GoKawiilThe development bank and others are starting to publish and verify the raw data and code accompanying their research-based reports and papers. Other agencies should do the same.
The World Bank’s headquarters in Washington DC.Credit: J. David Ake/Getty
For years, the social and behavioural sciences have lived with an open secret: findings from studies in these fields can be difficult to reproduce. When independent researchers are unable to arrive at the same result as the original team despite using identical data and the same analysis techniques, this raises doubts about the robustness of the underlying findings. Comprehensive analyses of reproducibility in various fields, published earlier this year, have added to these concerns (O. Miske et al. Nature 652, 126–134 (2026) and A. Brodeur et al. Nature 652, 151–156; 2026).
Reproducibility difficulties are a problem in any field of science. But when studies in social and behavioural sciences cannot be reproduced, there are direct implications for health and well-being.
Take, for example, research published by the World Bank, an international financial institution in Washington DC that has a remit to end poverty and promote prosperity globally. Every year, it provides loans and grants worth roughly US$120 billion, mostly to low- and middle-income countries (LMICs). The bank also employs researchers who produce working papers and reports, which help to inform its policy decisions. And it has a separate department that evaluates the effectiveness of its work to ensure its funds are achieving their aims.
Reliable research in the social and behavioural and sciences
Both its own research staff and external researchers who study the bank’s operations are concerned about reproducibility. Since 2023, the World Bank has been taking concrete steps to increase the reproducibility of its research, and since 2025, all of the findings it publishes have been accompanied by ‘reproducibility packages’. These packages contain reports that document how researchers arrive at their conclusions from original data sets, and include the underlying code, data and other supplementary material.
Each reproducibility package is citable and, where appropriate, contains instructions on how to rerun code to verify studies — a process that is done by a dedicated internal team before certain materials are released publicly. Packages that pass this internal check get a ‘badge’ stating that the research is reproducible, a mark of quality that is helpful to editors when the studies are submitted to journals and sent out for peer review. The publication of materials that aid reproducibility is good practice for any public-facing institution that wants to establish or improve trust by sharing its inner workings.
In June, the World Bank announced that more than 500 reproducibility packages have been released since 2023 (see go.nature.com/4gqjen8). “It took nearly three years to reach the first 500 packages, but we expect the next 500 to arrive much faster,” write the bank’s researchers. It has also launched a reproducible research repository to host the packages (see go.nature.com/4ytc3he).
Beyond the bank
... continue reading