SUTVA

SUTVA, the Stable Unit Treatment Value Assumption, is a foundational assumption in the potential outcomes framework for causal inference. It asserts that the observed outcome for a unit under a given treatment equals the unit’s potential outcome under the assigned treatment, enabling causal effects to be identified from experiments and observational studies under certain conditions.

SUTVA comprises two components. The first is no interference between units: a unit’s potential outcomes depend only on its own treatment, not on the treatments assigned to other units. The second is no hidden variations of treatment: for each unit, there is a unique version of the treatment, so the potential outcomes depend only on the treatment level, not on how the treatment was delivered or other contextual factors.

Implications: When SUTVA holds, observed outcomes under different assignment schemes map to the corresponding potential outcomes, allowing unbiased estimation of causal effects such as the average treatment effect. Violations can bias estimates. Interference occurs in settings with spillovers or network effects. Variations in treatment versions, dosages, noncompliance, or multi-faceted implementations also violate SUTVA. To address violations, researchers may design studies to minimize interference (for example, cluster randomization), explicitly model interference (defining direct and indirect effects), or use methods that accommodate treatment heterogeneity.

SUTVA is an assumption rather than a directly testable proposition; its plausibility depends on context and it is common to assess robustness through sensitivity analyses or alternative modeling approaches.