment is a persistent change to every user’s feed composition, and we need to measure long-term outcomes such as 7-day retention, session length, and cumulative ad revenue. If we randomized at the session level, a single user would bounce between treated and control sessions, creating within-user contamination, inconsistent user…
Checking your access…