ritest

Overview

ritest provides fast randomization inference (RI) tools for linear models and arbitrary statistics. It supports weights as well as stratified and clustered designs. Reports coefficient confidence interval by default.

📑 Documentation: https://tabarecapitan.com/projects/ritest.

Features

• Linear-model RI with efficient computation.
• Generic RI for arbitrary scalar statistics via stat_fn.
• Ultra-fast coefficient bounds and bands.
• Stratified, clustered, and stratified-clustered designs.
• Weighted least squares (WLS) support.
• Deterministic seeding, reproducible permutations.

Installation

From PyPI:

pip install ritest-python

Optional extras:

pip install ritest-python[plot]      # plotting support
pip install ritest-python[numba]     # Numba acceleration

Quickstart

Linear model (formula interface)

import pandas as pd
from ritest import ritest

df = pd.DataFrame({
    "y": [1, 2, 3, 4],
    "treat": [0, 1, 0, 1],
    "x": [5, 6, 7, 8],
})

res = ritest(
    df=df,
    permute_var="treat",
    formula="y ~ treat + x",
    stat="treat",
    reps=1000,
)

print(res.summary())

Custom statistic (stat_fn)

def my_stat(df):
    return df["y"].corr(df["treat"])

res = ritest(
    df=df,
    permute_var="treat",
    stat_fn=my_stat,
    reps=1000,
)

print(res.pvalue)

Citation

A software citation entry will be provided once the package reaches its first stable release and the accompanying paper/notes are finalized. For now, cite the GitHub repository:

Tabaré Capitán (2025). ritest: Randomization inference in Python.
https://github.com/tabareCapitan/ritest

Disclaimer

Use this software at your own risk. I make no guarantees of correctness or fitness for any purpose. I use it in my own work, but you should review the code to ensure it meets your needs. If you find an issue, please report it.

Contributing and issues

Bug reports and feature requests can be filed at:

https://github.com/tabareCapitan/ritest/issues

Pull requests should be focused and include tests for new behavior.

License

MIT License. See the LICENSE file for full terms.