Machines Learn The Chicago School: Modeling Multidimensional Neighborhood Change as a Spatial Markov Process

Elijah Knaap, Sergio Rey, Levi Wolf, Nicholas Finio

Despite lively interest and much active research, there remains little consensus on the appropriate ways to measure gentrification and neighborhood change, and even less on the best ways to model the phenomenon. In this paper, we enter the debate on gentrification by considering a novel model of neighborhood change. Drawing from regional science, social theory, and unsupervised machine learning, we construct a model of gentrification that accounts simultaneously for multiple dimensions of change and incorporates both spatial and temporal effects. The crux of our approach is the consideration of a neighborhood as a bundle of demographic attributes which together describe a discrete ’neighborhood state’ rather than a single or series of continuous variable(s). To measure gentrification, we thus develop a spatial Markov Chain to examine the ways in which neighborhoods transition between states as a function of their previous state and the states of the surrounding neighborhoods. We develop our model using annual, block-level LEHD data which include information about the location of both workers and employers in the USA. As a result, our model captures a wide variety of crucial information often overlooked in quantitative studies of neighborhood change. We model the nuanced process of residential turnover in concert with economic restructuring using data with high spatial and temporal resolution and we incorporate concepts of neighborhood spillovers into our model. We develop such models for the 15 largest metros in the U.S. and describe how the application of modern geographic data science can lend both insight and forewarning into the process of neighborhood change.