Estimating Contextual Effects from Ego Network Data: The Case of Cohesion in Schools

• In Event: 4258 - Social Network Analysis: New Methodologies

Abstract

Network methods are a natural fit for contextual models. Global network measures offer a rich picture of a social context, making them ideal contextual-level predictors of health, mental health and the like. Such richness comes at a cost, however, as network data are difficult to collect, requiring information on all actors and all ties between actors. A contextual study pushes this to the extreme, as one would have to collect full network data in every context of interest. This paper considers a middle-ground, one that avoids the heavy data collection toll, while still retaining the best features of a contextual-network approach. The basic idea is to combine hierarchical linear models with network sampling, where one uses sampled, ego network data to infer the network features of each context, and then uses the inferred network features as second-level predictors in an HLM. I test the validity of this idea using a complete dataset (Add Health) on adolescents in school. I predict two individual-level outcomes, school attachment and behavioral problems in school, as a function of cohesion, defined at the contextual level (measured as density and bicomponent size). The results on the whole are encouraging. Across all models, it is possible to approximate the true coefficients of density and bicomponent size just using the ego network data. The hope, going forward, is that researchers will find it easier to incorporate holistic, network measures into traditional regression models.

Author

• Jeffrey A. Smith, University of Nebraska-Lincoln