statsjunk

A collection of statistical odds and ends.

Framework-free Python functions for common statistical tasks — correlation, regression, spatial autocorrelation, electoral fragmentation indices, and (soon) sample size calculations. Every function takes plain arrays and returns a typed Pydantic result model; invalid input raises ValueError rather than failing silently.

Installing

pip install statsjunk

Usage

from statsjunk import compute_pearson_correlation

result = compute_pearson_correlation([1, 2, 3, 4], [2, 4, 6, 8])
print(result.r, result.pvalue)

What's here

One folder per statistical domain, one module per technique inside it. Everything is also re-exported from the top-level statsjunk package, so from statsjunk import compute_pearson_correlation always works regardless of where a function actually lives.

• correlation — pearson: Pearson correlation, with confidence intervals, and a summary-statistics variant (r, n) for when the raw arrays aren't available. spearman: rank correlation, for monotonic-but-not-linear relationships or data with outliers. fisher: the Fisher z-transformation used internally by both, also usable standalone.
• regression — simple: simple linear regression with slope diagnostics. multiple: ordinary least squares multiple linear regression, with standardized coefficients and variance inflation factors.
• spatial — morans_i: global Moran's I spatial autocorrelation.
• elections — laakso_taagepera: effective number of parties/candidates.
• samplesize — a Python port of pmsampsize (binary, continuous, survival), for minimum sample size calculations when developing a prediction model (Riley et al. 2019, 2020).

Developing

uv sync --dev
uv run pytest
uv run ruff check .
uv run ruff format .