Propensity Score Matching
Reduce confounding in observational studies
About Propensity Score Matching
PSM estimates propensity scores (probability of treatment given covariates), then matches treated units to structurally similar control units. This reduces confounding bias in observational studies where randomization is not possible. The matched sample estimates the Average Treatment Effect on the Treated (ATT).
Parameters
Propensity Score Distribution (Mirror Plot)
Treated (57) shown above axis, control (343) below. Overlap region indicates matched region of common support.
Covariate Balance (Love Plot)
SMD < 0.1 indicates good balance. Red = before, green = after matching.
Matching Results
56
98%
Treatment Effect Estimates
ATT = 12.31
ATT = 4.76
5
Covariate Balance Table
| Covariate | SMD (Before) | SMD (After) | Balanced? |
|---|---|---|---|
| age | 0.324 | 0.026 | |
| income | 0.389 | 0.021 | |
| education | 0.278 | 0.032 | |
| bmi | 0.17 | 0.032 |
Interpretation
Before matching, the naive ATT of 12.31 may be biased by confounding (confounding strength = 1). After propensity score matching with caliper = 0.1, 56 pairs were matched. The adjusted ATT is 4.76 (true ATT = 5). Covariate balance improved across all covariates, with SMDs reduced toward the 0.1 threshold — indicating that the matched sample is more comparable between treated and control groups.