Record Re-Identification of Swapped Numerical Microdata

Government agencies, researchers, healthcare providers, and other organizations release data for public use. To protect the privacy of the data subjects, these organizations mask the data prior to release. One popular masking procedure is data swapping, by which values of records are exchanged before being released. Data swapping is one of the preferred techniques since it is simple, easy to implement, and---based on prior studies---provides a reasonable balance between disclosure risk and data utility. In this study, we investigate the ability of an adversary with limited knowledge (of just a single record) to re-identify a record in the swapped data by using a procedure that reverse engineers the data-swapping process. The study also provides the adversary with the ability to evaluate the effectiveness of the re-identification. We empirically evaluate the effectiveness of data swapping using a dataset that has been used previously to evaluate the effectiveness of masking techniques. Our results demonstrate that data swapping can be vulnerable to disclosure even against this limited knowledge adversary.