psmpy - Propensity Score Matching for Python

0.3.16 verified Sat May 09 auth: no python

psmpy provides propensity score matching for observational studies, including functions to compute propensity scores, perform matching (with/without replacement, caliper), and generate graphical plots (balancing, histogram). Version 0.3.16 is the latest release. Release cadence is low; last update was 2024.

pip install psmpy

Common errors

error TypeError: __init__() got an unexpected keyword argument 'idcol' ↓

cause Parameter 'idcol' renamed to 'indx' in version 0.3.16.

fix

Use 'indx' parameter instead of 'idcol'.

error AttributeError: module 'psmpy' has no attribute 'Psmpy' ↓

cause Wrong import statement; psmpy is a package, not a module.

fix

Use 'from psmpy import Psmpy'.

Warnings

breaking In version 0.3.15 and earlier, the 'indx' parameter was called 'idcol'. It was renamed in 0.3.16. Using 'idcol' raises TypeError. ↓

fix Use 'indx' instead of 'idcol' when initializing Psmpy.

gotcha The 'logistic_model' parameter expects a scikit-learn LogisticRegression instance with solver='liblinear' (or similar) that supports predict_proba. Using default solver may raise warnings or errors on small datasets. ↓

fix Explicitly instantiate LogisticRegression(solver='liblinear').

gotcha The 'exclude_cols' parameter should include any columns not used for propensity score estimation (e.g., outcome variables, IDs). Forgetting to exclude the treatment column can cause errors. ↓

fix List all columns that are not predictors in 'exclude_cols'.

Imports

Psmpy
wrong
```
import psmpy
```
correct
```
from psmpy import Psmpy
```
Direct import fails; must import the class explicitly.
Psmpy
wrong
```
from psmpy.psmpy import Psmpy
```
correct
```
from psmpy import Psmpy
```
Older documentation suggested submodule path; top-level import works.

Quickstart

Basic workflow: initialize with data, compute scores, match, and inspect matches.

import pandas as pd
from psmpy import Psmpy
from sklearn.linear_model import LogisticRegression

# Sample data
df = pd.DataFrame({
    'treatment': [0, 1, 0, 1, 0, 1],
    'age': [30, 40, 35, 45, 25, 50],
    'income': [50000, 60000, 55000, 65000, 45000, 70000]
})

# Initialize Psmpy
psm = Psmpy(data=df, treatment='treatment', indx='age',
            exclude_cols=['income'],
            logistic_model=LogisticRegression(solver='liblinear'))

# Compute propensity scores
psm.pscore()

# Perform matching
psm.match(method='nearest', caliper=None, replace=False)

# View matched pairs
print(psm.matched_ids)