Horsefield, Olivia
(2024)
Improving estimates of alcohol-related crime with geographic and data science methods.
PhD thesis, University of Liverpool.
|
Text
201112851_Jul2024.pdf - Author Accepted Manuscript Download (10MB) | Preview |
Abstract
Alcohol is considered by the UK Government as one of the six key drivers of crime. Many studies have recognised alcohol-related crime exists in some areas more than others, implying that certain area-level characteristics, known as geographic determinants, may be driving alcohol consumption and alcohol-related crime. The geographic determinants of alcohol-related crime include the availability of alcohol in an area and area-level deprivation amongst others. Previous studies have focused on finding the main geographic determinant of alcohol-related crime to find the most effective approaches needed for crime prevention. These studies have not always considered the role of local context and how different areas have different crime rates and characteristics, meaning that geographic determinants may have varying impacts depending on the local context. As space is expected to play a significant role in understanding these area-level drivers of alcohol-related crime, space needs to be integrated into any analysis. Moreover, in order to understand alcohol-related crime and its geographic determinants, we need access to data on alcohol-related crime. Although police data is the best source of detailed spatial data on alcohol-related crime, the measure of this crime type is inaccurate as they are known for being under-counted on the system. The overarching aim of this thesis is to determine if geographic and data science methods can improve estimates and our understanding of alcohol-related crime. The novelty of this work involved implementing these novel and maturing Geographic Data Science (GDS) and Data Science methods to consider these contextual issues and improve data on alcohol-related crime. The first empirical analysis implemented a GDS method called Geographically Weighted Regression to capture the spatially varying associations between geographic determinants and violent crime as a proxy measure of alcohol-related crime. GDS methods revealed how associations for each type of alcohol outlet varied in strength and direction spatially, demonstrating the importance of local context, which hasn’t always been considered in previous studies. High sales alcohol outlets had not previously been considered in the evidence base, and I found these outlets to increase alcohol-related crime in city centre night-time economies and less so elsewhere. In the second empirical analysis, I aimed to improve the identification of alcohol-related crimes in official police records of crime occurrences using text-based algorithms. A Support Vector Machine (SVM) algorithm classified police notes describing crimes investigated as whether they were alcohol-related or not. The SVM estimated a higher proportion of alcohol-related crime (31%) than current police estimates (13%). This helps to build on the wider literature as there are few sources of what figures one would expect to find in the wider evidence base. A final empirical analysis used spatial clustering methods to determine the value that SVM can bring to the police and the wider evidence base’s understanding of alcohol-related crime. The chapter used GDS methods which revealed how the SVM was detecting more harmful alcohol-related crimes types than the police in the most deprived areas. The thesis has demonstrated how GDS and Data Science methods can improve estimates and our understanding of alcohol-related crime, which has implications for the evidence base and police forces. GDS methods have highlighted how alcohol-related crime prevention efforts could be most effective if the local drivers of crime were targeted, rather than a catch all approach of targeting one specific issue. Data Science methods have also shown their potential for detecting alcohol-related crime in large datasets, saving police officers the manual task of recording this crime type, so police time can be spent on higher priority tasks. In combination, GDS and Data Science methods uncovered the spatial distribution of alcohol-related crime and its inequalities, some of which were not fully understood in the current police system. Doing so can assist the police in understanding and targeting this crime type as well as providing insight to the field’s overall understanding and prevalence of alcohol-related crime.
| Item Type: | Thesis (PhD) |
|---|---|
| Uncontrolled Keywords: | alcohol-related crime, machine learning, police data, spatial data science |
| Divisions: | Faculty of Science and Engineering > School of Environmental Sciences |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 27 Aug 2024 09:30 |
| Last Modified: | 08 Feb 2025 03:05 |
| DOI: | 10.17638/03183224 |
| Supervisors: |
|
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3183224 |
Altmetric
Altmetric