Back to Blog
Technical Guide

SUTVA Violations and Network Effects in A/B Tests

Updated December 2026
15 min read
TL;DR

SUTVA (Stable Unit Treatment Value Assumption): One user's treatment shouldn't affect another's outcome. Violated in: marketplaces, social networks, referral programs. When violated, standard A/B tests give biased results. Solutions: cluster randomization, geo-based splits, or time-based experiments.

What is SUTVA?

SUTVA is a core assumption of A/B testing:

"The treatment assigned to one unit does not affect the outcome of another unit."

In plain English: What you show User A shouldn't affect User B's behavior.

When SUTVA is Violated

Marketplace (Uber, Airbnb)

Violation: Driver availability affects rider experience

Solution: Geo-based randomization

Social networks

Violation: User A sees content from User B in different variant

Solution: Cluster randomization

Two-sided platforms

Violation: Supply side affects demand side

Solution: Time-based or market-based splits

Referral programs

Violation: User A refers User B, both in different variants

Solution: Household/network-level randomization

Real Example: Marketplace

Scenario: Testing a new driver incentive on Uber.

  • Variant A (control): Normal driver pay
  • Variant B: +20% driver pay

The problem: Drivers in Variant B are more available. This affects riders in Variant A (control) who get faster pickups because more drivers are online.

Result: Your test underestimates the true effect because the control group benefits from spillover.

Solutions

Cluster Randomization

Randomize by city, region, or social cluster instead of individual users

Time-Based Experiments

Run variant A for 2 weeks, then variant B for 2 weeks (switchback design)

Accept the Bias

For most websites, SUTVA violations are minimal—proceed with standard A/B testing

When to Worry About SUTVA

High Risk:

  • • Marketplaces
  • • Social networks
  • • Referral programs
  • • Shared inventory systems

Low Risk:

  • • E-commerce product pages
  • • SaaS pricing pages
  • • Landing pages
  • • Most marketing sites

For Most Sites

SUTVA violations are rare outside of marketplaces and social networks. Standard A/B testing with ExperimentHQ works fine.

Share this article

Ready to start A/B testing?

Free forever plan available. No credit card required.