Why buffer zones exist
Geographic spillovers violate SUTVA: treatment in one market can affect outcomes in nearby markets through travel, media overlap, or local competition. Buffer zones create geographic separation by excluding untreated regions adjacent to treated markets.
How wide should a buffer be?
The buffer width depends on how fast spillovers decay with distance:
- Short-range spillovers: narrow buffers are sufficient.
- Long-range spillovers: buffers must be wide, which reduces effective sample size and power.
This is a design trade-off between contamination risk and statistical precision.
Buffer zones are not a universal fix
Buffer zones only address spatial spillovers. They do not solve:
- Network spillovers (social sharing, word-of-mouth).
- Competitive spillovers (national pricing responses).
When these are plausible, explicit spillover models with exposure mappings $h_g(D_{-g,t})$ are required.
When spillovers are part of the estimand
If spillovers are substantively important, you should measure outcomes in buffer zones rather than exclude them. A common design is three groups:
- Treated markets
- Neighboring buffer markets
- Distant controls
This enables estimation of direct and spillover effects, provided the neighborhood structure is correctly specified.
Sensitivity to neighborhood definitions
The three-group design depends on how you define “neighboring” and “distant” markets. Misspecification biases both direct and spillover estimates. Sensitivity checks across alternative neighborhood definitions are essential.
Takeaway
Buffer zones are a useful design tool, but they only handle spatial spillovers. When spillovers are central or non-spatial, the estimand must incorporate exposure and the design must measure spillover paths directly.
References
- Shaw, C. (2025). Causal Inference in Marketing: Panel Data and Machine Learning Methods (Community Review Edition), Section 3.3.3.