When and why to use overlap weighting: clarifying its role, assumptions, and estimand in real-world studies

Abstract

Objectives

To examine the strengths and limitations of overlap weighting in observational studies and to clarify when it is appropriate to use this method based on the target estimand.

Study Design and Setting

This is a narrative commentary that reviews recent methodological developments and real-world examples to highlight how overlap weighting operates, when it provides advantages over methods like inverse probability of treatment weighting, and the importance of aligning analytic methods with the causal question and estimand.

Results

Overlap weighting produces bounded, stable weights and achieves exact mean covariate balance in the subset of patients with overlapping treatment probabilities near 0.5—those considered to be in clinical equipoise. However, it targets the average treatment effect in the overlap population (ATO), a statistically defined subgroup that is difficult to characterize clinically. Use of this method without prespecifying interest in the ATO may lead to misinterpretation of results. While overlap weighting improves statistical performance, it limits generalizability and interpretability. Study design and inclusion/exclusion criteria remain critical for addressing violations of positivity.

Conclusion

Overlap weighting is most appropriate when the research question explicitly targets the overlap population. It should not be adopted solely to resolve estimation issues with average treatment effect or average treatment effect in the treated methods. Researchers must define their target estimand before choosing a method and clearly report the characteristics of both the unweighted and overlap-weighted populations to ensure valid causal inference.

Plain Language Summary

Overlap weighting is a statistical method used in health research to compare treatments when people are not randomly assigned to different options. It focuses on patients who could realistically receive either treatment and helps improve the fairness and precision of comparisons. However, the results apply only to this specific group and not everyone in the study. Researchers should choose this method only when it fits the question they are asking.