You work on a mobile game with multiplayer interactions and want to test a new feature that changes how players engage with one another in matches or social groups. A standard user-level A/B test may be biased because treated players can affect control players through matchmaking, guild activity, or friend invites.
How would you think about experimentation when the feature has strong network effects or multiplayer interactions? How would that change your choice of randomization unit, metrics, and interpretation of results?
Reasoning about network interference and SUTVA violationsChoosing the right unit of randomizationDefining primary and guardrail metrics under spilloversSizing a clustered experiment with an explicit MDE