The Dynamics of Urban Neighborhoods: A Survey of Approaches for Modeling Socio-Spatial Structure

Abstract

For close to a century, researchers from across the disciplines of Urban Studies have developed empirical models for understanding the spatial extent and social composition of urban neighborhoods--and how these dimensions change over time. Unfortunately, however, these techniques have often been developed within disciplinary silos and without broad exposure to other potentially interested constituencies. In this paper, we traverse the literatures of social science, computer science, and statistics to examine a variety of modeling techniques for understanding neighborhood dynamics. We begin our review by examining early concepts of spatial structure first outlined in the Chicago School and discuss how the notions of social ecology and quantitative neighborhood analysis permeated the urban studies for several decades to come. Our survey continues by reviewing contemporary statistical approaches for identifying urban neighborhoods, culminating with the state of the art in subfields known as `geodemographics' and `regionalization'. Following this review, we offer insight into the field's persistent conceptual issues, identify areas ripe for additional research, and highlight newly-developed computational methods that can inform more just and socially equitable public policy, community development, and accountable governance.

Learning Geographical Manifolds: A Kernel Trick for Geographical Machine Learning

“…Developments in "spatially-encouraged machine learning" will have profound consequences for geodemographics, a domain heavily involved in dimension reduction in cases where a compromise between geographical regularity, temporal stability, and feature homogeneity can be difficult to strike (Voas and Williamson 2001;Singleton and Spielman 2014;Singleton, Pavlis, et al 2016), as well as other urban data science topics. Clustering the results of a hybrid geographical-manifold dimension reduction, where nearby observations are both similar and feature-near, may yet blend the dichotomy between spatial constraint and feature optimality (Knaap et al 2019).…”

“…This is equally true of geography, where machine learning models are applied ever more to multivariate problem domains contextualized by geographic space. Neighborhood analysis is a subfield of geography where this trend has been partic- Knaap et al 2019). Apart from these new methods, however, neighborhoods have both a rich theoretical tradition and a long history of quantitative analysis, much of which exists outside the realm of spatial effects or other geographic considerations (Shevky and Bell 1955;Raudenbush and Sampson 1999;Sampson et al 2002;Sampson 2012).…”

“…A widespread practice in the literature is the application of ML methods to reduce multivariate neighborhood data into a smaller, more tractable and interpretable dataset. Such dimensionality reduction techniques have been common in the urban social sciences for decades, with sociologists gravitating toward factor analysis in the form of "ecometrics" (Raudenbush and Sampson 1999;O'Brien et al 2015) and geographers favoring cluster analysis often termed "geodemographics" (Singleton, Pavlis, et al 2016;Knaap et al 2019) but rarely has either of these traditions attempted to incorporate spatial information into the modeling frameworks 1 . Thus while the longevity of the neighborhood analysis literature provides strong evidence for the value of classic and emerging aspatial machine learning methods, the fact that they lack consideration of geography's first law suggests there is a wide berth for improvement (Tobler 1970).…”

“…of course, there are important exceptions to this statement, most notably the field of geographic regionalization. For a recent review seeKnaap et al 2019 …”

Adaptive Choropleth Mapper: An Open-Source Web-Based Tool for Synchronous Exploration of Multiple Variables at Multiple Spatial Extents

“…Also, unlike ACM, LNE includes an interface that allows users to select variables and regions ( Figure 3). The development of ACM and LNE are parts of The GeoSpatial Neighborhood Analysis Package (GEOSNAP) [36], a suite of spatial statistics and visualization tools for longitudinal neighborhood analysis.…”