Policy Evaluation Propensity Score Matching
Propensity Score Matching for Policy Evaluation · Also known as: PSM policy evaluation, policy PSM, propensity matching for program evaluation, PSM treatment evaluation
Policy evaluation propensity score matching applies the propensity score framework — originally developed by Rosenbaum and Rubin (1983) and operationalized for program evaluation by Heckman et al. (1997) — to estimate the causal effect of a policy intervention. It constructs a credible comparison group from non-participants by matching them to participants on their estimated probability of receiving the treatment, enabling unbiased effect estimation without random assignment.
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 policy evaluation PSM when you have observational data on programme participants and a pool of non-participants with rich pre-treatment covariates, and the policy was not randomly assigned. It is appropriate for labour market programmes, health interventions, educational policies, and social transfer evaluations where randomisation was infeasible. The key requirement is that assignment to the policy is ignorable conditional on observed covariates (the unconfoundedness assumption). Do NOT use PSM when important confounders are unobserved (e.g., motivation, ability) and no instrument or proxy is available; when the treated and control populations have non-overlapping covariate distributions (lack of common support); or when the outcome data were collected only after treatment, with no pre-treatment covariates to match on.
Strengths & limitations
- Eliminates selection bias on observed characteristics by constructing a comparable control group without requiring a regression model of the outcome.
- Transparent and interpretable: the propensity score summarises covariate balance and matching quality in a single diagnostic.
- Flexible matching algorithms (nearest-neighbour, caliper, kernel) allow researchers to trade off bias and efficiency based on data structure.
- Widely accepted in policy evaluation, development economics, and public health, with well-developed software implementations.
- Can be combined with difference-in-differences to control for time-invariant unobservables in addition to observed selection.
- Only removes bias from observed confounders; unobserved selection (e.g., self-selection on motivation) is not addressed and can invalidate causal claims.
- Requires common support: if treated and control units occupy different covariate regions, matching discards treated units or relies on poor matches.
- The unconfoundedness assumption is untestable and must be defended with domain knowledge and sensitivity analyses.
- Efficiency loss: matched samples are smaller than the original data, reducing statistical power compared to regression-based approaches.
- Results can be sensitive to propensity score model specification and the choice of matching algorithm.
Frequently asked
What is the propensity score and why does matching on it balance covariates?
The propensity score is the conditional probability of receiving the treatment given observed covariates. Rosenbaum and Rubin (1983) proved that if treatment assignment is ignorable given covariates X, it is also ignorable given the scalar propensity score p(X). Matching on p(X) therefore balances the full covariate distribution between treated and matched controls, without requiring exact matching on every variable.
What is the unconfoundedness assumption and when is it plausible for policies?
Unconfoundedness (also called conditional independence or selection on observables) states that, after conditioning on observed pre-treatment covariates, treatment assignment is independent of potential outcomes. It is plausible when researchers observe the key determinants of both programme participation and the outcome — such as baseline income, prior employment, and demographics in a labour programme evaluation. It is implausible when self-selection on unobservable motivation or ability is the main driver of participation.
How do I choose between nearest-neighbour, caliper, and kernel matching?
Nearest-neighbour matching minimises bias by using the closest control but can include poor matches when controls are scarce. Adding a caliper (e.g., 0.25 standard deviations of the log-odds score) excludes bad matches at the cost of dropping some treated units. Kernel matching uses all controls weighted by distance, improving efficiency but averaging in more dissimilar units. Check balance diagnostics for each and prefer the algorithm that achieves the best balance on substantively important covariates.
Can I combine PSM with difference-in-differences?
Yes. Matching on pre-treatment covariates to select a comparison group and then applying DiD on the matched sample is a doubly robust strategy. It removes selection on observed covariates via matching and removes time-invariant unobserved confounders via the before-after differencing, making the causal estimate more credible than either method alone.
How do I assess sensitivity to hidden bias?
Rosenbaum bounds compute the maximum amount of unmeasured confounding (expressed as the odds ratio of treatment assignment for two identical-looking units) that would render the estimated treatment effect statistically insignificant. A large bound indicates the findings are robust; a small bound signals that even mild hidden bias could explain the result.
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 ↗
- Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme. Review of Economic Studies, 64(4), 605-654. DOI: 10.2307/2971733 ↗
How to cite this page
ScholarGate. (2026, June 3). Propensity Score Matching for Policy Evaluation. ScholarGate. https://scholargate.app/en/causal-inference/policy-evaluation-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.
- Coarsened Exact MatchingCausal inference↔ compare
- Difference-in-DifferencesEconometrics↔ compare
- Doubly Robust EstimationCausal inference↔ compare
- Inverse Probability WeightingCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Propensity Score WeightingCausal inference↔ compare