Coarsened Exact Matching (CEM)
You'll learn how CEM simplifies matching by temporarily grouping continuous variables into bins, making exact matches possible while maintaining balance between treatment and control groups.
The Core Problem
Regular exact matching (covered earlier) requires finding observations with identical values on all covariates. With continuous variables like age or income, finding perfect twins is nearly impossible. You'd reject most of your data, losing statistical power.
How CEM Works
Coarsened Exact Matching solves this through a clever three-step process:
- Coarsen: Temporarily bin continuous variables into meaningful categories (e.g., age 20-30, 31-40, etc.)
- Match: Apply exact matching on these coarsened bins—much easier now!
- Estimate: Use the original, uncoarsened data for analysis, but only keep matched observations
Think of it like organizing a matchmaking event: instead of finding people with exactly the same age, you group "people in their 30s" together. Once matched, you still know everyone's actual age.
Why It Preserves Balance
CEM guarantees balance on the coarsened variables because it uses exact matching logic. Unlike propensity score matching (which balances on average), CEM ensures no treated unit exists in a stratum without at least one control unit—and vice versa.
The researcher controls the coarsening: finer bins mean stricter matches but fewer matches overall; coarser bins increase sample size but reduce precision.
The Trade-off
You're trading precision (exact age) for feasibility (age ranges). But critically, the final analysis uses real values, not bins—the coarsening is just for the matching step.
Key Takeaway: CEM bridges exact matching and practical feasibility by temporarily coarsening variables into strata, enabling exact matches while preserving balance and using original data for causal effect estimation.