Online Program

WITHDRAWN: Examining statistical disclosure possibilities of an empirical ROC graph in the presence of auxilliary data

Ofer Harel, University of Connecticut 
*Gregory J Matthews, University of Massachusetts 

Keywords: Statistical Disclosure, Privacy, ROC Curve

Often, when researchers publish their results, summaries of their findings are presented in tables and figures hopefully in such a way that useful findings are communicated while at the same time no private information about an individual is disclosed. However, many statistical summaries present scenarios in which private information can be learned through the use of auxiliary information in combination with the statistical summary. Specifically, in many areas including diagnostic testing research, a common summary is the receiver-operating characteristic curve. Previously, it has been demonstrated that if an individual possesses the exact points of the empirical ROC curve and a subset of the true raw data, that they can attempt to reconstruct the full data set. However, in practice the exact values of the points on an empirical ROC curve will be unavailable. Rather, a malicious data user will have access to only an image of the empirical ROC curve. Here, we examine what a malicious data user can learn about a data set when they possess both the image of the empirical ROC curve and a subset of the true data set.