Link Prediction and Missing Data in Social Network Surveys: An Initial Exploration

• Activity Number: 696
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Taniecea Arceneaux*+ and Burton H. Singer
• Companies: U.S. Census Bureau and University of Florida, Emerging Pathogens Institute
• Keywords: social network analysis ; missing data ; imputation ; link prediction

Abstract

An important problem in the analysis of social networks, particularly those modeled using survey response data, is that of non-response. Recent studies have shown the negative effects of missing actors and missing ties on the structural properties of networks [4]. To overcome these problems, we adapt link prediction methodology [6] to the problem of single imputation of item non-response in social network survey data. Empirically, we examine the accuracy of link prediction methods on relational data for social groups in Zablocki's Urban Communes Data Set [13]. Furthermore, we extend the methodology to include the use of supplementary information for link prediction in the case of networks with multiple types of relations on the same individuals.