Dynamics of Population Flow Networks

Taking a relational and systemic approach, this dissertation offers theoretical, methodological, and empirical advancements in understanding the social forces that drive or inhibit human migration. We consider migration flows among geographical areas as a network system, analyzing its dynamics using the exponential-family random graph models (ERGMs) and simulation methods. Chapter 2 grapples with the computational hurdle for modeling valued/weighted networks using ERGMs. We propose and implement an efficient parallelizable subsampled Maximum Pseudo-Likelihood Estimation (MPLE) scheme, which enables fast and accurate computation of ERGMs for big valued networks with high edge variance. The comparative simulation experiments further show whether and how the performance of existing computational approaches vary by the network size and the variance of edge values, providing guidelines for choosing and tuning those methods for different use cases. Chapter 3 applies the implemented method to study intercounty migration in the United States (U.S.), whose migration rates have declined for decades and reached a historical low. We found a pattern of "segmented immobility," where fewer people migrate between counties with dissimilar political contexts, levels of urbanization, and racial compositions. We also propose a "knockout experiment" framework to quantify the impact of segmentation on population immobility. The chapter reveals the social and political cleavages that underlie population immobility in the U.S., bridging the scholarly domains of (im)mobility, segregation, and polarization. Motivated by debates about California’s net migration loss ("California Exodus"), Chapter 4 examines the scale of and the mechanisms behind the migration-induced population redistribution among U.S. states. We combine ERGMs, knockout experiments, and a protocol for functional form visualization to understand the complex effects of political climates, housing costs, racial dynamics, and ur