Approaches to Handling Missing Data in Two-Time Point Social Network Analyses

• In Event: 67.040 - Missing Data Considerations, Solutions, and Issues

Abstract

This paper investigates missing data handling approaches for a 2-time point social network analyses using the separable temporal exponential random graph model (STERGM). Monte Carlo simulation conditions included two levels of nodes, two sample sizes, two density levels, and six levels of missingness amounts. For each cell, six approaches to handling missing data were applied: node deletion, inserting 0s, inserting 1s, inserting a random 0/1 mix based on 50% or true density, and borrowing information from T1 or T2. A STERGM was then used on each set of T1-T2 data. Preliminary results show that using a random mix of 0s/1s was best at keeping formation and persistence parameters unbiased while retaining the best precision.

Authors

• Nathan Abe, University of Washington - Seattle

• Elizabeth A. Sanders, University of Washington

• Elizabeth A. Dietrich, University of Washington - Seattle

• Jessica J. Thompson, University of Washington