Social selection models for multilevel networks
Abstract
Social selection models (SSMs) incorporate nodal attributes as explanatory covariates for modelling network ties (Robins et al., 2001). The underlying assumption is that the social processes represented by the graph configurations without attributes are not homogenous, and the network heterogeneity maybe captured by nodal level exogenous covariates. In this article, we propose SSMs for multilevel networks as extensions to exponential random graph models (ERGMs) for multilevel networks (Wang et al., 2013). We categorize the proposed model configurations by their similarities in interpretations arising from complex dependencies among ties within and across levels as well as the different types of nodal attributes. The features of the proposed models are illustrated using a network data set collected among French elite cancer researchers and their affiliated laboratories with attribute information about both researchers and laboratories (0070 and 0075). Comparisons between the models with and without nodal attributes highlight the importance of attribute effects across levels, where the attributes of nodes at one level affect the network structure at the other level.