Matching Methods (CEM / Optimal / Genetic)
General Matching Methods (CEM / Optimal / Genetic) · Also known as: coarsened exact matching, optimal matching, genetic matching, CEM, Genel Eşleştirme Yöntemleri (CEM / Optimal / Genetic)
Matching Methods are a family of causal-inference techniques beyond propensity-score matching that pair treated and control units with similar covariates so that a treatment effect can be read off the balanced sample. The family includes Coarsened Exact Matching (Iacus, King & Porro, 2012), optimal matching, and genetic matching.
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 matching methods on observational (cross-sectional or longitudinal) data when you want to estimate a treatment effect and you can credibly assume conditional independence (ignorability), with at least about 80 observations so that enough matched units remain. They fit continuous, binary, or categorical covariates, and are a strong alternative to plain propensity-score matching when you need an explicit guarantee on covariate balance (CEM) or the smallest total distance between matches (optimal matching). For CEM, the cut points used to coarsen the matching variables must be chosen meaningfully.
Strengths & limitations
- Coarsened Exact Matching guarantees covariate balance by construction, removing the trial-and-error balance checking that propensity-score matching often needs.
- Optimal matching minimises the total within-pair distance, producing the closest possible overall set of matches.
- Works with continuous, binary, and categorical covariates, and any residual imbalance can be mopped up with a weighted regression after matching.
- With fewer than about 80 observations, too few units survive matching and the treatment-effect estimate becomes unstable.
- When the region of common support is narrow, good matches cannot be found and balance (SMD) cannot be achieved; inverse-probability weighting is then preferable.
- Identification still rests on the untestable conditional-independence assumption — matching only balances observed covariates, not hidden confounders.
Frequently asked
How is Coarsened Exact Matching different from propensity-score matching?
Coarsened Exact Matching coarsens each covariate into bins and matches units exactly within those bins, which guarantees a chosen level of covariate balance directly. Propensity-score matching instead collapses the covariates into a single score and usually requires repeated balance checking to confirm the matches worked.
What does optimal matching optimise?
Optimal matching selects the set of treated-control pairings that minimises the total covariate distance summed across all matches, so the matched sample is as close as possible overall rather than pairing units greedily one at a time.
How do I confirm the matching worked?
Check covariate balance after matching using the standardised mean difference (SMD) between treated and control groups. If some imbalance remains, run a weighted regression on the matched sample to adjust for it.
When should I not use matching?
With fewer than about 80 observations, switch to propensity-score matching; and when the common-support region is narrow so good matches cannot be found, use inverse-probability weighting instead.
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 ↗
- Hansen, B. B. (2004). Full Matching in an Observational Study of Coaching for the SAT. Journal of the American Statistical Association, 99(467), 609-618. DOI: 10.1198/016214504000000647 ↗
How to cite this page
ScholarGate. (2026, June 1). General Matching Methods (CEM / Optimal / Genetic). ScholarGate. https://scholargate.app/en/causal-inference/matching-methods
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.
- Heterogeneous Treatment EffectsCausal inference↔ compare
- Inverse Probability WeightingCausal inference↔ compare
- Local Average Treatment EffectCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Sensitivity Analysis for Unmeasured ConfoundingCausal inference↔ compare