Four checks for low-fidelity synthetic data: recommendations for disclosure control and quality evaluation

Area of Science:

• Data Science
• Information Security
• Research Methodology

Background:

• Confidential administrative data is typically restricted to researchers within Trusted Research Environments (TREs).
• Low-fidelity synthetic data (LFSD) is proposed for external use, enabling code-testing and data discovery.
• Transparency regarding LFSD creation and characteristics is essential for user understanding.

Purpose of the Study:

• To outline essential quality standards and transparency measures for releasing LFSD.
• To ensure LFSD is useful for preliminary analyses while maintaining data security.
• To guide data controllers in preparing and releasing LFSD responsibly.

Main Methods:

• Propose checks for LFSD release, including clear labelling as synthetic data.
• Mandate disclosure risk evaluation and mitigation strategies for LFSD.
• Emphasize maintaining structural similarity between LFSD and TRE data, with documented differences.

Main Results:

• LFSD release requires careful consideration of labelling, disclosure risks, data structure, and comprehensive documentation.
• Variable relationships are not preserved in LFSD, posing potential disclosure risks.
• Clear communication of LFSD's limitations compared to TRE data is necessary for accurate preliminary analyses.

Conclusions:

• Data controllers must implement checks to ensure LFSD is well-documented, useful, and non-disclosive before release.
• A flexible, context-aware approach to LFSD release rules is recommended over strict, rule-based methods.
• Effective LFSD management balances data accessibility for research with robust privacy protection.