32.8 Inference in SCM
Unlike traditional methods, SCM does not rely on standard statistical inference due to:
- Undefined sampling mechanism (e.g., only one treated unit).
- SCM is deterministic, making p-values difficult to interpret.
32.8.1 Permutation (Placebo) Inference
To perform inference:
- Iteratively reassign the treatment to units in the donor pool.
- Estimate placebo treatment effects for each synthetic control.
- Compare the actual treatment effect to the placebo distribution.
The treatment effect is considered statistically significant if it is extreme relative to the placebo distribution.
32.8.2 One-Sided Inference
- Recommended due to the limited number of treated cases.
- The permutation test is more robust than standard p-values.
For benchmark distributions (e.g., uniform permutation distributions), see (Firpo and Possebom 2018).
References
Firpo, Sergio, and Vitor Possebom. 2018. “Synthetic Control Method: Inference, Sensitivity Analysis and Confidence Sets.” Journal of Causal Inference 6 (2): 20160026.