permuted-brunnermunzel 0.1.0

Python implementation of the permuted Brunner-Munzel test for small sample sizes

Permuted Brunner-Munzel Test

A Python implementation of the permuted Brunner-Munzel test, a nonparametric test for comparing two independent samples.

When to Use

The permuted Brunner-Munzel test is best suited for:

• Small sample sizes (7-10 observations per group)
• When standard parametric assumptions don't hold
• Comparing distributions that may differ in shape, not just location

For larger samples (>10 per group), consider using scipy.stats.brunnermunzel.

Installation

pip install permuted_brunnermunzel

Quick Start

from permuted_brunnermunzel import permuted_brunnermunzel

# Sample data
x = [0, 0, 0, 1, 1, 1, 0]
y = [30, 20, 19, 18, 15, 10]

# Run the test
estimate, pvalue = permuted_brunnermunzel(x, y, alternative="less")

print(f"Effect size estimate: {estimate:.4f}")
print(f"P-value: {pvalue:.6f}")

Output:

Effect size estimate: 0.8571
P-value: 0.000583

Parameters

Parameter	Type	Default	Description
x	list	required	First sample observations
y	list	required	Second sample observations
alternative	str	"two_sided"	"two_sided", "greater", or "less"
nan_policy	str	"propagate"	"propagate", "raise", or "omit"
est	str	"original"	"original" or "difference"
force	bool	False	Force test even for large samples

Returns

Value	Description
estimate	Effect size: P(X < Y) + 0.5*P(X = Y) for est="original", or P(X < Y) - P(X > Y) for est="difference"
pvalue	The p-value for the test

Interpreting Results

• estimate = 0.5: No difference between groups
• estimate > 0.5: Values in Y tend to be larger than X
• estimate < 0.5: Values in X tend to be larger than Y
• pvalue < 0.05: Statistically significant difference (at alpha=0.05)

Dependencies

• numpy >=1.20
• scipy >=1.7

References

This is a Python reimplementation of the R package brunnermunzel.

Brunner, E. and Munzel, U. (2000). The nonparametric Behrens-Fisher problem: Asymptotic theory and a small-sample approximation.

License

MIT License - see LICENSE for details.