Hierarchical Network Models for Interventions on Social Networks

• In Event: 46.028 - Quantitative Methods for the Analysis of Social Network Data in Education

Abstract

Observations and interventions are sometimes focused on the way students or teachers interact with one another (e.g. friendship networks after a bullying intervention, changes in professional learning communities over time, diffusion of innovation and reform practices in schools, etc.). Many whole school initiatives aim for some type of structural change in relationships such as less isolated friendship cliques, increased collaboration, resource sharing, advice seeking and receiving or move toward small learning communities, i.e. a change in the social networks in schools.

To model the multiple networks of classrooms and/or schools in these types of studies as well as estimate treatment effects, we developed a class of statistical social network models, Hierarchical Network Models (HNM) using a hierarchical Bayes approach. These models are very broad in that any single network statistical model can be extended using this framework and a variety of dependence assumptions about the networks can be incorporated.

We present a variety of HNMs in this talk and describe the types of substantive research questions they can each address. In particular, we will introduce the HLSM and HMMSBM and describe how they can be used to evaluate an intervention on social networks. For example, the HLSM can detect treatment effects on overall tie probability whereas the HMMSBM can detect effects on the subgroups within each network.

Author

• Tracy Sweet, University of Maryland - College Park