Propensity Score Matching (PSM)

Propensity score matching (PSM) estimates a treatment effect from observational data by pairing treated and untreated units with similar estimated probabilities of receiving treatment given observed covariates.

성향점수매칭(PSM)은 관측된 공변량으로 추정한 처치받을 확률이 비슷한 처치·비처치 단위를 짝지어 관찰자료에서 처치효과를 추정하는 방법이다.

How do I estimate a treatment effect from observational data when groups differ at baseline?

기초 특성이 다른 집단이 있는 관찰자료에서 처치효과를 어떻게 추정하나요?

Propensity score matching pairs treated and untreated units with similar probabilities of treatment based on observed covariates, creating comparable groups before outcomes are compared. It fits observational studies where all important confounders are measured, and covariate balance must be checked after matching.

When to use Propensity Score Matching (PSM)

• A binary treatment, an outcome and pre-treatment covariates in observational data
• Treated and control groups that differ on measured characteristics
• Estimating the average treatment effect on the treated

When to use something else

• Important confounders are unmeasured; use instrumental variables, difference-in-differences or regression discontinuity
• Keeping all observations is preferred; use inverse probability weighting

Assumptions and requirements

• No unmeasured confounding (conditional ignorability)
• Overlap (common support) in propensity scores
• Covariates measured before treatment
• Stable unit treatment value assumption (SUTVA)

What the analysis reports

• Propensity score distribution and common support plot
• Standardized mean differences before and after matching
• Number of matched units
• Average treatment effect on the treated with confidence interval

How to report it

After matching, all standardized mean differences were below [value], and the average treatment effect on the treated was [value], 95% CI [[value], [value]].

Related methods

• Inverse Probability Weighting
• Coarsened Exact Matching
• Difference-in-Differences

Run Propensity Score Matching (PSM) on Quorant