StatPilot 🧭

Automated, transparent statistical analysis for researchers.

StatPilot picks the right statistical test for your data, explains why it chose it, and generates a publication-ready report — all from a single function call.

The problem

Running statistics correctly involves a sequence of decisions most researchers make inconsistently:

1. Check normality — but which test? Shapiro-Wilk? Visual inspection?
2. Check variance homogeneity — but only when normality holds?
3. Pick the right test — but which of t-test, Welch, Mann-Whitney, ANOVA, Kruskal-Wallis?
4. Calculate effect size — but Cohen's d, eta-squared, or rank-biserial?
5. Write it up — in a reproducible, auditable way.

StatPilot automates this entire chain with a transparent decision engine that shows its work.

Quick start

pip install statpilot

import pandas as pd
from statpilot import compare

df = pd.read_csv("my_data.csv")

result = compare(df, target="score", group="treatment")

result.summary()  # prints a rich table to the terminal
result.plot()  # shows a boxplot + distribution
result.to_report()  # returns a Markdown string ready to paste into your paper

What the output looks like

┌─────────────────────────────────────────────────────┐
│               StatPilot Result                      │
├─────────────────────────────────────────────────────┤
│ Test selected:   Independent samples t-test         │
│ Statistic:       t = 4.21                           │
│ p-value:         0.0003 ***                         │
│ Effect size:     Cohen's d = 0.87 (large)           │
├─────────────────────────────────────────────────────┤
│ Why this test?                                      │
│  • 2 independent groups detected                    │
│  • Normality: passed (Shapiro-Wilk, α=0.05)        │
│  • Variance equality: passed (Levene, α=0.05)      │
│  → Independent samples t-test is appropriate       │
└─────────────────────────────────────────────────────┘

CLI usage

# Compare two groups from a CSV file
statpilot compare --data my_data.csv --target score --group treatment

# Save a Markdown report
statpilot compare --data my_data.csv --target score --group treatment --report report.md

# Paired comparison
statpilot compare --data my_data.csv --target score --group condition --paired

Supported tests (v0.1)

Why not just use pingouin or statsmodels?

pingouin and statsmodels are excellent libraries — StatPilot uses them under the hood. The difference is the automated decision layer: with pingouin, you still choose which function to call. StatPilot runs the assumption checks and makes that choice for you, and documents the reasoning in the output.

Installation for development

git clone https://github.com/your-org/statpilot.git
cd statpilot
pip install -e ".[dev]"
pytest

Documentation

Full documentation at statpilot.readthedocs.io — including the decision engine concept guide, API reference, and example notebooks.

Contributing

See CONTRIBUTING.md. Bug reports and feature requests welcome via GitHub Issues.

Citation

If you use StatPilot in your research, please cite it. A DOI is available via Zenodo after each tagged release.

License

MIT — free to use in academic and commercial projects.