A Comparison of Statistical Network Models

• In Event: Methods and Models for Your Social Network Data

Abstract

In social network analysis, statistical network models are often used to make inferences about the relationship between individual characteristics and the probability of ties in a network. The three commonly used statistical network models for this purpose are Exponential Random Graph Model (EGRM), Stochastic Block Model (SBM) and Latent Space Model (LSM). There are several attempts to link these three models conceptually or theoretically, but there is a lack of detailed model comparison to address their connection and distinction, under what circumstances one model is preferred than the other, and whether the covariate effect inferred by each model is consistent.

The purpose of our work is to fill in this gap by providing a comprehensive simulation study to compare these three models with regards to the covariate effect, the model-assumed network structure, goodness-of-fit to data with different network structures, as well as easiness of implementation. Also, we will use several sets of commonly used real-world data as examples to demonstrate the conclusions we draw from simulation studies.

The results of this work can provide useful information to applied researchers on model selection and interpretation, as well as serve as a reference for methodology researchers to better understand and extend existing models, or to develop a completely new class of models.

Author

• Qiwen Zheng, University of Maryland - College Park