Humans act in ways that are a direct consequence of our surrounding environment, and social interactions play a key role in shaping our behavior. While simple, individual acts or public movements may not seem to contribute to the broader social milieu, sustained patterns in the ways humans live and move can result in significant spatial variability. Geographers, and human geographers in particular, have been keenly interested in social factors such as race, health, and political identity that drive spatial variation, particularly as they contribute to processes of intentional separation, such as segregation and dispersion.
Despite its importance in segregation and dispersion research, social interaction remains difficult to model. Challenges include the necessity of obtaining highly dimensional movement trajectories or survey responses of activity, both of which can be costly to collect and analyze. This has resulted in the development of many methods that focus on area or zone-based measures of group differences, which can only produce a single estimate for all individuals in an area. But due to the ever expanding growth of computational power, the last several decades have seen the development of numerous methods devoted to better capturing interaction, and network frameworks and analyses, which can model multiple relationships through node and edge representations, have gained immense popularity. Coupled with the advancement of spatial modeling capabilities, especially within geographic information systems (GIS), measures of segregation and dispersion have been and have the potential to be vastly improved.
The goal of this dissertation is to examine compelling dimensions of segregation and dispersion using network and computational methods in an effort to more accurately capture the complexity of human interaction in ways that were previously difficult to do. The first chapter of this dissertation explores how measures of racial segregation can be captured from summaries of movements, and proposes a network-based metric. The second chapter delves into the realm of spatial optimization, and introduces how measurements of potential interaction can be used to more systematically and safely disperse individuals from one another in a shared office setting during a pandemic. The third chapter more critically interrogates the social and spatial definitions of existing measures of partisan exposure and isolation, and demonstrates that even with more data and more computing power, fundamental coding decisions related to social and spatial parameters can greatly affect estimates of partisan segregation.
Of course, limitations still exist within the conceptual models of each of these chapters, as abstracting reality into a digital environment implies some loss of accuracy. This can be seen in Chapter 1’s abstraction of neighborhoods to points and diverse travel activity into pairwise neighborhood trips, Chapter 2’s estimate of unknown intersecting movements into a single measure of overlap, and Chapter 3’s modeling of one thousand nearest neighbors to represent a ``neighborhood.'' Still, this dissertation shows that (1) networks and computational methods provide versatile frameworks for modeling complicated and often unknown human behaviors and (2) that despite the challenges of representing interactions, our developed network and computational methods help better triangulate interaction potential through thoughtful specification of modeling decisions. My goal with this dissertation is to ameliorate the constant research design tensions that exist between overlapping social and spatial definitions, and by developing methods that more accurately capture human interaction, help advance research and policy on the inequities created by social and spatial variation.
