Panel Data Propensity Score Matching
Propensity Score Matching with Panel Data · Also known as: PSM with panel data, longitudinal PSM, panel PSM, difference-in-differences PSM
Panel data propensity score matching combines the bias-reduction of PSM with the longitudinal structure of panel data, enabling causal estimation of treatment effects by matching treated and control units on observable pre-treatment characteristics and then differencing within matched pairs over time. Developed in the framework of Heckman, Ichimura, and Todd (1998), it is especially valuable when randomisation is infeasible and both selection on observables and time-varying confounding must be addressed simultaneously.
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 panel data propensity score matching when you have longitudinal observational data, treatment is non-random but selection is plausibly driven by observed pre-treatment characteristics, and you want to control for both selection on observables and time-invariant unit heterogeneity. It is well-suited to programme evaluation, labour economics, and health studies where participants self-select and repeated measurements are available. Do not use it when key confounders are unobserved and time-invariant (fixed-effects alone would be preferred), when the panel is too short to capture meaningful pre-treatment trends, when the overlap between treated and control propensity scores is negligible, or when the treatment mechanism is genuinely random (an RCT needs no matching).
Strengths & limitations
- Controls for observable selection bias through propensity score matching and for time-invariant unobservable differences through within-pair differencing.
- Non-parametric in spirit: does not impose a functional form on how covariates affect the outcome.
- Diagnostic transparency — covariate balance tables and overlap plots make it easy to see whether matching succeeded.
- The ATT estimate is directly policy-relevant, targeting the effect on those who actually received treatment.
- Combines two well-established quasi-experimental strategies, making the approach credible to reviewers familiar with either method.
- Rests on the strong ignorability assumption: all confounders must be observed and included in the propensity score model; hidden bias from unobserved time-varying confounders cannot be removed.
- Requires adequate common support — if treated and control units have very different covariate distributions, few or poor matches result.
- Bootstrapped standard errors can be computationally intensive for large panels.
- Dropping unmatched units reduces sample size and can limit external validity.
- Pre-treatment period must be long enough to support reliable before-after differencing within matched pairs.
Frequently asked
How is panel data PSM different from standard cross-sectional PSM?
Standard PSM matches units on covariates and compares outcomes at a single point in time. Panel data PSM adds a longitudinal step: once matched, the estimator uses the before-after change within each matched pair, removing any time-invariant unobserved differences that ordinary PSM cannot touch.
Does panel data PSM require the parallel-trends assumption like DiD?
Yes. When the within-pair differencing step mimics a DiD estimator, the implicit assumption is that matched treated and control units would have followed the same trend absent treatment. Verifying pre-treatment trend similarity in the matched sample is therefore important.
Which matching algorithm should I choose?
Nearest-neighbour matching with replacement and a tight caliper (e.g., 0.2 of the standard deviation of the log-odds) is a common default. Kernel or radius matching can recover more control observations and reduce variance at the cost of some bias. The choice should be guided by balance achieved and sample size.
What if balance is still poor after matching?
Re-specify the propensity score model (add interactions, polynomials, or additional covariates), tighten the caliper, or switch to an entropy balancing or inverse probability weighting approach which directly targets covariate balance rather than estimating a score model.
How many pre-treatment periods do I need?
At least one pre-treatment period is required for within-pair differencing, but two or more pre-treatment periods are strongly preferred so you can verify that matched units trended similarly before treatment, supporting the parallel-trends assumption.
Sources
- Heckman, J. J., Ichimura, H., & Todd, P. (1998). Matching as an Econometric Evaluation Estimator. Review of Economic Studies, 65(2), 261-294. DOI: 10.1111/1467-937X.00044 ↗
- Caliendo, M., & Kopeinig, S. (2008). Some Practical Guidance for the Implementation of Propensity Score Matching. Journal of Economic Surveys, 22(1), 31-72. DOI: 10.1111/j.1467-6419.2007.00527.x ↗
How to cite this page
ScholarGate. (2026, June 3). Propensity Score Matching with Panel Data. ScholarGate. https://scholargate.app/en/causal-inference/panel-data-propensity-score-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
- Matching EstimatorCausal inference↔ compare
- Panel Data Difference-in-DifferencesCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Propensity Score MatchingResearch Statistics↔ compare