A1570
Title: Network autoregression for binary responses
Authors: Lijia Wang - City University of Hong Kong (Hong Kong) [presenting]
Abstract: Studying the propagation of binary responses on nodes in a large-scale social network is critical for understanding how individual behaviors and decisions are shaped by social structures and for predicting collective outcomes. A network autoregressive model for binary-valued responses is proposed, in which the probability of response at each node is influenced by its neighbors' past decisions, its own past decision, and node-specific covariates, through a logistic link function. The model accounts for network noise and community structure by assuming the underlying network is generated from a block model, with autoregressive parameters that are community-specific. Conditions are established under which the long-term behavior of the high-dimensional binary vector converges to a community-specific distribution and the associated convergence rate, illustrating when individuals in the same community or across the whole network reach a consensus regardless of their initial positions. Given an observed network and response vectors, asymptotic consistency and normality of the maximum likelihood estimators are established. The efficiency and validity of the inference procedure are demonstrated through simulated and real data. In particular, the model is shown to be applicable to studying the dynamics of strike occurrences in China and highlights the impact of online social networks in facilitating collective actions.
