Multi-period Matching Estimator
Also known as: panel matching estimator, longitudinal matching, multi-wave matching, repeated-cross-section matching
The multi-period matching estimator extends the standard matching framework to settings with multiple time periods, pairing each treated unit to similar untreated units based on pre-treatment covariates or propensity scores, then using within-pair before-after differences to estimate the average treatment effect on the treated (ATT). Leveraging repeated observations, it simultaneously controls for observed confounders and time-invariant unobserved heterogeneity.
Key highlights
- Simultaneously removes observed confounding through matching and time-invariant unobserved confounding through the before-after differencing, yielding doubly robust bias reduction.
- Using multiple pre-treatment periods to construct or validate matches greatly strengthens the plausibility of the parallel-trends assumption.
- Event-study plots across post-treatment periods reveal how treatment effects evolve over time, not merely an average.
- Non-parametric in spirit — does not require specifying the functional form of the outcome model.
- Naturally handles staggered treatment adoption by matching within cohorts defined by treatment timing.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the multi-period matching estimator when you have panel or repeated cross-sectional data spanning several time periods, a treatment that begins at a single wave or staggered across cohorts, and a rich set of pre-treatment covariates or lagged outcomes to match on. It is appropriate when you expect time-invariant unobserved confounders and want to combine the bias-reduction of matching with the longitudinal control offered by differencing. Do not use it when no pre-treatment periods are available (then standard cross-sectional matching is the only option), when the treatment is continuous rather than binary, or when the treated sample is too small to find adequate matches.
Strengths & limitations
- Simultaneously removes observed confounding through matching and time-invariant unobserved confounding through the before-after differencing, yielding doubly robust bias reduction.
- Using multiple pre-treatment periods to construct or validate matches greatly strengthens the plausibility of the parallel-trends assumption.
- Event-study plots across post-treatment periods reveal how treatment effects evolve over time, not merely an average.
- Non-parametric in spirit — does not require specifying the functional form of the outcome model.
- Naturally handles staggered treatment adoption by matching within cohorts defined by treatment timing.
- Requires sufficiently many untreated units with similar covariate profiles; poor overlap (common support) degrades match quality and biases estimates.
- Cannot control for time-varying unobserved confounders that differ between treated and matched controls after matching.
- Matching on high-dimensional pre-treatment histories can suffer from the curse of dimensionality unless dimension-reduction tools such as propensity scores are used.
- With very few treated units, variance is high and inference is unreliable even with good matches.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is multi-period matching different from standard propensity-score matching?
Standard matching compares treated and control units at a single point in time. Multi-period matching also differences within matched pairs over time, removing time-invariant unobserved confounders that single-period matching cannot handle. It therefore combines the strengths of matching and difference-in-differences.
Do I need a balanced panel?
No, but missing observations complicate the computation of before-after differences. With an unbalanced panel, restrict the analysis to units observed in all relevant periods, or use inverse-probability-of-observation weights to account for attrition.
How many pre-treatment periods improve the method?
More pre-treatment periods improve match quality because you can match on outcome trajectories rather than a single baseline value, and you can conduct a more credible parallel-trends test. Even two or three pre-treatment waves are substantially better than one.
What if treatment timing is staggered across units?
Match within cohorts sharing the same treatment-onset period. Each cohort's matched controls should be units that are not yet treated at that time. Aggregating cohort-level ATTs by cohort size yields an overall estimate robust to the pitfalls of two-way fixed-effects DiD with staggered adoption.
Which software implements multi-period matching?
In R, the PanelMatch package by Kim et al. directly targets this design. Standard matching packages (MatchIt, Matching) can be adapted for panel settings. In Stata, teffects with panel options and user-written routines support longitudinal matching workflows.
Sources
- 1.Abadie, A. (2005). Semiparametric Difference-in-Differences Estimators. Review of Economic Studies, 72(1), 1-19.
- 2.Imbens, G. W., & Wooldridge, J. M. (2009). Recent Developments in the Econometrics of Program Evaluation. Journal of Economic Literature, 47(1), 5-86.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Multi-period Matching Estimator. ScholarGate. https://scholargate.app/causal-inference/multi-period-matching-estimator