Time Series Clustering for Exploring Neighborhood Dynamics: The Case of U.S. Neighborhood Racial and Ethnic Trends, 1990–2020

Abstract

This article introduces a time-series clustering approach for classifying, visualizing, and exploring neighborhood dynamics. We illustrate the method for the case of racial and ethnic dynamics of neighborhoods in 64 U.S. metropolitan areas from 1990 to 2020. We establish typologies of continuous attribute trajectories for the share of Black, White, and Hispanic populations at the census tract level and explore generalizability versus specificity tradeoffs when varying the cluster analysis scale. Our results affirm a consistent decline in White population shares in neighborhoods across most metropolitan areas, accompanied by varied increases in Black and Hispanic populations. We also highlight the importance of metropolitan context in shaping neighborhood trends. While all cities show a trend towards increased diversity, the specific patterns and rates of change vary considerably, highlighting the unique demographic dynamics at play in each metropolitan area. The time-series clustering approach offers some advantages over previously used methods for visualizing and classifying longitudinal neighborhood dynamics like sequence analysis or growth change modeling in that it clusters the full continuous time series and does assume a pre-determined functional form.