About the client
The IRS manages hundreds of millions of personal records from individuals and corporations. The Statistics of Income Department of the IRS needs to be able to release statistical summaries to other government agencies to further research and support evidence-based policymaking. However, these releases must adhere to stringent privacy constraints encoded in U.S. Law.
The challenge
The IRS asked Tumult Labs to design and implement a unique solution for one of their highly visible data products. On The Department of Education’s College Scorecard website, the organization publishes annual college graduate income summaries to help students and their parents evaluate potential schools. The DOE reaches out annually to the Statistics of Income division of the IRS to request degree-based income data summaries on graduates from thousands of institutions across the country.
While the IRS is no stranger to requests for data releases, the organization’s traditional privacy techniques have begun to breakdown as requests grow in both number and complexity.
- The team’s legacy approach relied on the suppression of any at-risk groups. Lately this required suppression rate was going beyond 70%, yielding potentially unusable data.
- As the number of detailed requests grew, the team was also concerned about their ability to verify privacy protection.
It became clear that the IRS needed a stronger methodology for crafting data releases—one that would provide accurate results while also meeting its strict regulatory requirements.
Tumult Labs’ Differential Privacy solution
Tumult Labs specializes in differential privacy solutions that meet the data sharing goals and compliance requirements of government agencies like the IRS.
- First, Tumult Labs assessed the data release needs at the IRS, taking into account its privacy budget constraints and data utility tradeoffs.
- Next, Tumult designed a solution for the safe release of counts and income quantiles. The solution was built on Tumult’s SafeTables platform, offering an end-to-end guarantee of privacy and optimized utility.
- With SafeTables deployed on-premises, IRS data never left the trusted enclave of the institution.
- Tumult’s experts then helped tailor the platform to the Statistics of Income department’s utility goals, empowering IRS officers to make precise, informed decisions around acceptable privacy risk and accuracy for each release.
The results
With its customized deployment of SafeTables, the IRS was able to share data with the Department of Education that was better protected, more accurate, and contained fewer suppressions.
- IRS officers had greater confidence in the reduced privacy risk resulting from the shared data, because of the provable guarantee provided by Tumult Labs' solutions.
- The informative data statistics are now live on the 2020 College Scorecard website! See it for yourself.