wrs2-py

A Python port of the R package WRS2, which provides robust statistical methods.

Features

This package currently implements the following methods from WRS2:

• Location Estimators:
  • trim_mean: Trimmed mean.
  • winmean: Winsorized mean.
  • winvar: Winsorized variance.
  • winval: Winsorized values.
  • winvarN: Rescaled Winsorized variance (for standard normal).
  • pbos: One-step percentage bend measure of location.
  • trimse: Standard error of the trimmed mean.
• Correlation:
  • wincor: Winsorized correlation.
  • winall: Winsorized correlation/covariance matrix.
  • pbcor: Percentage bend correlation.
• ANOVA (Trimmed Means):
  • t1way: Heteroscedastic one-way ANOVA.
  • t2way: Two-way ANOVA.
  • rmanova: One-way repeated measures ANOVA.
  • bwtrim: Between-within subjects (split-plot) ANOVA.
• Group Comparison:
  • yuen: Yuen's test for independent groups.
  • yuenbt: Bootstrap Yuen's test for independent groups.

Implementation Equivalence Validation

The implementation has been comprehensively verified against the original R package using simulation data. The following table summarizes the equivalence tests:

*Note: Bootstrap results (effsize, yuenbt p-value) may vary slightly due to random sampling, but are statistically consistent with the R implementation.

Installation

pip install .

Usage

import numpy as np
from wrs2_py import t1way, yuen

# Sample data
g1 = np.random.normal(0, 1, 30)
g2 = np.random.normal(0.5, 1, 30)
g3 = np.random.normal(1, 1, 30)

# One-way ANOVA
res_anova = t1way([g1, g2, g3])
print(res_anova)

# Yuen's test
res_yuen = yuen(g1, g2)
print(res_yuen)

References

• Wilcox, R. R. (2017). Introduction to Robust Estimation and Hypothesis Testing. Academic Press.
• Mair, P., & Wilcox, R. (2020). Robust Statistical Methods in R Using the WRS2 Package. Behavior Research Methods.