Heterogeneous Treatment Effect Synthetic Control Method
Also known as: HTE-SCM, heterogeneous SCM, heterogeneous synthetic control, SCM with HTE
The Heterogeneous Treatment Effect Synthetic Control Method (HTE-SCM) extends the classical synthetic control framework by allowing the causal effect of an intervention to vary across time periods, subgroups, or outcome dimensions rather than collapsing it to a single average estimate. It combines the counterfactual donor-pool matching logic of Abadie et al. (2010) with modern heterogeneous-effects machinery to recover time-varying or subgroup-specific treatment paths.
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 HTE-SCM when you have a single treated unit (or a small number) and a reasonably large pool of untreated comparison units observed over many pre- and post-treatment periods, and you need to recover how the causal effect evolves over time or differs across contexts — for instance, whether a policy's impact faded, grew, or varied by subgroup. It is most appropriate for aggregate-level interventions (countries, states, firms) where randomisation is impossible and the pre-treatment fit can be made tight. Avoid it when the donor pool is very small (fewer than 10 units), when the pre-treatment fit is poor, when the treated unit is an outlier in the covariate space, or when you need unit-level individual heterogeneity rather than aggregate time-path heterogeneity.
Strengths & limitations
- Recovers a full time path of treatment effects rather than a single average, revealing whether the policy impact grew, faded, or was heterogeneous across the post-treatment horizon.
- Transparent and visually interpretable: the gap between the treated unit's outcome and its synthetic counterpart is plotted and immediately readable.
- Permutation-based placebo inference does not rely on large-sample normality assumptions, making it valid even with a small number of post-treatment periods.
- Augmented variants (e.g., Ben-Michael et al. 2021) reduce bias from imperfect pre-treatment fit through a regression bias-correction term.
- Robust to time-invariant unobserved confounders, as the pre-treatment matching absorbs fixed differences between treated and donor units.
- Requires a large, comparable donor pool: results are unreliable if fewer than roughly 10 untreated units are available for placebo inference.
- The convex-weights constraint can prevent a good pre-treatment fit if the treated unit lies outside the convex hull of donors; extrapolation beyond the donor distribution is needed but is not supported by the standard method.
- Subgroup-level heterogeneity analysis is indirect: the method was designed for aggregate units, not individual-level variation, and recovering covariate-driven HTE requires additional modelling assumptions.
- Long post-treatment horizons risk poor counterfactual accuracy as donors and the treated unit drift apart for reasons unrelated to the intervention.
Frequently asked
How does HTE-SCM differ from standard synthetic control?
Standard SCM typically reports an average post-treatment effect or a single gap at a chosen horizon. HTE-SCM explicitly estimates and analyses the period-by-period treatment-effect path, and may further decompose that path by subgroup or moderating variable, treating the heterogeneity in time or context as the primary object of inference rather than a nuisance.
How many donor units do I need?
At least 10-20 donor units are recommended to generate a meaningful placebo distribution for inference. Fewer than 10 donors makes it difficult to assess whether the treated unit's effect path is unusual relative to chance, weakening any causal claim.
What if my pre-treatment fit is imperfect?
An imperfect fit inflates post-treatment gaps and can simulate treatment effects that are actually artefacts of the mismatch. The Augmented SCM (Ben-Michael et al. 2021) adds a regression bias-correction that partially addresses poor fit; alternatively, revisit the choice of predictor variables and the donor pool composition.
Can HTE-SCM handle multiple treated units?
Yes, by running a separate synthetic control for each treated unit and then summarising the resulting time-path estimates across units. This multi-unit extension supports richer heterogeneity analysis across units, though each unit requires its own donor pool that excludes other treated units.
Is this the same as the generalised synthetic control or matrix completion method?
Related but distinct. Generalised synthetic control (Xu 2017) uses an interactive fixed-effects model that can handle multiple treated units and naturally recovers time-varying effects. Matrix completion methods (Athey et al. 2021) similarly allow heterogeneous effects. HTE-SCM is an umbrella term covering SCM-based analyses that target heterogeneous or time-varying effects, including augmented and generalised variants.
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 ↗
- 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). Heterogeneous Treatment Effect Synthetic Control Method. ScholarGate. https://scholargate.app/en/causal-inference/heterogeneous-treatment-effect-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.
- Causal Impact AnalysisCausal inference↔ compare
- Difference-in-DifferencesEconometrics↔ compare
- Heterogeneous Treatment Effect Difference-in-DifferencesCausal inference↔ compare
- Matching EstimatorCausal inference↔ compare
- Panel Data Synthetic Control MethodCausal inference↔ compare
- Synthetic Control MethodCausal inference↔ compare