Modeling Social Network Change over Time: A Comparison of Methods

Abstract

This study contributes to the gap in educational research methods knowledge by 1) assessing prevalence of using analytical social network methods in educational research, 2) being the first to directly compare four possible approaches for modeling longitudinal stochastic social networks: temporal exponential random graph model with bootstrapping (TERGM-B), temporal exponential random graph model with Markov Chain Monte Carlo maximum likelihood estimation (TERGM-M), separable temporal exponential random graph model (STERGM), and stochastic actor-based model (SABM), and 3) providing a demonstration of each method analyzing a small-sample, low-density network of professional development participants who were followed for four years. The field of education has been slower to adopt social network analysis (SNA) as a research tool compared to the health and social sciences. To date, few studies employ quantitative SNA, and none have reported using longitudinal SNA models. The present study examines the prevalence of stochastic network models in educational research and compares options for conducting stochastic models for longitudinal data using both simulated and observed network data. The simulated network data included two and four time points, with and without linear and quadratic change in density. The observed network data involved a small sample size, a sparse network, and four time points. For both kinds of datasets, exponential random graph model (ERGM) approaches, including temporal and separable temporal ERGM (TERGM and STERGM), as well as stochastic actor-based model (SABM) were applied. Results of the simulated datasets has shown that the TERGM appears to perform closest to the simulation parameter settings, with STERGM performing second best. The STERGM and SABM presents challenges in practice. Advantages and disadvantages of each approach are discussed.