Multi-period Synthetic Control Method
Also known as: multi-period SCM, extended synthetic control, synthetic control with multiple treatment periods, staggered synthetic control
The multi-period synthetic control method extends the classic synthetic control framework to settings where treatment occurs across several distinct periods or where the researcher needs to track causal effects over a prolonged post-treatment window. It constructs a weighted combination of untreated units that reproduces the treated unit's pre-treatment trajectory, then uses that synthetic counterfactual across all post-treatment periods to estimate time-varying treatment effects.
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 the multi-period synthetic control when you have a single treated aggregate unit (or a small number of them) — such as a country, region, or firm — and want to estimate causal effects across multiple distinct post-treatment periods rather than collapsing them into one. It suits aggregate panel data with a long pre-treatment window (ideally more pre-periods than predictors) and a clearly defined treatment date or staggered rollout. Do not use it when the number of donor units is very small (fewer than five or six), when the pre-treatment fit is poor, when treatment assignment is endogenous and correlated with time-varying shocks, or when you need unit-level rather than aggregate-level inference.
Strengths & limitations
- Provides a transparent, data-driven counterfactual without imposing parametric functional forms on the outcome model.
- Tracks how the treatment effect evolves over multiple post-treatment periods, revealing whether impacts grow, fade, or reverse.
- Permutation-based inference is exact and does not rely on large-sample approximations, making it valid even with few units.
- The pre-treatment fit serves as a built-in falsification check: a poor fit exposes an implausible counterfactual before any causal claims are made.
- Readily accommodates heterogeneous treatment timing across units when combined with staggered synthetic control extensions.
- Requires a sufficiently large and diverse donor pool; with fewer than five or six untreated units the method loses credibility and inference power.
- Demands a long pre-treatment period to achieve good fit; sparse pre-treatment data leads to an unreliable synthetic twin.
- Extrapolates if the treated unit lies outside the convex hull of donor units, potentially producing nonsensical counterfactuals.
- Cannot handle a large number of simultaneously treated units without substantial methodological extension (e.g., matrix completion or staggered variants).
Frequently asked
How does the multi-period variant differ from standard synthetic control?
The standard method typically focuses on a single aggregate post-treatment effect or a fixed post-treatment window. The multi-period variant explicitly estimates the treatment effect gap in each successive post-treatment period, producing a time profile of effects. This requires careful attention to whether the pre-treatment weights remain valid throughout an extended post-treatment horizon.
How many pre-treatment periods do I need?
More is better. As a practical rule, you need enough pre-treatment periods to fit the treated unit's trajectory without overfitting. Abadie (2021) recommends that the number of pre-treatment periods substantially exceed the number of predictors used in the weight-optimization step. Fewer pre-treatment periods than predictors almost always yield poor out-of-sample performance.
Can I use this with staggered treatment rollout across units?
With some care, yes. Extensions such as the staggered synthetic control (Cattaneo et al.) or matrix completion with staggered treatment adapt the framework to settings where different units adopt treatment at different times. The key challenge is ensuring that early adopters are not used as donors for late adopters who are still in their pre-treatment period.
What should I do if pre-treatment fit is poor?
Poor fit is a warning that no credible synthetic counterfactual exists for this treated unit given the available donor pool. Options include expanding the donor pool, adding predictors, using the augmented synthetic control (Ben-Michael et al. 2021) which combines weighting with outcome-model imputation, or reconsidering whether synthetic control is the right design for this application.
Is the method valid for micro-level (individual) data?
Synthetic control is designed for aggregate units such as countries, states, or firms where each unit represents a distinct time series. Applying it to individual-level microdata is generally inappropriate; difference-in-differences or matching estimators are better suited to that context.
Sources
- Abadie, A. (2021). Using synthetic controls: Feasibility, data requirements, and methodological aspects. Journal of Economic Literature, 59(2), 391-425. DOI: 10.1257/jel.20191450 ↗
- Ben-Michael, E., Feller, A., & Rothstein, J. (2021). The augmented synthetic control method. Journal of the American Statistical Association, 116(536), 1789-1803. DOI: 10.1080/01621459.2021.1929245 ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-period Synthetic Control Method. ScholarGate. https://scholargate.app/en/causal-inference/multi-period-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
- Dynamic Synthetic Control MethodCausal inference↔ compare
- Multi-period Difference-in-differencesCausal inference↔ compare
- Panel Data Synthetic Control MethodCausal inference↔ compare
- Synthetic Control MethodCausal inference↔ compare