Reproducibility of social science research using aggregate statistics with noise infused for differential privacy.

Approximately 91% of social science regression findings remain statistically significant after applying typical differential privacy noise to aggregate data.

• Why it matters: This work addresses concerns that privacy-preserving noise might undermine the validity of social science research, which is crucial for balancing privacy and scientific integrity.
• What they did: The authors analyzed 93 published social science studies by simulating privacy noise infusion at industry-standard privacy budgets and assessing whether original findings replicated.
• The result: The findings suggest that privacy noise causes smaller distortions than measurement errors and traditional replication discrepancies, enabling most original claims to be maintained under privacy protections.