Abstract

Speaker: Eric Kolaczyk
Title: Coevolving Latent Space Network with Attractors Models for Polarization
Abstract: We develop a broadly applicable class of coevolving latent space network with attractors (CLSNA) models, where nodes represent individual social actors assumed to lie in an unknown latent space, edges represent the presence of a specified interaction between actors, and attractors are added in the latent level to capture the notion of attractive and repulsive forces. We apply the CLSNA models to understand the dynamics of partisan polarization on social media, where we expect US Republicans and Democrats to increasingly interact with their own party and disengage with the opposing party. Our analysis confirms the existence of partisan polarization in social media interactions among both political elites and the public. Moreover, while attractive partisanship is the driving force of interactions across the full periods of study for both the public and Democratic elites, repulsive partisanship has come to dominate Republican elites' interactions since the run-up to the 2016 presidential election.

Speaker: Maggie Niu
Title: Learning Network Properties without Network Data -- A Correlated Network Scale-up Model
Abstract: The network scale-up method based on "how many X's do you know?" questions has gained popularity in estimating the sizes of hard-to-reach populations. The success of the method relies primarily on the easy nature of the data collection and the flexibility of the procedure, especially since the model does not require a sample from the target population, a major limitation of traditional size estimation models. In this talk, we propose a new network scale-up model which incorporates respondent and subpopulation covariates in a regression framework, includes a bias term that more accurately estimates response biases under fewer assumptions, and adds a correlation structure between subpopulations. In addition to estimating the unknown population sizes, our proposed model depicts people's social network patterns in an aggregated level without using the network data.

Speaker: Gongjun Xu
Title: High-dimensional Factor Analysis for Network-linked Data
Abstract: Factor analysis is a widely used statistical tool in many scientific fields, including psychology, economics, and sociology. Meanwhile, observations connected by a network are becoming increasingly common, and how to incorporate the network structure in factor analysis remains an open question. Focusing on high-dimensional factor analysis with network-connected observations, we propose a generalized factor model with latent factors capturing both the network structure and the dependence structure among the high-dimensional variables. We develop a computationally efficient estimation procedure and establish asymptotic inferential theories. In particular, we show that by borrowing information from the network, the proposed estimator of the factor loading matrix achieves the optimal asymptotic variance under much milder identifiability constraints than the existing literature.

Speaker: Yi Yu
Title: Change Point Localisation in Dynamic Multilayer Random Dot Product Graphs
Abstract: In this talk we are concerned with change point localisation in a dynamic multilayer network, which is modelled through a sequence of random dot product graphs in tensor forms, with temporal dependence and cross-layer dependence. We will demonstrate the minimax optimality of our proposed change point estimators and support the theoretical findings through extensive numerical experiments.