MetaPy: an open-source Python tool for transparent, provenance-tracked preparation of heterogeneously reported data and meta-analysis
Before studies can be combined in a meta-analysis, each study's results must be expressed as a common effect size, yet studies report their findings in many ways: means and standard deviations at baseline and follow-up, mean differences with confidence intervals, medians with interquartile ranges, or effect sizes whose design is not stated. Converting these by hand is slow and a well-documented source of error, and most software either performs such conversions without recording how each value was obtained or requires programming. We present MetaPy, a free, open-source Python package with a browser interface. It reads a review team's extraction spreadsheet and derives every obtainable effect size by chaining 180 cited formulas, applying exact formulas first, published approximations only when the researcher accepts them, and assumptions only when switched on. The formula, input values and citation behind every derived number are recorded. MetaPy checks reported standardized mean differences against raw data in the same row to infer their design, asks for more data only when nothing can be derived, and provides pairwise, rare-event, dependent-effects, network, publication-bias and Bayesian meta-analyses with an auditable report. Derived effect sizes agreed with an independent implementation to within rounding error, and 250 results from 21 analyses agreed with the R packages metafor, clubSandwich, netmeta and bayesmeta, nearly all to four or more significant digits. MetaPy makes the preparation stage of a meta-analysis inspectable by reviewers and editors as well as authors, and its design could be reused by evidence-synthesis tools in any discipline.
Authors
- Sadia Nadeem (ORCID: https://orcid.org/0009-0008-5510-9456)
Publication Details
- Journal
- MetArXiv (OSF Preprints)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.31222/osf.io/6r24v
- Primary Topic
- Meta-analysis and systematic reviews
- Type
- preprint