Protecting Privacy and Preserving the Value of Federal Statistics for Research and Evidence-Building

ASA’s Committee on Privacy and Confidentiality

Currently, the US federal government is undergoing efforts to streamline its operations by consolidating certain agencies and significantly reducing the size of others. This restructuring aims to eliminate redundancy, improve efficiency, and cut costs. Alongside this, there is a growing push to centralize and integrate federal administrative data through initiatives such as data lakes—large, unified repositories that store vast amounts of structured and unstructured information from multiple agencies. These efforts are intended to enhance data analysis, improve public services, and inform policymaking more effectively.

As more sensitive personal and government data are pooled into shared repositories, however, the risk of misuse, unauthorized access, and large-scale breaches increases. Critics warn that without robust oversight, transparency, and strong data governance policies, centralizing federal data could compromise individual privacy rights and reduce accountability in how personal information is handled and safeguarded across agencies. Balancing efficiency with privacy protection remains a central challenge in these modernization efforts.

One critical privacy issue that agencies, especially those within the federal statistical system, need to address is how to protect against the disclosure of confidential information in data products derived from these and other federal data assets. Greater reliance on administrative and blended data introduces new disclosure vulnerabilities, and rapidly evolving technologies make it easier for third parties to attempt to reidentify specific individuals and businesses in “deidentified” government data. Improvements in Statistical Disclosure Limitation methods can mitigate these risks, but stronger and more effective protections carry costs of their own, both in terms of the technical expertise necessary to apply them effectively and with regards to the techniques’ impact on the usability and value of the resulting data.

Successfully navigating these challenges and ensuring the ongoing value of federal data for decision-making, evidence-building, and research will require ongoing investment and engagement by statistical agencies, academia, and the broader statistical community.

Background

Statistical agencies and other government agencies responsible for producing public data have the following dual mandate:

1. Providing useful statistics for researchers, data practitioners, and public policymakers
2. Ensuring that those published statistics do not disclose or allow others to discover confidential information about individual households and businesses

This dual responsibility of protecting confidential information for these agencies is grounded in longstanding legal frameworks dating to the Privacy Act of 1974 and the Confidential Information Protection and Statistical Efficiency Act of 2002, which established violations as felonies punishable by steep fines and prison terms. More recently, the Evidence Act in January 2019 reaffirmed these protections and the requirement that statistical agencies “protect the trust of information providers by ensuring the confidentiality and exclusive statistical use of their responses.”

The challenge for government agencies today is how to balance the growing demand for accessible and detailed public data with the need to maintain robust privacy protections against a rapidly evolving computing landscape that preserves public trust in these data.

Historically, government agencies have protected data subject confidentiality using traditional SDL methods. These methods include top-coding, swapping, and omitting geographic or industry details from public-use microdata and tabular files. However, due to advances in computational capabilities, minimal oversight and guidance of artificial intelligence, and the increasing availability of detailed personal information from commercial sources, traditional SDL methods are insufficient for protecting respondent confidentiality.

Government agencies are exploring new SDL frameworks to address the emerging and evolving privacy risks (e.g., responding to the prevalence of AI) while continuing to meet their data users’ needs for quality statistics (e.g., complying with the Evidence Act). These efforts include investigating methods to strengthen traditional approaches to SDL, researching and embracing newer SDL methods (e.g., synthetic data and formally private methods), and creating new and innovative ways of providing approved researchers with secure access to the data they need (e.g., virtual data enclaves, controlled access online data tools, and secure multiparty computation).

These new ways of accessing data offer opportunities for greater transparency around how data are transformed and the impact it has on data usability, increasing public trust and scrutiny. In other words, this transparency allows for methodological innovations in drawing valid inferences from privacy-protected statistics and invites more public scrutiny and discussion of agencies’ disclosure avoidance decision-making. These new SDL methods also present challenges for data users who may rely on the statistics remaining available in the same format, with the same level of granularity and precision. Therefore, this new landscape requires a collaborative effort among researchers, policymakers, data users, and statistical agencies to navigate the challenges and opportunities associated with changing SDL methods.

To that end, members of the ASA Committee on Privacy and Confidentiality encourage ASA members and the broader statistical community to engage in an ongoing dialogue about these developments and make the following recommendations to its members, statistical agencies, and the broader statistical community.

Recommendations

• Expand awareness, training, and education.
  • Members of the ASA, other professional organizations, and the broader official-statistics community can promote awareness of new SDL methods and their implications for statistical research. Researchers whose work relies on public-use microdata files or tables will need to understand these changes and, where appropriate, adapt their methodologies. This will help ensure researchers are informed and motivated to engage in discussions and efforts in their self-interest.
  • Members of the ASA, other professional organizations, and the broader official-statistics community can encourage journals to establish policies and practices for reviewing manuscripts based on privacy-protected data.
  • Members of the ASA, other professional organizations, and the broader official-statistics community can train our profession by expanding graduate and continuing education courses on the topic.
• Develop collaborative communication strategies.
  • The statistical community can help agencies develop communication and dialog strategies to increase input and support from impacted groups and individuals. In doing so, ASA members can help generate the public discourse and input federal agencies need from academics, practitioners, and data users to inform decision-making around statistical disclosure.
  • Statistical agencies can inform the ASA and professional organizations in other disciplines to raise awareness about upcoming changes in SDL methods.
  • Statistical agencies can collaborate with academics and professionals across disciplines to develop a communication and dialog strategy about the importance of protecting confidentiality and the role of new SDL methods. Engagement on these issues needs to be multi-faceted, multidisciplinary, and ongoing.
• Advance methodological development and adaptation.
  • The statistical community can develop and adapt statistical methods that allow valid inferences from privacy-protected data. This effort should account for privacy protections as an additional form of error in the measurements alongside traditional notions of sampling error or measurement error.
  • The statistical community can develop methodological tools to better capture the tradeoff between publishing useful statistics and protecting confidentiality when evaluating alternative privacy protection strategies.
  • The statistical community can assist statistical agencies in adopting formal privacy protections as part of the overall ongoing transformation of our national data infrastructure toward more blended statistical products.
• Provide sustained financial support.
  • Funders, academics, private institutions, and other organizations can help by supporting efforts to develop appropriate privacy protection methods. This includes sponsoring research projects, seminars, workshops, interagency personnel agreements, training, and opportunities for researchers to collaborate with statistical agency staff. Such funding will facilitate the smooth implementation and acceptance of new SDL methods.