pymatchit-causal 0.5.0

Propensity Score Matching (PSM), Full Matching, Genetic Matching, Cardinality Matching, CEM, and Causal Inference in Python. A port of R's MatchIt.

pymatchit-causal: Propensity Score Matching in Python

Scalable Causal Inference, Propensity Score Matching (PSM), and Coarsened Exact Matching (CEM).

pymatchit-causal is a Python port of the standard R package MatchIt. It allows data scientists to preprocess data for causal inference by balancing covariates between treated and control groups using state-of-the-art matching methods. With the 0.5.0 release, it emphasizes a cohesive, publication-ready visual diagnostic suite.

Why use pymatchit?

If you are looking for Propensity Score Matching in Python, this library provides a robust, "R-style" workflow including:

• Propensity Score Estimation: Logistic Regression (GLM), Random Forest, GBM, Neural Networks.
• Matching Algorithms: Nearest Neighbor (Greedy), Optimal Matching, Exact, Subclassification, and Coarsened Exact Matching (CEM).
• Diagnostics: Publication-ready visual alignment—featuring Love Plots (Covariate Balance), Propensity Density Plots, ECDF plots, and the newly added Jitter Plots for intuitive match verification.

Features

• Matching Methods: Nearest Neighbor, Optimal Matching, Exact, Coarsened Exact Matching (CEM), Subclassification.
• Distance Metrics: Logistic Regression (GLM), Mahalanobis, Random Forest, GBM, Neural Networks, etc.
• Diagnostics: Cohesive diagnostic plots including visually aligned Love Plots, Jitter Plots, and Summary Tables (SMD, Variance Ratios).
• Parity: Designed to mirror the R MatchIt API (matchit(formula, data, method=...)).

---

Installation

pip install pymatchit-causal

Dependencies: numpy, pandas, scipy, statsmodels, matplotlib, scikit-learn, seaborn, patsy.

---

Example Workflow

Scenario: You have a dataset healthcare_data.xlsx with a binary treatment variable took_drug, an outcome recovery_time, and confounders like age, severity, and income.

1. Load Data

import pandas as pd
from pymatchit import MatchIt

# Load your dataset
df = pd.read_excel("healthcare_data.xlsx")

2. Initialize and Match

We will use Nearest Neighbor matching using a Random Forest to estimate the propensity score, applying a caliper to ensure good matches.

# Initialize the matching model
m = MatchIt(
    data=df,
    method='nearest',           # 1:1 Nearest Neighbor matching
    distance='randomforest',    # Use Random Forest for Propensity Scores
    distance_options={'n_estimators': 500},
    caliper={'distance': 0.1, 'age': 2},
    replace=False,
    random_state=42
)

# Fit the model using an R-style formula
m.fit("took_drug ~ age + severity + income + gender")

3. Assess Balance (Diagnostics)

Verify that the treatment and control groups are balanced with cohesive visualization tools.

# 1. Statistical Summary
summary = m.summary()

# 2. Visual Inspection: Love Plot
m.plot(type='balance', threshold=0.1)

# 3. Visual Inspection: Propensity Jitter Plot (New in 0.5.0!)
m.plot(type='jitter')

# 4. Visual Inspection: Propensity Density Overlap
m.plot(type='propensity')

# 5. Visual Inspection: ECDF Plot
m.plot(type='ecdf', variable='age')

4. Extract Matched Data

If balance is satisfactory, extract the data for analysis.

# Get the final dataset containing only matched units
matched_df = m.matches(format='long') 

# Get data with weights and subclass
final_analysis_set = m.matched_data

5. Downstream Inference

Calculate cluster-robust standard errors in your final effect estimation:

import statsmodels.formula.api as smf

model = smf.wls("recovery_time ~ took_drug", data=final_analysis_set, weights=final_analysis_set['weights'])
results = model.fit(cov_type='cluster', cov_kwds={'groups': final_analysis_set['subclass']})
print(results.summary())

---

Citation

If you use pymatchit-causal in your research, please cite it:

Tünnermann, J. (2026). pymatchit: Propensity Score Matching and Causal Inference in Python (Version 0.5.0). Zenodo. https://doi.org/10.5281/zenodo.17839522

BibTeX:

@software{pymatchit_causal,
  author       = {Jonas Tünnermann},
  title        = {pymatchit: Propensity Score Matching and Causal Inference in Python},
  year         = 2026,
  publisher    = {Zenodo},
  version      = {0.5.0},
  doi          = {10.5281/zenodo.17839522},
  url          = {https://doi.org/10.5281/zenodo.17839522}
}