Missing Data Handling in Weighted Social Network Analyses

• In Event: Advanced Statistical Methods for Complex Data

Abstract

Prior research (Authors, 2019) examined missing data handling for binary social network analysis, and found that imputation using sample density was preferred considering bias, precision, and practicality. The present paper extends this work by investigating missingness handling approaches for weighted network data (e.g., ratings) using Monte Carlo simulation with three missingness approaches: node deletion, imputation using a discrete uniform distribution, and imputation using a truncated multivariate normal (MVN) distribution. Results showed that MVN imputation was optimal considering both bias and precision. Importantly, imputation for network data is not currently available in R’s statnet suite of packages. As such, a proposed stand-alone R package, netImp, is presented to assist researchers in missingness handling prior to their applied social network analyses.

Authors

• Nathan Abe, The Energy Authority

• Elizabeth A. Sanders, University of Washington