metafopy

Meta-analysis in Python: a re-implementation of the R package metafor.

Status: work in progress.

Documentation

Example

import numpy as np
from metafopy import datasets, rma, confint, predict

dat = datasets.load('dat.hackshaw1998')  # already has yi / vi columns
res = rma(data=dat, method='REML')
print(res)
print(confint(res))
print(predict(res, transf=np.exp))

Building

Requires Python >= 3.14.7, < 3.15.

git clone https://github.com/bnomis/metafopy
cd metafopy
uv sync
source bin/activate.sh
pytest --mpl   # --mpl also compares the plots with the baseline images (Linux only)
mypy
ruff check .

Building the docs

The user manual is in docs/ (installation, tutorial, user guide, worked analyses, a guide for users coming from R and the function reference).

Build and browse it with:

uv pip install -r docs/requirements.txt
mkdocs serve

pytest src/tests/test_docs_examples.py runs every Python example in the manual.

Validation against R

R metafor is the numerical reference. src/tests/r/ holds the R scripts that generate JSON fixtures in src/tests/reference/ (currently metafor 5.2-1, metadat 1.6-0). The fixtures are committed, so the normal test run needs no R.

To regenerate them:

Rscript src/tests/r/export_datasets.R     # refresh bundled datasets
Rscript src/tests/r/generate_fixtures.R   # refresh src/tests/reference/*.json

or reproducibly with the pinned image in src/tests/r/Dockerfile:

docker build -t metafopy-r src/tests/r
docker run --rm -v "$PWD":/work -w /work metafopy-r Rscript src/tests/r/generate_fixtures.R

License

GPL-2.0-or-later, the same as metafor (GPL >= 2).

The bundled datasets come from the R package metadat (GPL >= 2).