StatForge

Automated Bayesian-frequentist statistics and publication-ready reports.

StatForge is an open-source Python library and command-line interface designed to automate statistical analysis and generate publication-ready reports. Built for academic researchers, biostatisticians, and data scientists, StatForge streamlines the process from raw data ingestion to formatted output (PDF, DOCX, HTML).

Overview

StatForge implements a robust six-stage execution pipeline:

1. DataLoader: Ingests data from CSV, Excel, SPSS (.sav), and Parquet formats.
2. AssumptionChecker: Performs statistical assumption checks (e.g., normality, homoscedasticity) utilizing a SHA-256 keyed caching layer (joblib.Memory) for optimized iterative checks.
3. MethodSelector: Automatically ranks and selects appropriate tests based on data characteristics and assumption results.
4. ModelFitter: Dispatches analysis to a plugin registry supporting both frequentist methods (SciPy, statsmodels) and Bayesian inference (PyMC).
5. ResultFormatter: Structures statistical output including effect sizes for standardized reporting.
6. ReportBuilder: Orchestrates the final document utilizing Jinja2 templates, generating APA or Vancouver styled tables, automated methods summaries, and figure captions.

Installation

pip install statforge

Quick Start

1. Interactive CLI Wizard

The easiest way to begin an analysis is via the interactive wizard. Navigate to your dataset and execute:

statforge run dataset.csv

The wizard will prompt you to:

• Select the outcome variable.
• Select grouping or predictor variables.
• Choose a report style (e.g., APA7).

2. Validating Data Quality

Before running a full analysis, generate a data quality report to flag missing values, outliers, or type mismatches:

statforge validate dataset.csv

3. Generating a Configuration File

For reproducible analyses, generate a configuration scaffold:

statforge config

This creates a statforge_config.yaml file that you can customize and version control.

Bayesian Analysis & PriorAdvisor

StatForge lowers the barrier to Bayesian analysis through its PriorAdvisor module.

• Guided Priors: PriorAdvisor suggests data-driven, weakly informative priors (e.g., assigning a Normal distribution with $\mu$ equal to the observed mean and $\sigma$ equal to twice the observed standard deviation).
• Transparency: The rationale for the selected priors is clearly documented and included in the generated report's methodology section.
• Sensitivity Analysis: The pipeline automatically evaluates posterior stability across weakly informative, uninformative, and highly informative prior variants to ensure robustness.

Model Plugin Registry

StatForge utilizes a @register decorator pattern, allowing seamless integration of custom analytical models. Users can drop custom .py model definitions directly into ~/.statforge/plugins/, and they will be dynamically loaded by the pipeline. See CONTRIBUTING.md for details on writing custom plugins.

Cite StatForge

If you use StatForge in your research, please cite our JOSS paper (DOI pending). See paper/paper.md and paper/paper.bib for citation details.

Made by Samvardhan Singh. Licensed under the Apache License 2.0.