B1368
Title: Spectral estimation of latent structure in networks with covariates
Authors: Swati Chandna - Birkbeck, University of London (United Kingdom) [presenting]
Abstract: Many real-world networks are observed with a large number of covariates, which possibly explain the intensity of interactions between pairs of nodes. For example, covariates such as the number of material and verbal conflicts between country pairs, etc., are observed in the study of alliance networks. It is important to understand the extent to which these covariates explain alliance formation. In such settings, interest also lies in the residual network structure that remains unaccounted for by the observed covariates. A simple dot product kernel is used to model this residual network structure and show how this model with covariates may be estimated via least squares. Further, bootstrap is employed to draw inferences on the homophily parameter and the residual network structure. Application to alliance networks illustrates the usefulness of the approach.
