Generating Poisson-Distributed Differentially Private Synthetic Data

The dissemination of synthetic data can be an effective means making information from sensitive publicly available while reducing the risk disclosure associated with releasing directly. While mechanisms exist for synthesizing that satisfy formal privacy guarantees, utility is often afterthought. More recently, use methods disease mapping literature has been proposed to generate spatially-referenced high utility, albeit without guarantees. objective this paper help bridge gap between and literatures. In particular, we extend existing approach generating formally private case Poisson-distributed count in a way allows infusion prior information. To evaluate data, conducted simulation study inspired by available, county-level heart disease-related death counts. results demonstrate differentially outperforms popular technique when counts correspond events arising subgroups unequal population sizes or event rates.