Synthetic Control Method for Policy Evaluation
Also known as: Synthetic Control Method, SCM, Synthetic Control, Abadie-Diamond-Hainmueller method
The Synthetic Control Method (SCM) is a causal inference technique for evaluating the effect of a policy or intervention on a single treated unit — such as a region, country, or firm — by constructing a weighted combination of untreated comparison units that closely mirrors the treated unit before the intervention. Introduced by Abadie and Gardeazabal (2003) and formalized by Abadie, Diamond, and Hainmueller (2010), it provides a data-driven, transparent counterfactual for comparative case studies.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use SCM when a policy or shock affects a single aggregate unit (region, country, industry) and you have panel data covering multiple pre-treatment periods and a set of unaffected comparison units. It excels when the treatment occurs at the aggregate level making randomization impossible, and when you want a transparent, data-driven counterfactual rather than an arbitrary control group. At least 15-20 pre-treatment periods are strongly recommended for reliable weight estimation. Do not use SCM when the treated unit is poorly reproduced by any convex combination of donor units (poor pre-treatment fit), when the donor pool is very small (fewer than five units), when the intervention affects most potential control units, or when your data is at the individual level rather than aggregate.
Strengths & limitations
- Provides a rigorous, transparent, data-driven counterfactual for single treated units where standard regression methods are ill-suited.
- Visually interpretable: the pre-treatment fit and post-treatment gap are directly visible in outcome plots.
- Restricts the synthetic control to a convex combination of donor units, which guards against extrapolation beyond the support of the data.
- Does not require a large number of treated units, making it ideal for evaluating aggregate-level policies.
- Permutation-based inference requires no distributional assumptions and is valid in small samples of units.
- Requires a sufficiently long pre-treatment period (ideally 15+ periods) to estimate reliable weights; short panels yield poor fit.
- Validity depends on no spillovers or anticipation effects: donor units must be unaffected by the treatment, and the treated unit must not adjust behavior before the intervention.
- Inference via placebo tests becomes imprecise when the donor pool contains very few units, making it difficult to assess whether the observed gap is unusual.
- Cannot accommodate multiple simultaneous treated units without extensions such as generalized synthetic control or matrix completion methods.
- When pre-treatment fit is poor — the synthetic unit cannot approximate the treated unit — the post-treatment gap is uninterpretable as a causal effect.
Frequently asked
How is SCM different from Difference-in-Differences?
DiD compares averages across multiple treated and control units and requires a parallel-trends assumption. SCM targets a single treated unit and constructs a tailored counterfactual by optimally weighting donor units to reproduce the treated unit's pre-treatment trajectory; the pre-treatment fit is observable and verifiable rather than assumed.
How do I choose the donor pool?
Include units that were not affected by the intervention, were not subject to other simultaneous major shocks, and are plausibly comparable to the treated unit. Exclude units that could themselves have been treated or that experienced large idiosyncratic events during the study period. Selection should be made using only pre-treatment information.
What counts as good pre-treatment fit?
The pre-treatment root mean squared prediction error (RMSPE) — the average discrepancy between the treated unit and its synthetic counterpart before the intervention — should be small relative to the scale of the outcome. There is no universal threshold, but a RMSPE that is much larger than the typical period-to-period variation in the outcome suggests the synthetic control is a poor match.
How do I perform inference when there are few donor units?
Use permutation (placebo) tests: apply SCM treating each donor unit as if it were the treated unit and compute a post-treatment RMSPE ratio for each. The p-value is the fraction of placebos with a ratio at least as large as the treated unit's. With fewer than 10 donor units this distribution is coarse, and conventional significance levels (5%) may be unattainable.
Can SCM handle multiple treated units?
The basic SCM is designed for a single treated unit. Extensions such as the Generalized Synthetic Control (Xu 2017) and matrix completion methods (Athey et al. 2021) accommodate multiple treated units and staggered adoption, and should be preferred in those settings.
Sources
- Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California's Tobacco Control Program. Journal of the American Statistical Association, 105(490), 493-505. DOI: 10.1198/jasa.2009.ap08746 ↗
- Abadie, A., & Gardeazabal, J. (2003). The Economic Costs of Conflict: A Case Study of the Basque Country. American Economic Review, 93(1), 113-132. DOI: 10.1257/000282803321455188 ↗
How to cite this page
ScholarGate. (2026, June 3). Synthetic Control Method for Policy Evaluation. ScholarGate. https://scholargate.app/en/causal-inference/policy-evaluation-synthetic-control-method
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Interrupted Time SeriesCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Propensity Score MatchingResearch Statistics↔ compare