While tax data is highly sought after by social scientists, it is costly, sensitive, and difficult to access. The IRS has historically released public-use files—privacy-protected databases of sampled individual income tax returns—but has stopped producing them due to high costs and high vulnerability to re-identification attacks. This grant provides ongoing support for Claire Bowen at the Urban Institute, who is working with the IRS to develop synthetic versions of individual income tax return data. Synthetic data has mathematical and statistical properties that are similar to those of the real data, but that contains almost no private information from the original dataset. Grant funds will allow Bowen to continue developing two synthetic datasets, making substantial methodological improvements and exploring the application of differential privacy methods to assess the privacy attributes of this methodology. In addition, Bowen will make open-source code available on GitHub, document the methodology for use by other agencies, and disseminate the work through a white paper, blog posts, presentations, and journal articles.

This grant supports work by V. Joseph Hotz at Duke University, Ruobin Gong at Rutgers University, and Ian Schmutte at the University of Georgia, who are leading an initiative through the National Bureau of Economic Research (NBER) to support research on privacy-protecting methods of data analysis. Of particular interest are how different implementations of such methods manage the inevitable trade-off between privacy and accuracy. Grant funds will allow the team to commission 18 research papers, including an NBER working paper series and an NBER proceedings volume; organize two in-person conferences and two virtual meetings on topics surrounding the use of privacy-protected data in empirical research; and foster interactions among researchers in economics, statistics, and computer science. Funding will be used primarily to cover expenses associated both with the commissioned papers and with the research conferences.

The mathematical theory of differential privacy describes methods and practices that allow researchers to query sensitive datasets while controlling how much each query compromises the privacy of individuals contained in the dataset. This approach represents the cutting edge of privacy-protection, but one that is mathematically subtle and challenging to implement. Widespread use of these methods will require lowering the cost of adoption and adaptation.  OpenDP  is therefore producing a tested, trustworthy, interoperable, and flexible library of software that will make it easier for users to set up differentially private access to sensitive data. This grant provides continuing support for Harvard computer scientist Salil Vadhan, creator of OpenDP, as well as a dedicated community of theorists, engineers, practitioners, and privacy experts that is aiming to increase adoption of differential privacy. Now in its third year, OpenDP is shifting from a minimum viable product to a prospering ecosystem with heightened impact and broadened support. Specifically, grant funds allow Vadhan to expand OpenDP’s library capabilities to meet new application needs; promote OpenDP adoption among social science researchers; and further strengthen the growing community of experts using and contributing to OpenDP.  Eventually, OpenDP will serve as a sustainable open-source library of tools and community dedicated to privacy-preserving data analysis.

The value of letting independent social scientists study the administrative data collected by companies of all sorts is hardly in doubt. Economists are particularly keen on basing hypotheses, models, and economic indicators on such information. Companies are often reluctant to share their data, however, in part due to concerns regarding data privacy, costs, inconvenience, reputational risk, or ethical issues. Recent regulatory measures, which give users control over their data, complicate matters even further. Without better data sharing mechanisms, we may soon live in a world where only a few large companies have access to that data and the insights such information provides. This grant supports Sara Jordan at the Future of Privacy Forum (FPF) Education and Innovation Foundation, a strictly nonpartisan and nonprofit organization, who is developing a strategy to further accelerate corporate-academic data sharing. Grant funds provide continued support for the Foundation’s Award for Research Data Stewardship, allow Jordan to prepare compelling use cases to demonstrate how insights generated by administrative data can advance research and evidence-based policymaking, and also allow the creation of a legislative tracker producing real-time analysis to be shared with the research community. Combined, Jordan’s efforts accelerate the safe and responsible sharing of administrative data between companies and academic researchers.