Computational Modeling approach for improved Social Network Analysis Model Specification
Elisa Bienenstock, Phillip BonacichSocial Network analysts, over time, have generated a broad collection of metrics to describe social structure. Selecting from among these metrics which is best to represent, summarize or describe data is an ongoing challenge. There are no clear guidelines that describe which metric, or set of metrics are optimal for revealing theoretically interesting features of a social system. In this paper we reveal how computational models can assist the researcher in determining which metrics are best suited to address a specific theoretical question. Here our focus is a comparison of eigenvector centrality with other centrality measures in a specific context: bias in the representation of attitudes or information in a network. The computational modeling approach reveals metric limitations, but also reveals how employing a combination of metrics can provide a more nuanced understanding about the relationship between network structure and outcomes.