A Simulation-Based Approach to Differential Privacy

This talk presents a simulation-based approach to statistical inference under differential privacy, with a focus on proportion estimation. After reviewing the limitations of existing DP inference methods—such as noisy approximations and accuracy loss—we introduce DP-JIMI, a differentially private adaptation of the Just-Identified Minimal Distance estimator. The method leverages simulation, fiducial matching, and exact distributional results to deliver more reliable inference in finite samples, outperforming common alternatives while remaining simple to apply.