Skip to content
Quantoros

How would you handle network effects / interference between treatment and control groups (e.g., in a social network or…

StatisticsLevel: IntermediateA/B testing and experimentation

How would you handle network effects / interference between treatment and control groups (e.g., in a social network or marketplace)?

Hints
  • Identify the random variable, sample and estimated parameter.
  • State the assumptions; distinguish an estimate from its uncertainty.

Answer

Standard A/B testing assumes the Stable Unit Treatment Value Assumption (SUTVA) — one user's treatment doesn't affect another's outcome. This breaks in social networks (a treated user's friends are indirectly affected) and marketplaces (treated sellers compete for the same limited buyer demand as control sellers, causing spillover). Solutions include cluster/graph-based randomization (randomize whole clusters of connected users together), geo-based experiments (randomize by region), or switchback experiments (randomize by time period rather than by unit) for marketplace-style interference.

Source and licence: Ishan Dutta — Awesome Data Science Interview Q&A · MIT

Related questions