SUTVA: Interference and Hidden Treatments

Chapter 11 — SUTVA: Interference and Hidden Treatments The Stable Unit Treatment Value Assumption underlies every standard estimate of the average treatment effect. This chapter examines what happens when it fails—either because of spillovers between units (interference) or because the treatment itself has hidden versions. It covers difference-in-means under interference, linear-in-means models, clustered randomized trials to attenuate spillovers, and randomization inference under interference. Crucially, it also shows how spillovers can be embraced rather than avoided using randomized saturation designs to measure spillover effects directly.

1. The Stable Unit Treatment Value Assumption (SUTVA) incorporates two distinct assumptions: no interference and no hidden versions of treatment.
2. The ATE is not well defined under SUTVA violations, but recovering other parameters of interest is possible with additional assumptions.
3. When SUTVA violations are a nuisance, in the design stage the experimentalist can use approaches to identify and attenuate violations.
4. A silver lining to SUTVA violations is that in certain cases they provide insights relevant to achieving EP1 and EP2 and can aid in the measurement of spillover effects.