Proceedings of the National Academy of Sciences · 9 October 2026 · From issue 49
Reproducibility of social science research using aggregate statistics with noise infused for differential privacy
Privacy noise added to data kept 91% of original study findings intact.
Ryan Steed, Eduardo Abraham Schnadower Mustri et al.
Many companies and agencies add random statistical noise to data to protect people's privacy. Researchers tested 93 published social science studies to see if their main results still held up after this noise was added. About 91% of the findings still supported the original conclusions. Results based on weaker original effects were more likely to change or flip. The noise caused smaller problems than normal errors that already happen in this type of research.
The share of 93 study findings that still held up after privacy noise was added to the data.
93 studies only, using one type of statistical method.
Integrity screen: passed (3 checks) Checked 12 October 2026. Retraction record: none. DOI resolves at doi.org. Metadata record found (Proceedings of the National Academy of Sciences). Read the source
This screen checks process signals. It does not tell you the findings are true.