Propensity Score Matching in Education Research
Propensity Score Matching Applied to Education Research · Also known as: PSM in education, educational PSM, PSM for program evaluation in schools, propensity matching education
Propensity Score Matching (PSM) in education research is a quasi-experimental technique that creates comparable treatment and control groups from observational student, teacher, or school data. By balancing groups on observed background characteristics, it enables credible causal estimates of educational interventions — such as tutoring programs, school choice policies, or teacher professional development — when random assignment is infeasible.
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 PSM in education research when you have a sufficiently large observational dataset with rich pre-treatment covariates and need to compare a program or intervention group against a non-participating group. It suits cross-sectional studies of school programs, teacher training effects, or scholarship receipt when randomisation was not possible. Do not use PSM when key confounders are unobserved (e.g., unmeasured motivation or ability), when the sample is small (fewer than roughly 100 per group), when there is insufficient overlap between treated and control propensity-score distributions, or when a stronger design such as a regression discontinuity or difference-in-differences is available.
Strengths & limitations
- Reduces selection bias from observed confounders without requiring a randomised experiment, making it applicable to existing administrative and survey datasets.
- Transparent covariate balance checks (SMD) allow researchers and reviewers to verify comparability of groups before reporting estimates.
- Flexible: compatible with many matching algorithms, and can be combined with regression adjustment (doubly robust) for additional precision.
- Well-accepted by education policy audiences and What Works Clearinghouse review standards when assumptions are documented.
- Preserves ecological validity since it uses real-world program participants rather than an artificially constrained experiment.
- Only controls for observed confounders; unmeasured variables such as student motivation, parental involvement quality, or teacher enthusiasm remain sources of bias.
- Requires sufficient common support — if treated and control propensity-score distributions barely overlap, matched samples may be small and unrepresentative.
- Results are sensitive to covariate specification: omitting an important predictor of treatment assignment can invalidate the causal interpretation.
- Matching discards unmatched control units, potentially reducing statistical power compared to full-sample regression.
- Does not address selection on gains — students who anticipate they will benefit most may self-select, a pattern matching cannot fully eliminate.
Frequently asked
How is PSM different from simply controlling for covariates in a regression?
Both approaches adjust for confounders, but PSM explicitly restricts comparisons to matched units with similar propensity scores, enforcing overlap and reducing model dependence. Regression extrapolates across the full covariate space, which can be misleading when treated and control groups barely overlap on key variables.
What covariates should I include in the propensity model for an education study?
Include all pre-treatment variables that predict both program participation and the outcome: prior test scores, demographics (race, gender, free lunch eligibility), prior attendance and behavior, and school-level characteristics. Avoid including post-treatment variables, which would introduce bias.
How do I know if balance is good enough after matching?
The standard benchmark is a standardized mean difference (SMD) below 0.10 for all covariates. Graphical checks such as love plots visually confirm balance. If balance is poor for any key variable, try a tighter caliper, kernel matching, or add the imbalanced covariate as a regression covariate on the matched sample.
Can PSM be used with school-level data instead of student-level data?
Yes. PSM can be applied at any unit of analysis — student, teacher, classroom, or school. However, at the school level sample sizes are often small, which limits statistical power and increases matching variance; more schools are needed for reliable estimates.
What if I suspect there is unobserved confounding?
Run a sensitivity analysis (e.g., Rosenbaum bounds) to assess how strong unmeasured confounding would need to be to overturn your conclusion. If the estimate is fragile to small amounts of hidden bias, report it with caution and consider supplementing with a stronger design such as an instrumental variable or regression discontinuity.
Sources
- Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41-55. DOI: 10.1093/biomet/70.1.41 ↗
- Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin. ISBN: 978-0395615560
How to cite this page
ScholarGate. (2026, June 3). Propensity Score Matching Applied to Education Research. ScholarGate. https://scholargate.app/en/causal-inference/propensity-score-matching-in-education-research
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
- Inverse Probability WeightingCausal inference↔ compare
- Matching EstimatorCausal inference↔ compare
- Propensity Score WeightingCausal inference↔ compare