Modeling Out-Degree of Node in Commuter Networks: The Impact of Socioeconomic Profiles and Spatial Autocorrelation in Shaping Regional Connectivity

Regional commuting networks are shaped by the distribution of population and employment as well as a host of other socioeconomic and demographic factors. A commuting network type of particular interest consists of residential origins and work destinations as nodes, with the desire lines between them functioning as links. A residential origin has links not only to adjacent towns and the main city, but to work destinations in many other towns as well. Although it is expected that highly connected towns are spatially clustered around the urban core, reflected in positive spatial autocorrelation (PSA), this pattern can be interrupted by origins with seemingly low demonstrated connectivity, reflecting negative spatial autocorrelation (NSA). Principal components analysis leading to regression analysis combined with Moran eigenvector spatial filtering provide an approach for achieving deeper insight into these varied patterns of connectivity in the Pittsburgh, Pennsylvania, region, while taking the role of PSA and NSA into account. It also highlights the role of NSA, which reveals spatial patterns that can at least appear to challenge the notion of Tobler’s law that closer things are more related, and a deeper analysis suggests that PSA and NSA jointly characterize connectivity, but not in equal measure. It is argued that actors such as population age, wealth, occupation type, household car ownership, distance from the core, and so on would offer refinement to the predictive ability of density or accessibility in modeling connectivity as represented by degree of node.