Description
Builds a weighted combination of untreated units that tracks a treated unit before a change, and uses its post-change path as the counterfactual.
When one market, city or customer segment receives a change, synthetic control finds weights on a donor pool of untreated units so that the weighted combination matches the treated unit's pre-period outcomes and predictors. The post-period gap between treated and synthetic unit estimates the effect. Inference uses placebo tests, reapplying the method to each donor and comparing the treated unit's post/pre RMSPE ratio with the placebo distribution. It is transparent (weights are inspectable) and suited to the few-treated-unit situations common in product launches by country.
Attributes
- Discipline
- Experimentation & Causal Inference
- Loss type
- Unmigrated
- Goal
- Prove causation
- Stage
- Post-launch
- Data
- Quantitative
- Effort
- Weeks
Steps
- Define the treated unit (e.g. a country), donor pool of untreated units, outcome and launch date.
- Exclude donors exposed to similar changes or spillovers.
- Choose predictors and fit weights minimising pre-period discrepancy (using e.g. the Synth or augsynth packages).
- Check pre-period fit; poor fit invalidates the analysis.
- Estimate post-period gaps and run in-space placebo tests on each donor, and in-time placebos on fake dates.
- Report effect paths, weights and the placebo-based p-value.
Sources
- Abadie, Diamond & Hainmueller, Synthetic control methods for comparative case studies (2010), JASA
- Abadie, Using synthetic controls: Feasibility, data requirements, and methodological aspects (2021), Journal of Economic Literature
Also known as
- Synthetic control
- SCM
- Synthetic counterfactual
Categories
- Analytics & Behaviour
- Experimentation & Causal Inference