Investigating the Spatiotemporal Distribution and Environmental Risk Factors of Harm-weighted Crime
MetadataShow full metadata
Background: Due to the recent, extensive use of geospatial information systems (GIS), questions about spatial criminology can be answered in greater detail than they have been in the past. However, most of this research is focused on single crime categories, or only examines these offenses as if each offense held the same weight relative to the amount of harm that is caused. There are still questions that have not been answered, specifically regarding the degree of variability in offenses in the observed crime hot spots, the amount of harm contained within those hot spots, and whether crime generators and crime attractors associated with hot spots vary based on the types and severity of the crimes that occur there.
Aims: This dissertation ultimately has one aim: to determine if the results in previous crime-related harm spots research are generalizable. It is well known at this time that unweighted crime follows a specific, non-random geographic distribution and is concentrated in time in very few people and places. Such findings have held in varied settings, but only recently has research begun examining how accounting for harm, or crime severity, affects such spatial and temporal patterns. The present dissertation subscribes to the notion that harm is a missing dimension in the geographic analysis of crime. Acknowledging that the narrow body of existing research has identified different non-random distributions of harm, this dissertation attempts to replicate these findings through three research studies focusing on: 1) the distribution of harm in space (Study #1); 2) the distribution and clustering of harm in space and time (Study #2); and 3) the identification of unique combinations of facilities and environmental features that are related to “high-harm” and “low-harm” harm spot locations (Study #3).
Data and Analysis: The data for the dissertation were obtained from public data portals. Specifically, publicly available geocoded facilities data collected and maintained by referenceUSA were one strand of public data used for the dissertation. The facilities selected for the analysis had been found to be associated with crime hot spots in previous research. These facilities include ATMs, convenience stores, drinking establishments, fast food restaurants, gas stations, lodging locations, liquor stores, banks and other financial institutions, pharmacies, police department locations, schools, and smoke shops. The crime data used for the dissertation was obtained from two separate sources. Specifically, the first set of crime data were obtained from the Open Data Portal for Washington, DC, and include all crimes reported to the police in 2016. The second set of crime data was obtained via an open records request submitted to the Austin (Texas) Police Department and include all calls for service from January 1, 2007 to December 31, 2017. The crime types included in the three studies described below were arson, aggravated assault, burglary, homicide, motor vehicle theft, robbery, sex assault, theft from a motor vehicle, and larceny/theft. These data were geocoded for the purposes of mapping harm spots.
The weights were calculated also using publicly available data/tools, namely the average recommended sentence for UCR Part I Index Crimes from the United States (U.S.) Sentencing Guidelines (United States Sentencing Commission, 2016), Wolfgang, Figlio, Tracy, and Singer’s (1985) seriousness scale, and the Cambridge Crime Harm Index developed by Sherman, Neyroud, and Neyroud, (2016).
Descriptive analyses, correlation analyses, and kernel density estimation are used to both validate the American Crime Harm Index and to identify spatial and temporal distributions of harm. Conjunctive analysis of case configurations (CACC) and logistic regression are used to identify unique combinations of facilities that correlate with the presence of harm spots.
Results: In Study #1, the results suggest harm spots are diffused away from the city center into more residential areas. This implies opportunities for more serious offenses could be higher in residential areas, and that different social ecological processes underlie the spatial distribution of more serious crime. This study also supports the continued use of the Cambridge Crime Harm Index based on the U.S. Sentencing Guidelines. Study #2 examined the spatial, temporal, and spatiotemporal distributions of harm. In general, the results were inconsistent with the distributional findings from Study #1, in that harm ultimately follows a different distribution than raw unweighted crime in space. Harm and unweighted crime generally followed major roadways in Austin. When examining the average harm scores for all the data considered at different time periods, the harm scores were highest most often on Sundays, during the winter months, and in the early mornings.
Study #3 examined possible contextual configurations surrounding harm spots using both logistic regression and CACC. The presence of all facilities, without considering the total count of facilities, was significant for all those included in the model, with the exception of law enforcement agencies. Drinking establishment or lodging facility increased the odds of the street segment having a harm score in the top 33% by approximately six and eight times, respectively. When considering the total count of these facilities, the presence of one additional Drinking establishments, hotels and other lodging facilities, and smoke shops increased the likelihood that a street segment was a harm spot by approximately four and five and a half times.
The CACC results indicate that there is evidence that there is an interactive effect that may result in street segments with harm scores in the top 33%; four of the five case configurations with a relative risk difference of 0.70 or greater included more than one facility type. ATMs were present in all configurations; pharmacies were present in all but one. Drinking establishments and financial institutions were only present in one of these case configurations each.
This dissertation contributes to the existing literature examining the clustering of crime in time and space, but while adding the consideration of the disparities in harm between different crime types. The evidence presented indicates that, in order to gain a full understanding of how dangerous an area is, it is important to consider both the raw crime counts and the weighted harm scores. Considering the relative harm of each offense provides a comprehensive assessment of a hot spot, both in time and space, as it not only considers the number of offenses that occur there, but the type of offense and how objectively dangerous these offenses are. Implications for theory, research, and policy are also detailed.