It’s likely that you wouldn’t raise an eyebrow at many research findings about criminal activity. Crimes are committed more often in cities. Many less serious crimes occur during the day, while violent offenses are more frequent at night. Yet new research from published by Diep Dao, Assistant Professor of Geography at the University of Colorado Colorado Springs, could turn our current understanding of crime on its head, thanks to a new analytical framework called CrimeScape.
Dao published the findings with Jean-Claude Thill, Knight Foundation Distinguished Professor of Geography at the University of North Carolina at Charlotte, in a publication titled “CrimeScape: Analysis of socio-spatial associations of urban residential motor vehicle theft.”
While the paper focuses specifically on one type of crime — urban residential motor thefts — its innovative analytical method has far-reaching implications.
Crime research frequently uses an existing method called association rule mining to analyze data and draw conclusions. The method is an unsupervised machine learning approach that finds relationships between variables, with a goal of sifting through large volumes of data to help crime researchers find connections between incidents and criminals.
Yet Dao and Thill, who both specialize in geospatial data science, found that association rule mining frequently overlooks an important component: location. So, the researchers added geospatial variables to their exploration of motor vehicle thefts. In doing so, they were able to combine social and spatial analysis — exploring complex interplays between the characteristics of neighborhoods where motor vehicle theft occurred, and the characteristics of those who live there. Dao and Thill dubbed it CrimeScape.
In the CrimeScape framework, neighborhoods are characterized according to their “socioeconomic fabric,” using variables such as average income, education level, employment levels, racial heterogeneity, housing types, home ownership and more. To study urban motor vehicle theft, Dao and Thill then added a new layer: geospatial relationships. As they mapped the neighborhoods and thefts in the study down to the street level, they explored spatial relationships that ranged from the simple — for example, whether or not a street segment was located within a high-income neighborhood — to the complex — for example, whether or not street segments had spillover impact from crime generators and attractors, like alcohol-serving establishments or motels.
From their findings, Dao and Thill extracted 50 rules — simple and complex if/then statements — associated with residential areas at high risk of motor vehicle theft, and 1,000 rules associated with low rates of theft. Then, they embedded a customized graph-based visualization tool into CrimeScape to visually cluster and examine the results.
Their findings not only confirmed previous research on crime, but turned other findings on their heads.
For example, the study confirmed existing literature indicating that residential vehicle theft often occurs in socioeconomically disadvantaged neighborhoods. Areas with “socio-spatial properties” including low income, low rates of home ownership, high rates of multiple-unit housing, high racial heterogeneity and a strong proportion of single-parent families formed the combination most likely to result in high rates of residential vehicle theft. This was an important finding, Dao and Thill write, because it confirmed that their novel analytical framework could accurately replicate existing findings.
But the study also overturned long-standing notions of high-crime areas. For instance, neighborhoods characterized by low income, a high proportion of renters and a racially diverse population are often considered to have higher rates of crime, like motor vehicle thefts. These neighborhoods, often home to communities of color, are frequently disproportionately targeted by police for their supposed crime prevalence. Yet Dao and Thill found that many neighborhoods with these characteristics are actually free from vehicle thefts, thanks to additional crime-preventing characteristics such as low business activity, high employment rates, a strong percentage of the population with a high school degree or more, and high levels of homeownership.
In other words, neighborhoods like these might be targeted by the police without good cause. Indeed, their crime-preventing characteristics could even make them safer than wealthier or socioeconomically advantaged neighborhoods — at least in terms of motor vehicle thefts.
“The interplays among discovered neighborhood characteristics trace a deeper layer of knowledge explaining the distribution of motor vehicle theft that adds contextual depth to the state of theoretical knowledge in this matter,” Dao and Thill write in the paper. “We find that the associative patterns to low motor vehicle theft present an ethical implication against racial profiling and disparate targeting of communities of color by policing practices.”
Importantly, Dao and Thill write that the enhanced spatial association rule mining framework developed in CrimeScape is general enough to be applied to research on other types of crime — and indeed, other types of social science research. By allowing researchers to identify precise, granular socio-spatial factors and then study the interplay, the analytical method can be easily applied to new research projects and fields. Indeed, Dao and Thill write that the new framework constitutes “a leap forward in social science research.”
Now, it only remains to be seen what kinds of new knowledge and understandings the framework generates. After all, nothing is more surprising than realizing how little you know about the place you live.
“CrimeScape: Analysis of socio-spatial associations of urban residential motor vehicle theft” was published in “Social Science Research” in Jan. 2022. Find the full publication online.
Diep Dao is an Assistant Professor within the Department of Geography and Environmental Studies at UCCS. Her research focuses on geographical information science (GIScience), spatial analysis and modeling, spatial big data analytics, urban regional analysis and GPS-based positioning and navigation. Learn more about her research on the UCCS website.