EM and Data Augmentation Algorithms for Social Network Analysis with Missing Data

• Activity Number: 138
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Junchi Guo*+ and Michael Larsen
• Companies: and George Washington University
• Keywords: Nodes ; Edges ; Labels ; Values ; Missing at random ; MLE

Abstract

Social network analysis (SNA) is used in a broad set of applications. Most of the statistical research in this field has been focusing on modeling the adjacency matrix. Not much of the discussion focuses on missing covariates associated with actors or nodes. In wireless service companies, such as AT&T and Sprint, call detail records (CDR) define connections in the social network. The companies have missing data problems of various kinds. Demographic information on subscribers based on self report is frequently missing. So the question is how to make use of the data to do inference about the social network structure when there are missing labels on the customers. We describe the EM algorithm for use with various social network models with missing data. Extensions are made enabling Bayesian analysis and multiple imputations through data augmentation. We explore the possibility of applying our methods in large, complex networks containing millions of edges, or links, between nodes, such as is found in the CDR data from wireless service companies.