Synthetic Control Analysis

Summary

The Synthetic Control Method is a quasi-experimental approach to causal inference and policy evaluation that is an increasingly popular alternative to traditional evaluation methods. The method aims to construct a counterfactual for a treated unit from a set of similar, but less-than ideal, untreated comparison units. Researchers first must determine the appropriate analysis time frame, intervention time, outcome rate transformation, and criteria for donor-pool inclusion. Once properly bounded, a quadratic programming routine assigns weights to the donor-pool units so as to minimize the difference between the treated unit and the constructed synthetic control across the pre-intervention segment as well as covariates supplied. To the extent that the constructed synthetic control series approximates the treated series in the pre-intervention segment, the gap that emerges post-intervention can be interpreted causally. Post-estimation tests for spuriousness and sensitivity to alternative specifications are also conventional.

The method does not employ traditional significance testing, instead employing a randomization inference procedure, and does not require a strict assumption about the shape intervention effect over time. Synthetic control methods overcome some key limitations of criminal justice policy evaluation literature, namely, the absence of comparable interventions or control groups, as well as the fragility of findings to alternative model specifications. Traditional applications involve a single treated unit, and the method can be seen as a formalization of the choice of comparison unit in comparative case-study designs.

Synthetic controls have become increasingly common in the criminology and criminal justice policy evaluation literature, both in the United States and internationally. The method is particularly useful in research lines that have stagnated due to a lack of a suitable comparison unit in nature as well as from disagreements parsing the continuation of pre-intervention trends from the intervention effect. Synthetic control methods are an attractive alternative in settings in which regression-based estimation struggles, such as when the causal nexus of the outcome is unknown or variable as well as when data for causal covariates are unavailable for the treated or donor-pool units.