Can we Trust Crime Predictors and Crime Categories? Expansions on the Potential Problem of Generalization

Abstract

City-driven open data initiatives have made spatially referenced crime and risk factor data more readily available online, allowing for significance tests to determine the relationship between environment and crime. This paper uses a variety of open source data to assess risk factors for specific violent crime types (assault, homicide, rape, robbery) in three different cities. The results contribute to our understanding of 1) variation in intra-city risk factors for each violent crime type, 2) the degree of spatial overlap for high-risk places for each violent crime type within a city, and 3) the generalizability of risk factors across crime types and cities. Risk Terrain Modeling (RTM) was used to determine the risk factors associated with each crime type at the micro-level and conjunctive analyses of case configurations (CACC) determined the unique behavior settings at the highest risk for each specific violent crime in each city. The findings indicate that intra-city risk factors vary greatly for each violent crime, highrisk places for different violent crimes tend to not overlap spatially within a city, and risk factors are not generalizable across crime types or across cities. Researchers and law enforcement need to examine local, crime-specific contexts when assessing crime problems and generating solutions.