psmpy

What it is and what it does

PsmPy is a Python library for propensity score matching, a statistical technique used in observational studies to estimate causal effects by matching treated and control subjects on their propensity scores—the predicted probability of receiving treatment given observed covariates. It reduces confounding bias by creating comparable treatment and control groups, mimicking the balance achieved in randomized trials.

The package provides logistic regression to compute propensity scores, KNN-based matching algorithms supporting both 1:1 and 1:many matching with optional caliper constraints, and visualization tools to assess covariate balance before and after matching. It integrates with pandas, scikit-learn, and matplotlib, and includes effect size calculations (Cohen's D) to quantify the standardized mean differences between groups.

Use it for:

• Estimate treatment effects in retrospective cohort studies where randomization is infeasible or unethical
• Assess whether an intervention (e.g., a medical procedure) causally affects outcomes (e.g., mortality, length of stay) in observational data
• Balance covariates between treated and control groups to reduce confounding in epidemiological analyses
• Visualize covariate imbalance before and after matching to validate the matching procedure
• Compare multiple matching strategies (1:1 vs. 1:many, with/without replacement) on the same dataset

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

PsmPy implements propensity score matching for observational studies, using logistic regression and KNN algorithms to match treated and control subjects on covariates and estimate causal effects.

Yes, if you are conducting observational studies in epidemiology or health research and need propensity score matching. The package is straightforward to use, well-integrated with the scientific Python stack, and permissively licensed. However, maintenance is aging (last release 275 days ago), so consider it stable for established workflows but not actively developed; for active support or cutting-edge features, evaluate alternatives or contribute to the repository.

Install

psmpy on PyPI

pip install psmpy

uv add psmpy

poetry add psmpy

Installing psmpy

Before you install

Low friction: pure Python wheel with six standard scientific dependencies (matplotlib, numpy, pandas, seaborn, scikit-learn, scipy). Maintenance status is aging—last release 275 days ago, 62 repository stars, no recent activity—so expect slower response to issues.

License in practice

MIT license (permissive) means you can use, modify, and distribute PsmPy freely in commercial and private projects with minimal restrictions.

Quickstart

pip install psmpy

from psmpy import PsmPy
import pandas as pd

data = pd.read_csv('your_data.csv')
psm = PsmPy(data, treatment='treatment', indx='pat_id')
psm.logistic_ps(balance=True)
psm.kdtree_matched(matcher='propensity_logit', replacement=False)
matched_df = psm.df_matched

Verify before relying

• Whether the package handles missing data or requires complete-case analysis
• Performance characteristics with large datasets (sample size limits or memory requirements)
• Whether Python version support is truly unspecified or has implicit constraints

Package facts

Evidence: psmpy-0.3.16-py3-none-any.whl

Keywords: propensity score matching, statistics, plotting

Tags

More Scientific/Engineering packages