Programme Evaluation and Spillover Effects

Causal Question / Estimand

The treatment effect when a program also affects untreated units — and the additional estimands spillovers create: the effect on the ineligible/untreated (the spillover or indirect effect) and the total program effect, alongside the direct treatment effect.

Identification Strategy

A practitioner’s guide to handling interference, which violates the no-interference part of SUTVA that standard evaluation assumes. Explains how spillovers (through markets, networks, social interactions, general equilibrium) bias naive treated-vs-control comparisons when controls are contaminated. Lays out designs that identify spillovers — most importantly partial-population / two-level randomization (randomize treatment intensity across clusters and treatment status within them), plus distance/network-based designs — so the direct and indirect effects can be separated.

Key Assumptions

A correctly specified interference structure (who can affect whom); a clean “pure control” group genuinely outside the spillover range; Randomization at the relevant level(s). Relaxes the no-interference assumption of SUTVA rather than maintaining it.

Threats to Validity

Contaminated controls (spillovers make the control group a poor counterfactual); mis-specified interference range/network; insufficient design (single-level randomization cannot separate direct from indirect effects).

Setting / Data

n/a — methodological guidance oriented to development field experiments (e.g. cash-transfer programs such as PROGRESA).

Key Claims

• Ignoring spillovers biases program-effect estimates and discards a policy-relevant quantity (the indirect effect).
• Multi-level / partial-population randomization is the workhorse design for identifying direct and spillover effects separately.
• Defining the unit of interference and a truly unaffected control group is the central design problem.

Connections

• Relaxes SUTVA (no interference); complements the experimental-design focus of BruhnMcKenzie2009-InPursuitOfBalance and AtheyImbens2017-EconometricsOfRandomizedExperiments.
• See also RCT, Foundations.

Citation

Angelucci, M., & Di Maro, V. (2016). Programme Evaluation and Spillover Effects. Journal of Development Effectiveness, 8(1), 22–43.