Synthetic Control: Building a Comparison From Many

How economists create an artificial "twin" of a place from a weighted mix of others to estimate the effect of a policy or event when no single good comparison exists.

To know whether a policy worked, economists need to know what would have happened without it, called the counterfactual. When a policy affects a single state or country, finding a good comparison is hard. The synthetic control method offers a solution.

The idea

Instead of comparing the treated place with one other place, economists build a synthetic version: a weighted combination of several untreated places that together closely match the treated place before the policy.

After the policy, the difference between the real place and its synthetic twin estimates the policy’s effect.

The Basque Country

In 2003, economists Alberto Abadie and Javier Gardeazabal used this method to estimate the economic cost of terrorism in Spain’s Basque Country. They built a synthetic Basque Country from other Spanish regions. After terrorism began in the late 1960s, the real Basque economy fell below its synthetic twin by about 10 percent of income per person.

California’s tobacco law

In 2010, Abadie and colleagues studied Proposition 99, a 1988 California law that raised cigarette taxes and funded anti-smoking campaigns. They built a synthetic California from other US states. Cigarette sales in real California fell much faster than in the synthetic version, suggesting the law substantially reduced smoking.

Other uses

Effects of German reunification on West Germany’s economy.
Impacts of minimum wage changes in particular cities or states.
Effects of economic reforms or natural disasters.

Strengths

Transparent: the weights show which places make up the comparison.
Useful when only one or a few places are treated.

Limits

Needs a long pre-policy period with good data.
Assumes no major other shocks hit only the treated place.
Works poorly if no combination of places closely matches the treated one.

The state's new law

A state introduces a new education policy. Researchers build a synthetic version using a mix of other states that tracked its test scores closely for ten years before. After the policy, the real state's scores rise faster than its synthetic twin's, suggesting the policy helped.

Thinking comparing with the most similar single state is always enough

One state may differ in important ways. A weighted combination can match the treated place more closely.

Key takeaways

Synthetic control builds a weighted "twin" of a treated place from untreated places.
Abadie and Gardeazabal estimated terrorism's cost to the Basque economy at about 10 percent.
A study of California's Proposition 99 found it cut cigarette sales.
The method needs long pre-policy data and good matches.