Synthetic Control & Synthetic DiD Services
When a policy or event affects a single unit—one region, one firm, one market—and no other unit is a clean comparison, the synthetic control method builds one: a weighted combination of untreated units that closely tracks the treated unit before the intervention. The post-intervention gap estimates the effect.
The synthetic control method estimates the effect of an intervention on a single treated unit by constructing a “synthetic” comparison—a weighted average of untreated units chosen to match the treated unit’s pre-intervention outcomes and characteristics. The difference between the treated unit and its synthetic control after the intervention estimates the treatment effect.
What synthetic control does
Some of the most important interventions happen to a single unit: a reform adopted by one region, a shock to one industry, a change affecting one firm, a policy in one country. Difference-in-differences needs a comparable untreated group, but for a single treated unit no other single unit is usually a convincing counterfactual—each differs from the treated unit in its own way. The synthetic control method addresses this by constructing a comparison rather than picking one.
It searches a pool of untreated units (the “donor pool”) for a weighted combination that closely reproduces the treated unit’s outcome path before the intervention, along with its key characteristics. That weighted combination—the synthetic control—serves as the counterfactual: an estimate of what the treated unit’s outcome would have been without the intervention. After the intervention, the gap between the treated unit and its synthetic control estimates the effect. The method’s appeal is transparency: the weights are explicit, the pre-intervention fit is visible, and a good match before treatment makes the post-treatment divergence credible.
When to use it, and synthetic DiD
Synthetic control suits comparative case studies: one or a few treated units, a clear intervention date, a pool of untreated donor units, and a reasonably long pre-intervention series so the match can be established and judged. It has become a standard tool in policy evaluation and aggregate-level management and economics research precisely for the single-unit case that DiD handles poorly.
Synthetic difference-in-differences (synthetic DiD) is a more recent method that combines the strengths of both approaches—it uses data-driven weights like synthetic control while retaining the two-way structure and robustness of DiD, and can perform well when neither pure method is ideal. A related, simpler design is the interrupted time series, which models a single unit’s own trajectory before and after an intervention without an external control; it is useful when no donor pool exists, but it rests on more assumptions about the counterfactual trend. We select among these based on the number of treated units, the availability of a donor pool, and the length and stability of the pre-intervention data.
Comparative designs for aggregate interventions
| Method | Counterfactual | Best when |
|---|---|---|
| Difference-in-differences | A comparable untreated group | Many treated & control units, parallel trends |
| Synthetic control | Weighted combination of donor units | One (or few) treated units, good donor pool |
| Synthetic DiD | Weighted donors + DiD structure | Want data-driven weights with DiD robustness |
| Interrupted time series | The unit’s own pre-trend, extrapolated | No external comparison available |
Doing synthetic control credibly
The credibility of a synthetic control rests first on pre-intervention fit: the synthetic unit must track the treated unit closely over a substantial pre-period. A poor pre-fit means the counterfactual is not believable and the post-intervention gap cannot be trusted, so the fit is examined and reported rather than assumed. The donor pool must also be chosen carefully—units should be plausibly unaffected by the intervention and comparable enough to be relevant—and the method should not be asked to extrapolate beyond the range of the donors.
Because there is typically one treated unit, conventional standard errors do not apply, so inference is done by placebo tests: the analysis is re-run pretending each donor unit was treated, and the treated unit’s estimated effect is judged against the distribution of these placebo “effects.” If the real effect is large relative to placebos, it is unlikely to be chance. We report placebo/permutation inference and pre-intervention fit statistics, and—where it strengthens the design—use synthetic DiD or complementary checks. As with all these designs, the method assumes no other shock hit the treated unit at the same time; that assumption is argued, not waved away.
A synthetic control is only as credible as its pre-intervention fit. If the synthetic unit does not closely track the treated unit before the intervention, the counterfactual—and the estimated effect—cannot be trusted. Fit is examined and reported, and inference is done through placebo tests, not conventional standard errors.
Software
We deliver synthetic control and synthetic DiD in established, reproducible tools—R (Synth, synthdid, tidysynth) and Stata—with the donor weights, pre-fit diagnostics, placebo/permutation inference, and clear treated-versus-synthetic plots, all with versioned code.
How we deliver a synthetic control study
Synthetic control sits within our wider causal-inference practice—so the counterfactual is constructed transparently, the pre-fit is judged honestly, and inference is done appropriately for a single treated unit.
We start from the treated unit, the intervention date, and a defensible donor pool of untreated units, and confirm there is enough pre-intervention data to build and judge a match. We construct the synthetic control (or fit synthetic DiD), examine pre-intervention fit, and run placebo/permutation inference along with the robustness checks the design calls for—including whether other contemporaneous shocks threaten the counterfactual.
Reporting sets out the donor pool and weights, the pre-intervention fit, the treated-versus-synthetic comparison, the placebo-based inference, and the assumptions—so the causal claim can be judged transparently.
You receive the estimated effect (the post-intervention gap) with placebo-based inference, the donor weights and pre-fit diagnostics, the treated-versus-synthetic and placebo plots, robustness checks, and reproducible analytical code and analysis-ready files (where appropriate and permitted)—with the strength of the pre-intervention match stated plainly.
Synthetic control across Management & Allied Studies
Single-unit interventions at an aggregate level—a region, market, industry, or firm—are common in policy and management research, and are exactly where synthetic control fits.
Economics & Public Policy
The classic setting—estimating the effect of a reform, tax, or programme adopted by a single region or country against a synthetic comparison.
Finance & Accounting
Effects of a regulatory change, market event, or firm-specific shock where one entity is treated and a weighted peer benchmark is built.
Management & Organizational Research
Effects of a major change at a single firm or business unit, compared against a synthetic combination of comparable units.
Marketing & Consumer Research
Effects of a campaign, entry, or platform change in one market, benchmarked against a synthetic control of other markets.
Strategy & Entrepreneurship
Effects of an ecosystem or policy shock on a single region or cluster, using untreated regions as the donor pool.
Operations & Information Systems
Effects of a system or process change rolled out to one site or region, compared to a synthetic combination of others.
Synthetic control: common questions
A single treated region, market, or firm?
When only one unit is affected and no single comparison is convincing, synthetic control constructs a credible counterfactual—with pre-intervention fit examined, placebo inference, and synthetic DiD where it strengthens the design.