Keywords

Geographic Information Systems, Social Network Analysis, Exponential Random Graph Models, Spatially Embedded Social Networks, Model Fit, Spatial Autocorrelation, Network Centrality, Community Detection, Location-Based Social Networks, Brightkite

Subject Categories

Geographic Information Sciences | Geography | Social and Behavioral Sciences

Abstract

Social networks are embedded in geographic space, yet traditional Social Network Analysis often separates relational structure from the spatial contexts in which social ties form. This dissertation evaluated whether Geographic Information Systems (GIS)-derived spatial information improves Exponential Random Graph Model (ERGM)-based analysis of spatially embedded social networks. Using the Brightkite location-based social network dataset, the study analyzed a contiguous United States network of 15,719 active users, 73,984 undirected friendship ties, and 2,769,396 check-in records. GIS-derived measures included user centroids, mobility dispersion, dyadic distance, shared place exposure, and spatial-cluster membership, with spatial-cluster matching used as the GIS-derived predictor in the final models. Four models were compared: a non-spatial benchmark, a GIS-enhanced benchmark, an activity-only benchmark, and a GIS-plus-activity benchmark. Corrected independent model-generated networks were used to evaluate model fit and adequacy, centrality reproduction, community structure and spatial-community alignment, and residual spatial autocorrelation in node-level network features. Results showed that GIS-derived spatial-cluster information improved comparative model fit and same-spatial-cluster tie reproduction under both direct and activity-adjusted comparisons. GIS enhancement also improved selected community-structure and spatial-community-alignment outcomes, indicating that geographic organization contributes to meso-level subgroup representation. However, GIS enhancement did not improve overall centrality reproduction, node-level centrality alignment, recovery of observed top-central users, or consistent reduction of residual spatial autocorrelation. Overall, the dissertation suggests that geographic context improves selected aspects of ERGM-based network representation, while spatial enhancement alone is insufficient to reproduce all node-level structural roles or eliminate residual spatial dependence.

Completion Date

2026

Semester

Summer

Committee Chair

Hahs-Vaughn, Debbie

Degree

Doctor of Philosophy (Ph.D.)

College

College of Community Innovation and Education

Department

Department of Learning Sciences and Educational Research

Format

PDF

Document Type

Dissertation

Language

English

Release Date

8-15-2027

Available for download on Sunday, August 15, 2027

Share

COinS
 

Accessibility Statement

This item was created or digitized prior to April 24, 2027, or is a reproduction of legacy media created before that date. It is preserved in its original, unmodified state specifically for research, reference, or historical recordkeeping. In accordance with the ADA Title II Final Rule, the University Libraries provides accessible versions of archival materials upon request. To request an accommodation for this item, please submit an accessibility request form.