PsmPy

PsmPy implements propensity score matching in Python to match treated and untreated subjects based on observed covariates for causal effect estimation in retrospective cohort studies.

Key Features:

• Logistic Regression-Based Propensity Score Calculation: Uses logistic regression to compute propensity scores representing the probability of treatment assignment conditional on observed covariates.
• K-Nearest Neighbors (k-NN) Matching Algorithm: Performs k-NN matching on propensity scores to pair treated and untreated cases with similar covariate profiles.
• Plotting and Visualization: Provides plotting capabilities for assessing propensity score distributions and post-matching balance.

Scientific Applications:

• Causal Effect Estimation in Retrospective Cohort Studies: Estimates treatment effects by matching treated and untreated subjects to approximate randomized comparisons.
• Confounding Control in Observational Studies: Reduces bias from observed confounders when estimating treatment effects through propensity score matching.

Methodology:

Computes propensity scores via logistic regression, applies k-nearest neighbors matching on propensity scores, and benchmarks performance against the R package MatchIt using a Mann–Whitney test (U=49, p<0.0001) reporting an average 10-fold improvement in reduction of residual effect sizes among covariates.