Multi-period Coarsened Exact Matching
Multi-period Coarsened Exact Matching Estimator · Also known as: Multi-period CEM, Longitudinal CEM, Panel CEM, Multi-wave CEM
Multi-period Coarsened Exact Matching (multi-period CEM) extends the CEM framework of Iacus, King, and Porro to longitudinal data with multiple pre- and post-treatment periods. It bins continuous covariates into coarsened categories, matches treated and control units that fall into the same cells across all relevant time periods, and then estimates a weighted average treatment effect that accounts for temporal structure.
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 multi-period CEM when you have panel or repeated cross-section data with multiple time periods, a binary treatment that affects some units but not others, and a rich set of pre-treatment covariates measured at more than one wave. It is especially valuable when the treatment assignment mechanism is complex enough that simple one-period matching would overlook time-varying confounding, and when you want nonparametric balance guarantees rather than relying on a correctly specified propensity model. Do not use it when covariates are measured at only one point in time (use standard CEM instead), when the number of matching cells explodes due to very many covariates with no natural coarsening (matching will prune excessively), or when the panel is very short and pre-treatment trends cannot be assessed.
Strengths & limitations
- Provides hard balance guarantees within matched strata without assuming a parametric form for the treatment assignment model.
- Extends nonparametric CEM to longitudinal settings, capturing time-varying confounding that single-period matching misses.
- The L1 imbalance statistic gives a transparent, model-free measure of covariate balance before and after matching.
- Pruning of unmatched units makes the estimation sample explicit, limiting extrapolation beyond the region of common support.
- Compatible with difference-in-differences or other panel estimators applied within matched strata to further remove residual confounding.
- Matching in multiple periods simultaneously can prune a large fraction of the sample if covariate profiles are heterogeneous across time, reducing statistical power.
- Coarsening choices — bin width and number of categories — affect matched sample composition and can be difficult to justify objectively.
- Only balances observed covariates; unmeasured confounders that differ between matched units remain a threat to causal identification.
- Requiring exact cell matches across multiple waves can fail entirely in small samples or with many covariates.
- No standard software package implements multi-period CEM end-to-end; researchers typically adapt the MatchIt or CEM R packages with custom period-weighting code.
Frequently asked
How does multi-period CEM differ from standard CEM?
Standard CEM matches units on covariates measured at a single point in time, typically just before treatment. Multi-period CEM extends this by matching on covariates from multiple time periods simultaneously, ensuring that treated and control units were comparable not just at one snapshot but throughout the observed window. This removes time-varying confounders that single-period CEM cannot address.
What is the L1 imbalance statistic?
L1 is a nonparametric measure of multivariate imbalance proposed by Iacus, King, and Porro. It captures the overall difference between the joint distributions of covariates in the treated and control groups by summing absolute differences across all coarsened cells. A value of 0 means perfect balance; 1 means complete separation. Matching aims to bring L1 close to 0.
How many time periods are needed?
At least two periods are required — one pre-treatment and one post-treatment — but multi-period CEM is most useful with three or more periods: at least two pre-treatment periods to assess trend comparability within matched strata, and one or more post-treatment periods for effect estimation.
Can I combine multi-period CEM with difference-in-differences?
Yes, and this combination is often recommended. Multi-period CEM creates a balanced subsample that satisfies the common support condition, and DiD is then applied within matched strata to additionally remove any remaining fixed group differences and common time trends. This two-step approach strengthens the parallel-trends argument.
What do I do if too many treated units are pruned?
Consider coarsening covariates more aggressively (wider bins) to widen strata and retain more matches. You can also restrict matching to the most critical covariates and control for the rest in a regression adjustment on the matched sample. If pruning remains severe, acknowledge the restriction in external validity and consider alternative methods such as propensity-score weighting, which does not discard unmatched units.
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 ↗
- Imai, K., Kim, I. S., & Wang, E. H. (2021). Matching methods for causal inference with time-series cross-sectional data. American Journal of Political Science, 67(3), 587-605. DOI: 10.1111/ajps.12685 ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-period Coarsened Exact Matching Estimator. ScholarGate. https://scholargate.app/en/causal-inference/multi-period-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.
- Coarsened Exact MatchingCausal inference↔ compare
- Difference-in-DifferencesEconometrics↔ compare
- Entropy BalancingCausal inference↔ compare
- Matching EstimatorCausal inference↔ compare
- Panel Data Coarsened Exact MatchingCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare