Policy Evaluation via Coarsened Exact Matching (CEM)
Coarsened Exact Matching for Policy Evaluation · Also known as: CEM, Coarsened Exact Matching, CEM policy evaluation, coarsening-based matching
Coarsened Exact Matching (CEM) is a quasi-experimental causal-inference technique that creates balanced treatment and control groups from observational data by temporarily coarsening covariates into bins, exactly matching units within those bins, and then pruning unmatched observations before estimating policy effects. Introduced by Iacus, King, and Porro, CEM belongs to the monotonic imbalance bounding family of matching methods and is especially popular in policy evaluation.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use CEM when you have observational data on a policy or intervention, a rich set of pre-treatment covariates on which to match, and enough untreated units to find matches. It is well-suited when you want explicit covariate balance guarantees rather than relying on a correctly specified propensity model, and when the dataset is moderately large (hundreds to thousands of units). Avoid CEM when the covariate space is very high-dimensional with few control units, since many strata will be empty and most treated units will be pruned, leaving a very small matched sample that may not represent the full treated population.
Strengths & limitations
- Provides monotonic imbalance bounding: coarsening the bins can only reduce imbalance, giving transparent and provable balance guarantees without iterative checking.
- Eliminates dependence on a correctly specified propensity model, making causal estimates more robust to model misspecification.
- Users retain full control over the coarsening cutpoints, making the matching assumptions explicit and auditable.
- Works with continuous, ordinal, and nominal covariates simultaneously in a single matching step.
- Reduces model dependence by pruning the sample to a region of common support before outcome analysis.
- High-dimensional covariate spaces can produce many empty strata, drastically reducing matched sample size and threatening external validity.
- Results depend on the user-chosen coarsening cutpoints; different choices can lead to different matched samples and ATT estimates.
- Pruning unmatched units means the estimated ATT applies only to matched treated units, not necessarily all treated units.
- Does not handle unobserved confounders; causal validity still requires the assumption that all relevant confounders are observed and included.
Frequently asked
How is CEM different from propensity-score matching?
Propensity-score matching (PSM) collapses all covariates into a single estimated score and matches on that scalar, meaning balance on individual covariates is not guaranteed and depends on model correctness. CEM matches exactly within coarsened covariate bins, guaranteeing local balance on every included covariate simultaneously and without relying on a correctly specified propensity model.
What does 'monotonic imbalance bounding' mean?
It means that coarsening the bins further (making them wider) can only reduce measured imbalance between the matched groups, never increase it. This gives users a provable, monotone control over post-matching covariate balance that no propensity-score or distance-based method provides.
How should I choose the coarsening cutpoints?
Cutpoints should reflect substantive knowledge about which covariate values are 'close enough' to be treated as equivalent for the study question. Automatic cutpoints (e.g., based on quantiles or Sturges' rule) are a defensible default, but researchers are encouraged to conduct sensitivity analyses with alternative coarsenings and report the matched-sample size for each.
What happens if many treated units are pruned?
If a large share of treated units have no matched control, the remaining matched sample may be unrepresentative of the full treated population, and the ATT estimate applies only to a select subgroup. This reduces external validity. Adding more control units, relaxing coarsening, or combining CEM with a regression adjustment can help.
Is CEM suitable for panel data?
Yes. CEM is typically applied to pre-treatment cross-sectional covariates to create a balanced matched sample, and the policy outcome can then be observed over time in panel or repeated cross-section data. It is commonly combined with difference-in-differences in the matched sample for added robustness.
Sources
- Iacus, S. M., King, G., & Porro, G. (2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1-24. DOI: 10.1093/pan/mpr013 ↗
- Iacus, S. M., King, G., & Porro, G. (2011). Multivariate matching methods that are monotonic imbalance bounding. Journal of the American Statistical Association, 106(493), 345-361. DOI: 10.1198/jasa.2011.tm09599 ↗
How to cite this page
ScholarGate. (2026, June 3). Coarsened Exact Matching for Policy Evaluation. ScholarGate. https://scholargate.app/en/causal-inference/policy-evaluation-coarsened-exact-matching
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Entropy BalancingCausal inference↔ compare
- Inverse Probability WeightingCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Synthetic Control MethodCausal inference↔ compare