Chapter 12: Multilevel Social Network Models: Incorporating Network-Level Covariates into Hierarchical Latent Space Models

Social network modeling is becoming increasingly common in education research, in some part due to the availability of multilevel social network models. Historically, social network methods have focused on a single network, but education research often involves several if not many networks such as multiple classrooms in one school or multiple schools in one district. Using multilevel social network models allows researchers to investigate patterns across multiple social networks. In this chapter, we consider one such model, the hierarchical latent space model (HLSM; Sweet, Thomas, & Junker, 2013) and focus on incorporating network-level covariates in HLSMs. We present several empirical analyses using real-world data to illustrate how network-level covariates can be used in practice, and then we explore some of the operating characteristics of the HLSM through two small simulation studies. The first isa sensitivity analysis to explore the effects of prior distribution specification on parameter recovery and bias, and the second is a power analysis to understand the relationship between network size, number of networks, and both parameter recovery and power.