A Quantitative Approach for Evaluating the Utility of a Differentially Private Behavioral Science Dataset

Social scientists who collect large amounts of medical data value the privacy of their survey participants. As they follow participants through longitudinal studies, they develop unique profiles of these individuals. A growing challenge for these researchers is to maintain the privacy of their study participants, while sharing their data to facilitate research. Differential privacy is a new mechanism which promises improved privacy guarantees for statistical databases. We evaluate the utility of a differentially private dataset. Our results align with the theory of differential privacy and show when the number of records in the database is sufficiently larger than the number of cells covered by a database query, the number of statistical tests with results close to those performed on original data increases.