Synthesizing Bipartite Graphs: An Application to Employer-Employee--Linked Data

• Activity Number: 8
• Type: Invited
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Jared Scott Murray*+ and Jerome P Reiter
• Companies: Duke University and Duke University
• Keywords: confidential data ; bipartite graphs ; social network

Abstract

We consider synthetic approaches to protecting data which are in the form of a bipartite graph, sometimes known in the context of social network data as two-mode or affiliation networks. Here the set of nodes is partitioned into two sets (individuals and "groups"). Edges in the graph represent an individual's association with a group. Such graphs are often characterized by very specific features (for example, almost all individuals have no more than two jobs, and most U.S. employers have fewer than 20 employees) and are typically accompanied by node-level attributes. The sparse but highly structured nature of this data make standard protection methods difficult to apply while maintaining data utility. We explore synthesis models which aim to preserve these important features while safeguarding confidentiality, with an application to the U.S. Census Bureau's Longitudinal Employer-Household Dynamics data.