Advancing Statistical Practice: Inference and Privacy in Synthetic Data

As privacy concerns rise, synthetic data offer a promising solution for protecting confidential information while facilitating analysis. However, ensuring valid statistical inference from synthetic data remains challenging.

At the 2025 Joint Statistical Meetings (JSM), a roundtable discussion brought together participants from various organizations to explore best practices for reducing bias and accurately estimating uncertainty in synthetic datasets. They shared insights, challenges, and priorities regarding the development of new methods and tools as synthetic data use expands.

Westat’s Tom Krenzke, MS, Vice President for Statistics and Data Science, led the roundtable, which was organized by Minsun Riddles, PhD, a Westat Principal Statistical Associate. Building on the discussion, Krenzke and Riddles wrote the proceedings paper “Inference Using Synthetic Data: Balancing Privacy, Bias, and Variance in Modern Statistical Practice.” The article, now available for download, concludes: “As the field continues to evolve, collaboration across disciplines will be essential to refine tools, share insights, and build confidence in synthetic data products.”