Spatial Clustering of Crime Trends in Urban Environments at Multiple Resolutions

• Activity Number: 531 - Urban Analytics: Modeling and Analysis of High Resolution Urban Data
• Type: Invited
• Date/Time: Thursday, August 6, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Social Statistics Section
• Author(s): Cecilia Balocchi* and Shane Jensen and Edward George
• Companies: University of Pennsylvania and The Wharton School, University of Pennsylvania and University of Pennsylvania
• Keywords: urbanism; crime modeling; Bayesian Nonparametrics; areal units; Dirichlet Process; multiresolution

Abstract

Understanding the spatial distribution of crime dynamics at a city neighborhood level is an important concern for urban planners and law enforcement officials. In particular, finding clusters of neighborhoods that display similar patterns allows for better interpretation and aids with dimensionality reduction. However, data is often aggregated at multiple granularities; for example, US cities are divided into the hierarchy of census tracts, block groups, and blocks, and the granularity level for the analysis is usually chosen arbitrarily. Instead of fixing the resolution level, we simultaneously partition each level of a city into regions with similar crime frequencies. Our model, which combines Nested and Hierarchical Dirichlet Processes, allows the sharing of information between partitions at different resolutions. These partitions also allow us to understand whether one level is more appropriate overall or if the spatial variation of crime happens at different resolutions in different parts of the city. We analyze crime frequencies in the City of Philadelphia over the 2006-18 period at the census tract, block group, and block levels.