Synthetic Control Policy Lab

For situations where randomization is impossible and parallel trends are questionable.

Overview

When evaluating policies that affect entire units (states, countries, organizations), traditional methods often fail. Synthetic control constructs a counterfactual from weighted combinations of donor units, providing rigorous causal inference even with a single treated unit.

Methods

Classic Synthetic Control

• Optimal weighting of control units to match pre-treatment outcomes
• No functional form assumptions
• Transparent weights reveal comparison strategy

Uncertainty Quantification

• Placebo tests using untreated units
• Permutation inference for p-values
• Confidence intervals via conformal inference

When to Use This

Good fit:

• Single treated unit (or few treated units)
• Good pre-treatment outcome data
• Pool of plausible comparison units

Not ideal:

• Treatment affects all potential donors (spillovers)
• Short pre-treatment period
• Multiple staggered treatments

Key Assumptions

1. No anticipation: Units don’t change behavior before treatment
2. No spillovers: Treatment doesn’t affect control units
3. Convex hull: Treated unit is within the range of controls