Robust Propensity Score Weighting
Robust Propensity Score Weighting Estimator · Also known as: robust PSW, robust IPW, robustness-augmented propensity score weighting, misspecification-robust weighting
Robust Propensity Score Weighting extends standard inverse probability weighting by incorporating safeguards against misspecification of the propensity score model and extreme weights. It combines techniques such as weight trimming, overlap weighting, or augmented outcome models to ensure that causal effect estimates remain reliable even when the propensity score model is imperfectly specified.
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 Robust Propensity Score Weighting when you suspect that the propensity score model may be imperfectly specified, or when diagnostic checks reveal extreme weights that inflate variance or bias estimates. It is appropriate with observational cross-sectional or panel data, a continuous or binary outcome, and sufficient sample size to support bootstrap inference. Do not use it as a substitute for addressing fundamental lack of overlap between treatment and control groups — when covariate distributions barely overlap, robust weighting can mitigate but cannot eliminate the problem, and a different design may be needed.
Strengths & limitations
- Protects against propensity score model misspecification by making the estimate less sensitive to incorrect functional-form assumptions.
- Overlap weights and trimming reduce variance inflation caused by extreme weights, improving finite-sample performance.
- Doubly robust augmentation ensures consistency if either the propensity model or the outcome model — but not both — is correctly specified.
- Sensitivity analysis tools provide explicit quantification of how robust the conclusion is to unmeasured confounding.
- Retains the interpretability of a weighted average treatment effect while adding protective layers against common failure modes.
- Trimming or overlap weighting changes the target population, so the estimand may no longer be the ATE or ATT for the full sample — the estimand shift must be reported.
- Doubly robust variants require specifying and fitting a second (outcome) model, increasing model complexity and computational cost.
- Even robust weighting cannot correct for unmeasured confounding; sensitivity analyses bound but do not eliminate this threat.
- Selecting trimming thresholds or weighting schemes involves researcher discretion, which can introduce specification searching if not pre-registered.
Frequently asked
What makes propensity score weighting 'robust'?
Robustness is achieved through one or more strategies: trimming or capping extreme weights to limit the influence of outlier observations; using overlap weights that focus on units with intermediate propensity scores where estimation is most reliable; or augmenting the IPW estimator with an outcome model so that misspecification of either model alone does not destroy consistency.
How does robust PSW differ from doubly robust estimation?
Doubly robust estimation is one type of robust PSW that augments the inverse probability weighted outcome with an outcome model. Robust PSW is the broader family that also includes trimming-based and overlap-weighting approaches that do not require an outcome model but achieve robustness through weight modification instead.
When should I trim weights versus use overlap weights?
Trimming is a simple fix for occasional extreme weights and preserves most of the original sample. Overlap weights are preferable when a substantial fraction of units have propensity scores near 0 or 1, as they smoothly down-weight poor-overlap units and define a clearly interpretable estimand (the overlap population). If the overlap population is too narrow relative to the policy question, neither approach is adequate.
Does robust PSW remove the need for a parallel-trends or unconfoundedness assumption?
No. Robust PSW still requires the selection-on-observables (unconfoundedness) assumption — that treatment assignment is independent of potential outcomes conditional on observed covariates. Robustness refers to resilience against propensity model misspecification, not to unmeasured confounding. A sensitivity analysis assesses the latter.
How large a sample do I need?
The same minimum as standard PSW — at least 40 observations is a rough lower bound — but doubly robust or bootstrap-based inference works best with larger samples (n > 200) because it estimates two models and requires stable bootstrap resampling distributions.
Sources
- Robins, J. M., Rotnitzky, A., & Zhao, L. P. (1994). Estimation of regression coefficients when some regressors are not always observed. Journal of the American Statistical Association, 89(427), 846-866. DOI: 10.1080/01621459.1994.10476818 ↗
- Zhao, Q., Small, D. S., & Bhattacharya, B. B. (2019). Sensitivity analysis for inverse probability weighting estimators via the percentile bootstrap. Journal of the Royal Statistical Society: Series B, 81(4), 735-761. DOI: 10.1111/rssb.12327 ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Propensity Score Weighting Estimator. ScholarGate. https://scholargate.app/en/causal-inference/robust-propensity-score-weighting
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.
- Doubly Robust EstimationCausal inference↔ compare
- Inverse Probability WeightingCausal inference↔ compare
- Marginal Structural ModelCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Propensity Score WeightingCausal inference↔ compare
- Sensitivity Analysis for CausalityCausal inference↔ compare