This prototype system will allow researchers with sensitive datasets to make differentially private statistics about their data available through data repositories using the Dataverse platform. A paper describing our system can be found here in this ArXiv link.
As an intern, I worked on the beginning stages of the interactive GUI and researched applications of differential privacy software. I collaborated with Nabib Ahmed, and I was advised by Dr. James Honaker and Jack Murtagh.
PSI is a system of interlocking statistical tools for data exploration, analysis, and meta-analysis. The first to be released is an interface for quantitative analysis, that allows users at all levels of statistical expertise to explore their data, describe their substantive understanding of the data, and appropriately construct statistical models. This integrates with Dataverse and the Zelig Project through a portable, lightweight, browser-based and gesture-driven interface, allowing users to run statistical models available in Zelig on data archived in Dataverse.
The demo of the Budget Tool Interface uses replication data from Fearon and Laitin's 2003 article "Ethnicity, Insurgency, and Civil War".
The second iteration Budget Tool GUI allows data depositors to:
- select variables of interest
- select statistics of interest
- allocate global privacy parameters ("the privacy budget")
- interactively view accuracy rate of statistics
- submit privacy budget
View the first-iteration demo here.
(Last Updated August 2016)
This project is part of the Privacy Tools Project, a broad effort to advance a multidisciplinary understanding of data privacy issues and build computational, statistical, legal, and policy tools to help address these issues in a variety of contexts. It is a collaborative effort between Harvard's Department of Computer Science, Center for Research on Computation and Society, Institute for Quantitative Social Science. A National Science Foundation Secure and Trustworthy Cyberspace Project, with support from the Sloan Foundation and Google, Inc.