Private Evidence Based Policy Making at the Department of Education

In 2017, the US Commission on Evidence-Based Policy-making unanimously recommended that inter-agency sharing of administrative data should be accompanied by enhanced privacy protections. Unfortunately, current approaches for doing so often involve outsourcing any computation that is performed on said data to third parties that are contractually obligated to safely and securely handle the data. In this pilot, we demonstrated to the department of education how multiparty computation can efficiently and securely perform any statistics needed for evidence base policy making in-between agencies with no recourse to anyone outside the department itself and without any privacy risks.

This work was completed during my internship at galois inc. in collaboration with Georgetown University's McCourt School of Public Policy. For more information you can checkout the [article], [technical report], and [codebase].

Privacy preserving analytics for assessing and addressing economic inequalities

In 2014, the Boston Women’s Workforce Council (BWWC) initiated a city wide study to analyze the wage gap among gender, ethnicity and seniority as part of an effort to advance the interests of women in the workforce. Initially, no third party was willing to undertake the risk of receiving the raw salary data coming from multiple companies partaking in the study. As such, we developed and deployed a web based MPC analytics system for the BWWC to allow them to run their statistics securely.

The success of this project was primarily due to relying our design decision heavily on the accessibility and comprehensibility of the system to a wider audience, and on understanding the various roles and dynamics of participants (e.g. asynchronicity of participation, possible errors in data entry, etc.).

In 2018, the Greater Boston Chamber of Commerce (GBCC) launched the Pacesetters Initiative which aimed to leverage the purchasing power of large and mid-sized companies to create and promote economic opportunities for local minority-owned businesses. In order to assess the progress of the initiative, we later adjusted and deployed this system to help the GBCC measure the initiative’s impact on supplier diversity practices, including ways to increase spending with minority-owned businesses.

Both the BWWC and GBCC studies are periodic and are run and deployed at least once a year. Later iterations of these deployments used JIFF, which I discuss below, as a backend.

This work was completed during my work at SAIL. For more information checkout our [paper] and our codebases[1][2]. My wonderful co-authors later formalized and extended this work in a paper on the usability of secure computation. You can checkout their work [here].