Abstract:
|
A number of interventions in social settings such as education seek to improve the structure of social networks as a means to improving more distal outcomes, such as test scores. The rapidly expanding field of network modeling offers a wide range of models to summarize networks, as well as model the effects of interventions on network structure or the effects of network structure on other outcomes. In this talk we will introduce a framework for modeling networks in their intended role: as causal mediators, mediating the relationship between an intervention and an outcome. Specifically, we will unify the hierarchical latent space model for networks with the Neyman/Rubin causal model, by modeling subjects' potential positions in the latent space in treatment and control conditions, respectively. This leads to potential values for network-based mediators. We will discuss options and issues in estimating network mediation models
|