Modelling Crime: A Spatial Microsimulation Approach

Modelling Crime: A Spatial Microsimulation Approach

Charatdao Kongmuang

Submitted in accordance with the requirements for the degree of Doctor of Philosophy

The University of Leeds School of Geography September 2006

The candidate confirms that the work submitted is her own and that appropriate credit has been given where reference has been made to the work of others.

This copy has been supplied on the understanding that it is copyright material and that no quotation from the thesis may be published without proper acknowledgement.

i

Academic Acknowledgements a. The 2001 Census statistics used in this thesis are Crown Copyright produced by the Office for National Statistics (ONS). Licensed for academic use by the ESRC/JISC Census Programme, which funded access to the data for researchers in the UK, free at the point of use. The ESRC/JISC Census Programme funds the Data Support Units which provide access to UK Census Data. The 2001 Census Area Statistics are provided by the Census Dissemination Unit (CDU) through the Manchester Information and Associated Services (MIMAS) of Manchester Computing, University of Manchester through an interface called CASWEB. b. All maps are based on data provided by the United Kingdom Boundary Outline and Reference Database for Education and Research Study (UKBORDERS) via Edinburgh University Data Library (EDINA) with the support of the Economic and Social Research Council (ESRC) and the Joint Information Systems Committee (JISC) and boundary material which is copyright of the Crown, Post Office and the EDLINE consortium. c. The 2001/2002 British Crime Survey, material from Crown Copyright records made available through the Home Office and the UK Data Archive has been used by permission of the Controller of Her Majesty’s Stationery Office and the Queen’s Printer for Scotland. d. The recorded crime datasets (2000/01, 2001/02, 2002/03, and 2003/04) and known offender dataset (2000-2004) are provided by West Yorkshire Police. e. Gipton and Harehills NRAs boundaries are in dBase and Shapefiles formats provided by Leeds Statistics, Leeds City Council’s Neighbourhood and Housing Department. All Neighbourhood Renewal Areas maps are based upon Ordnance Survey material with permission of Ordnance Survey on behalf of the Controller of Her Majesty's Stationery Office.

ii

Acknowledgements This research would not have been possible without the help, support and advice of many people whom I am extremely grateful. First and foremost, I would like to express my gratitude to my supervisors, Prof. Graham Clarke and Dr. Andy Evans for working so hard on reading my thesis and appreciating what this thesis is all about. Their advice, guidance, and comments have been invaluable and helped me to complete this research. Thanks to my research support group, Dr. Dimitris Ballas and Dr. Debbie Phillips for their advice. In particular, thanks to Dimitris for helping me to build on his spatial microsimulation work and other suggestions at various stages. I would also like to express my gratitude to Prof. Phil Rees not only for examining this work, but also for his kindness, comments, and suggestions which were also of great value. I would like to acknowledge the funding that was provided by Naresuan University, Thailand. Without such generous financial support my PhD would not have been possible. Thanks P’Bid, Assoc. Prof. Kanchalee Jetiyanon, for her long-distance support and caring. Thanks Nikki for sharing lots of experiences when we were in the UK. It is very hard to individually acknowledge everyone who has helped me to reach this point. However, I would like to thank all the people who have directly or indirectly helped me in completing this thesis. Firstly, thanks Pete, Dan, Alison, Dianna, and Andy Turner for your very helpful comments at various stages. Thanks Jin for sharing your programming expertise. Secondly, thank to those with whom I have shared the last four years: Dear Thai friends, you are too many to list here, but you know who you are. Thank you my other friends at the School of Geography for providing an inspiring and friendly atmosphere. A further special thanks to Jin, my very best friend. Your help, care, support, company and patience over the past two and a half years has been wonderful. Our friendship will never be forgotten I promise. Last, but not least, I would like to thank my family for their love, support, and encouragement that create whom I am. They have always been there when I needed them. People always ask me why I am doing a research on crime. My dad, Police Major General Sanam Kongmuang, is my inspiration. This thesis is the result of a four-year long journey in Leeds, a chapter of my life which I will always remember with a smile.

iii

Abstract Spatial microsimulation offers a potentially powerful framework for modelling crime at small area levels. It is more powerful than traditional crime analysis in that it can make policy-centred predictions.

This thesis presents SimCrime; a spatial microsimulation model for modelling crime and an analysis using this software at the ward level in Leeds. The model is based on the UK 2001 Census and the 2001/2002 British Crime Survey. The model effectively adds ‘geography’ to the British Crime Survey data (which is not currently released below the national level). Adding geo-references into the British Crime Survey makes it more valuable, with the spatial aspect of the data enabling an analysis of the geographical variations of factors of interest to policy-makers at a range of scales.

Within this innovative framework, victims, offenders and locations are examined. The results show the geographical distribution of the likelihood of being a victim of burglary dwelling, the risks for different types of household and the reporting rate of each crime type. The addition of a spatial interaction model allows for the analysis and prediction of offender flows, and in combination with the microsimulation of victims, the complete framework thus provides a predictive capacity which can be used to inform policy making.

iv

Table of Contents Academic Acknowledgements………………………………………………………………..i Acknowledgements…………………………………………………………………………...ii Abstract………………………………………………………………………………………iii Table of Contents…………………………………………………………………………….iv List of Figures……………………………………………………………………………….vii List of Tables…………………………………………………………………………………x Abbreviations………………………………………………………………………………..xii Chapter 1: Introduction…….…………………………………………………………….... 1 1.1 Background………………………………………………………………………......... 1 1.2 Aims and Objectives of the Research…………………………………………………. 3 1.3 Thesis Structure ……………………………………………………………………... 4 Chapter 2: Geography of Crime and Modelling Crime: A Literature Review …………8 2.1 Introduction……………………………………………………… ……………………. 8 2.2 The Geographical Approaches to Crime………………………………………………. 8 2.3 Crime Attributes……………………………………………………………………… 11 2.3.1 Demographic Characteristics………………………………………………... 11 2.3.2 Socio-economic Characteristics……………………………………………... 12 2.3.2.1 Employment………………………………………………………… 12 2.3.2.2 Tenure Type……………………………………………………….... 13 2.3.2.3 Poverty and Deprivation………………………………………….… 13 2.3.3 Neighbourhood/Area Characteristics and Offence Locations...…………...... 15 2.4 Movement of Offenders: Journey to Crime………………………………………….. 17 2.5 Crime Victimisation………………………………………………………………….. 18 2.6 Modelling Crime……………………………………………………………………... 20 2.6.1 Multivariate Regression Model……………………………………………....20 2.6.2 Poisson and Negative Binomial Regression Model…………………………. 25 2.6.3 Logistic Regression Model……………………………………………… …. 26 2.7 Concluding Comments………………………………………………………...…....... 29 .

..

Chapter 3: Microsimulation Modelling: A Literature Review……………………….… 31 3.1 Microsimulation: An Introduction……………………………………………....…… 31 3.2 Types of Microsimulation…………………………………………………………..... 31 3.2.1 Static and Dynamic……………………………………………………….…. 32 3.2.2 Spatial and Aspatial………………………………………………….……… 34 3.3 A Review of Selected Microsimulation Models…………………………………….. 35 3.4 Advantage and Disadvantages………………………………………..………….…... 40 3.5 The Creation of Synthetic Microdata……………………………………………..….. 41 3.5.1 Synthetic Reconstruction…………………………………………………..... 41 3.5.2 Combinatorial Optimisation…………………………………….…………... 42 3.6 Combinatorial Optimisation using Simulated Annealing Method…………………… 45 3.7 Concluding Comments…………………………………………………….……….… 48 Chapter 4: Modelling Crime: Data Sources and Issues…………………………...……. 49 4.1 Introduction……………………………………………………………..…….….…... 49 4.2 Data Sources and Issues……………………………………………………..…….…. 49 4.2.1 The 2001 Census………………………………………………………..…… 49 4.2.2 The 2001/2002 British Crime Survey…………………… ………...………... 55 4.2.3 Police Recorded Crime Datasets……………………………………..……....60 4.2.4 Offender Dataset………………………………………………………….…. 63 4.3 Concluding Comments……………………………………………………………….. 64 .

v Chapter 5: Geography and Determinants of Crime in Leeds………………………….. 66 5.1 Introduction…………………………………………………………………..….…… 66 5.2 Leeds Crime Figures and Trends…………………………………………….….….. . 67 5.3 Geographical Variations………………………………………………………………73 5.3.1 Police Divisions………………………………………………………...…… 73 5.3.2 Wards…………………………………………………………………...…… 78 5.4 Findings by Crime Type……………………………………………………………… 80 5.4.1 Burglary Dwelling……………………………………………………………80 5.4.2 Burglary Elsewhere………………………………………………………….. 81 5.4.3 Criminal Damage……………………………………………………………. 81 5.4.4 Drug Offences……………………………………………………………….. 82 5.4.5 Fraud and Forgery…………………………………………………………… 82 5.4.6 Handling……………………………………………………………………... 83 5.4.7 Homicide…………………………………………………………………….. 83 5.4.8 Other Crime…………………………………………………………………..84 5.4.9 Other Theft…………………………………………………………………... 84 5.4.10 Robbery……………………………………………………………………… 85 5.4.11 Sexual Offences……………………………………………………………... 85 5.4.12 Theft from Motor Vehicle…………………………………………………… 86 5.4.13 Theft of Motor Vehicle……………………………………………………… 86 5.4.14 Violent Crime……………………………………………………………….. 87 5.5 Known Offenders and Victims Characteristics……………………………...………. 87 5.5.1 Known Offenders Characteristics…………………………………………… 87 5.5.2 Known Victims Characteristics……………………………………………... 91 5.6 The Relationship between Crime and its Related Determinants………………...…... 93 5.6.1 Population/Household Density………………………...….……...…………. 95 5.6.2 Demographic Characteristics……………………………………...……....… 96 5.6.3 Percentage of Students……………………………………………...……….. 97 5.6.4 Rented Tenure Type………………………..………………………………... 98 5.6.5 Number of Cars per Household……………………………….…………...... 98 5.6.6 Unemployment………………………………………………………………. 99 5.6.7 Deprivation…………………………………………………………….........101 5.6.8 Number of Offenders……………………………………………………..... 103 5.6.9 Multiple Regression Model.………………………………………………...104 5.7 Concluding Comments………………………………………………………….…...107 Chapter 6: SimCrime: A Spatial Microsimulation Model for Crime in Leeds….……108 6.1 Introduction…………………………………………………………………………. 108 6.2 SimCrime Model Specification……………………………………………………... 108 6.2.1 Input………………………………………………………………………... 109 6.2.2 Input Adjustment……………………………………………………………114 6.2.3 Model Execution Process…………………………………………………... 116 6.2.4 Model Output………………………………………………………………. 121 6.3 Evaluation of Synthetic Microdata…………………………………………………..124 6.4 Concluding Comments……………………………………………………………… 130 Chapter 7: Modelling Crime at the Small Area Level…………………………...……..131 7.1 Introduction…………………………………………………………………………. 131 7.2 Comparing the Victim Estimation and Police Recorded Crime……………………. 131 7.3 Victim Estimation…………………………………………………………………... 132 7.3.1 Victim of Burglary Dwelling Estimation………………………………...… 133 7.3.2 Index of Wealth…………………………………………………………..…139 7.3.3 Burglary Dwelling Victimisation Rate…………………………………….. 143 7.3.4 Risk of Becoming a Victim of Burglary Dwelling: Household Characteristics Most at Risk………………………….………... 145

vi 7.4

7.5 7.6

Recorded Crime Estimation………………………………………………………… 148 7.4.1 Recorded Burglary Dwelling Estimation…………………………………... 148 7.4.2 High Risk Areas for Burglary Dwelling…………………………………… 151 Reporting Crime in Leeds…………………………………………………………... 154 Concluding Comments………………………………………………………………156

Chapter 8: Movement of Offenders and Spatial Interaction Modelling……….…….. 158 8.1 Introduction…………………………………………………………………………. 158 8.2 Movement of Offenders…………………………………………………………….. 159 8.2.1 Crime Travel Patterns……………………………………………………… 159 8.2.2 Crime Type Areas………………………………………………………….. 162 8.2.3 Offender Flows…………………………………………………………….. 166 8.2.4 Inflow/Outflow Ratio………………………………………………………. 171 8.2.5 Self-Containment…………………………………………………………... 172 8.3 Spatial Interaction Model for Burglary Dwelling: Model Specification…………… 174 8.3.1 Model Formulation………………………………………………………… 175 8.3.2 Attractiveness Factors……………………………………………………… 176 8.3.3 Scaled Attractiveness Factors……………………………………………… 177 8.3.4 Model Calibration………………………………………………………….. 181 8.3.5 Goodness-of-fit Statistic…………………………………………………… 184 8.3.6 Model Summary…………………………………………………………… 185 8.4 Concluding Comments……………………………………………………………… 190 Chapter 9: What-if Analyses………………………………………………………….…. 191 9.1 Introduction………………………………………………………………………..... 191 9.2 Policy and Scenario Issues………………………………………………………….. 191 9.3 Modelling the Neighbourhood Renewal Strategy…………………………………... 196 9.3.1 Scenario 1: Economic Activity Changes in Gipton NRA………………….. 196 9.3.2 Scenario 2: Economic Activity Changes in the Harehills NRA…………… 201 9.3.3 Scenario 3: A Reduced Number of Offenders Committing Burglary Dwelling in the Gipton NRA………….....……….……………… 204 9.3.4 Scenario 4: A Reduced Number of Offenders Committing Burglary Dwelling in the Harehills NRA……………………….…………. 208 9.4 Concluding Comments……………………………………………………………… 211 Chapter 10: Conclusions……………………………………………………………...…. 215 10.1 Introduction…………………………………………………………………………. 215 10.2 Summary of the Research Findings………………………………………………… 215 10.3 Evaluation and Limitations of the Research………..………………………………. 219 10.4 Possibilities for Future Research…………………………………………………….221 10.5 Concluding Statements………………………………………………………………222 References………………………………………………………………………………… 223 Appendix A: Recorded Crime by Crime Type at Ward Level…………….…...………… 236 ..

Appendix B: Crime Rate per 1,000 Population by Crime Type at Ward Level……....….. 240 Appendix C: Demographic and Socio-economic Variables from the 2001 Census…...… 242 Appendix D: Model File……………………………………………………………....….. 244

vii

List of Figures Figure 1.1: Framework for modelling crime at the small area level………………………….3 Figure 1.2: Chapter linkage…………………………………………………………………...7 Figure 2.1: A model of repeated, multiple crime victimisation…………………………….. 22 Figure 3.1: Microsimulation procedure for the allocation of employment status…………...42 Figure 3.2: A simplified combinatorial optimisation process…………………...…………. 44 Figure 3.3: Flowchart of simulated annealing algorithm…………………………...……… 47 Figure 4.1: Discrepancies in census counts between tables………………………………… 55 Figure 5.1: Leeds Crime……………………………………………………………………. 68 Figure 5.2: Leeds Police Division…………………………………………………………...73 Figure 5.3: Crime rates per 1,000 population by police division……………………………75 Figure 5.4: Percentage changes (between 2002/03 and 2003/04) of selected crime types….76 Figure 5.5: Leeds wards ……………………………………………………………………. 77 Figure 5.6: Crime rate per 1,000 population (2003/04)…………………………………….. 79 Figure 5.7: Burglary dwelling: a) Rate per 1,000 households 2003/04 b) Change, 2002/03 to 2003/04.…...... 80 Figure 5.8: Burglary elsewhere: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 81 Figure 5.9: Criminal damage: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 81 Figure 5.10: Drug offences: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 82 Figure 5.11: Fraud and forgery: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 82 Figure 5.12: Location of homicides from 2000/01 to 2003/04……………………………... 83 Figure 5.13: Other crime: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 84 Figure 5.14: Other theft: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 84 Figure 5.15: Robbery: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 85 Figure 5.16: Sexual offences: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 85 Figure 5.17: Theft from motor vehicle: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 86 Figure 5.18: Theft of motor vehicle: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 86 Figure 5.19: Violent crime: a) Rate per 1,000 population 2003/04 b) Change, 2002/03 to 2003/04……… 87 Figure 5.20: ‘Population density’ and burglary dwelling…………………………………... 95

viii Figure 5.21: ‘Household density’ and burglary dwelling…………………………………... 95 Figure 5.22: Distribution of ‘young adult’ and burglary dwelling………………………..... 96 Figure 5.23: Distribution of ‘male young adult’ and burglary dwelling……………...……. 96 Figure 5.24: ‘Percentage of students’ and burglary dwelling………………………………. 97 Figure 5.25: ‘Rented tenure’ and burglary dwelling…………………………………...…... 98 Figure 5.26: ‘Number of cars per household’ and crime…………………………………… 99 Figure 5.27: ‘Number of cars per household’ and criminal damage…………………...….. 99 Figure 5.28: Distribution of ‘unemployment rate’ in Leeds………………………………. 100 Figure 5.29: ‘Unemployment’ and criminal damage…………………………………...… 100 Figure 5.30: ‘Male-unemployed’ and criminal damage…………………...……………... 100 Figure 5.31: Correlation between ‘male unemployed’ and criminal damage……...……... 101 ..

Figure 5.32: ‘Index of Multiple Deprivation’ and criminal damage………………...……. 101 Figure 5.33: ‘Number of offender’ and criminal damage…………………...……………. 103 ..

Figure 5.34: Correlation between ‘number of offenders’ and criminal damage……...…... 103 Figure 6.1: Constraint table adjusted method…………………………………………….. 115 Figure 6.2: The process to check each individual fits the column constraints…………… 118 Figure 6.3: SimCrime Framework………………………………………………………... 120 Figure 6.4: Distribution of female single, widowed, or divorced aged 25-49 living in rented house by output area in Leeds…………………………………...……. 122 Figure 6.5: Distribution of full-time student aged 20-30 living in rented house by output area …………………………………………………………………... 122 Figure 6.6: Distribution of high-class households with owner occupier having at least 1 car……………………………………………………………………………123 Figure 6.7: Distribution of male aged 16-24 unemployed and living in the rented house by ward…………...…………………………….………………………123 Figure 6.8: Spatial distribution of SAE for age and sex by living arrangement at output area level……………………………………………………………….128 Figure 6.9: Spatial distribution of SAE for NS-SEC by tenure type at output area level… 128 Figure 6.10: Spatial distribution of SAE for tenure type and car or van availability by economic activity at output area level……………………………………. 129 Figure 6.11: Spatial distribution of SAE for all constraints at output area level………….. 129 Figure 7.1: Crime recording process………………………………………………………133 Figure 7.2: Comparing estimated number of victims and police recorded burglary dwelling 2001/02………………..………………………………….. 134 Figure 7.3: Index of Multiple Deprivation by ward……………………………………….136 Figure 7.4: Number of burglars resident by ward………………………………………… 136 Figure 7.5: Self-containment for burglary dwelling by ward………………………...…... 137 Figure 7.6: Relationship between self-containment of burglary dwelling and Index of Multiple Deprivation…………………………………………………………. 137 Figure 7.7: Catchment area of burglary dwelling for Roundhay ward………...………… 138

ix Figure 7.8: Leeds surrounding areas……………………………………………………… 140 Figure 7.9: Index of Wealth………………………………………………………………. 140 Figure 7.10: Estimated victim rate per 1,000 households by ward of burglary dwelling in Leeds………..…………………………………………………………….. 143 Figure 7.11: Stepwise multiple regression analysis……………………………………….. 149 Figure 7.11: Estimated burglary dwelling rate per 1,000 households……………...……... 153 Figure 7.12: Recorded burglary dwelling rate per 1,000 households (2001/02)………….. 153 Figure 7.13: SimCrime estimated reporting rate of each crime type comparing with the results from the BCS……...………..……………………………………. 154 Figure 8.1: Distance decay for burglary dwelling……………………………...….……… 161 Figure 8.2: Area classification of output areas: Super-group……………………….…….. 164 Figure 8.3: Catchment area for crime in Burmantofts…………………………………….. 170 Figure 8.4: Catchment area for crime in Headingley………………………………..……..170 Figure 8.4: Relationship between ‘inflow/outflow ratio’ and ‘degree of self-containment’…………………………………………………………....... 174 Figure 8.5: Burglary dwelling flows to be compared in the calibration process…………...183 Figure 9.1: Neighbourhood Renewal Areas (NRAs) in Leeds……………………………. 193 Figure 9.2: Effects of socio-economic changes on crime…………………………………. 194 Figure 9.3: Spatial distribution of people aged 16-74 and economically active by output area……………………………………………………………………...195 Figure 9.4: Spatial distribution of unemployed people aged 16-74 by output area……….. 195 Figure 9.5: Gipton Neighbourhood Renewal Area………………………………………... 197 Figure 9.6: How number of victims of burglary dwelling in the Gipton NRA would change under Scenario 1…………………………………………………...….. 199 Figure 9.7: Change in the number of victims of burglary dwelling in the Gipton NRA…. 199 Figure 9.8: Harehills Neighbourhood Renewal Area……………………………………... 200 Figure 9.9: How the number of victims of burglary dwelling in the Harehills NRA Would change under Scenario 2……………………………………..…….…. 202 Figure 9.10: The changing number of victims of burglary dwelling in the Harehills NRA under Scenario 2……………...………………………………203 Figure 9.11: Decrease in burglary dwelling under scenario 3……...…………………….. 205 Figure 9.12: ‘Percentage share’ change under Scenario 3………………………………… 207 Figure 9.13: Decrease in burglary dwelling under Scenario 4…………………………..... 208 Figure 9.14: ‘Percentage share’ change under Scenario 4………………………………… 210 Figure 9.15: Change in recorded burglary dwelling (2001/02 and 2003/04)……………... 213 Figure 9.16: Change in ‘percentage share’ of recorded burglary dwelling (2001/02 and 2003/04)……………………………………………………….. 213

x

List of Tables Table 1.1: Thesis outline……………………………………………………………………... 4 Table 2.1: Best fitting final model for HIA data after including spatially averaged variables………………………………………………………………………..... 27 Table 2.2: Crime attributes…………………………………………………………………. 30 Table 3.1: Requirement profile for a static microsimulation model………………………... 33 Table 3.2: Synthetic reconstruction versus combinatorial optimisation……………………. 45 Table 4.1: Topics in the 2001 Census………………………………………………………. 51 Table 4.2: Census Area Statistics dataset tables available from CASWEB ………………...53 Table 4.3: Selected topics in the British Crime Survey…………………………………….. 56 Table 4.4: Comparing the British Crime Survey and police recorded crime………………..60 Table 4.6: Details on recorded crime……………………………………………………….. 63 Table 4.7: Detailed data on known offenders………………………………………………. 64 Table 5.1: Number of crime and crime rate (all crime) in Leeds compared to West Yorkshire and England and Wales………………………………………………. 67 Table 5.2: The effects of NCRS on West Yorkshire……………………………………….. 68 Table 5.3: Leeds Crime figures and trends (2000/1-2003/04)……………………………… 69 Table 5.4: Recorded crime in England and Wales by offence 2000/01 to 2003/04 and percentage change between 2002/03 and 2003/04………………………………. 70 Table 5.5: Detection rates in Leeds compared to England and Wales………………………72 Table 5.6: Number of crime by division (2000/01 to 2003/04)…………………………….. 74 Table 5.7: Crime rates (per 1,000 population) by division (2000/01 to 2003/04)………….. 74 Table 5.8: Leeds Crime rates by ward (2003/04)………………………………………..…. 79 Table 5.9: Age group of known offenders (2000-2004)………………………………...….. 88 Table 5.10: Gender of known offenders (2000-2004)…………………………………...…. 88 Table 5.11: Ethnicity of known offenders (2000-2004)……………………………...….…. 89 Table 5.12: Age groups of known victims (2003/04)………………………………..…..…. 90 Table 5.13: Gender of known victims (2003/04)….…………………………………..……. 91 Table 5.14: Victimisation by major crime types and ethnic group in Leeds (period 2001/02, 2002/03, and 2003/04)…………………………………..……92 Table 5.15: Correlation coefficient of crime and its related determinants…………………. 94 Table 5.16: Indices of deprivation for Leeds wards 2000 (ranking from high to low)…….102 Table 5.17: Burglary dwelling model (model summary)…………………………………..105 Table 5.18: Coefficients of the models……………………………………………………. 106 Table 6.1: SimCrime constraint variables…………………………………….………….. 112 Table 6.2: SimCrime constraint tables……………………………………….……..….… 113 Table 6.3: Comparing the distribution of constraint table and synthetic microdata to get the Total Absolute Error (TAE)……………………………………..….. 125

xi 6.3a: Constrainted table………………………………………………….…… 125 6.3b: The distribution of synthetic population……………………………….. 125 6.3c: Compare constraint table and the synthetic microdata to get TAE of each area………………………………………………………………... 125 Table 6.4: Standardised Absolute Error (SAE) between runs………………..…………… 127 Table 7.1: Victim of burglary dwelling estimation from the SimCrime………………...... 134 Table 7.2: Index of Wealth values………………………………………………………… 142 Table 7.3: Victimisation rate of burglary dwelling (estimation)……...………………….. 144 Table 7.4: Proportion of households being victims of burglary dwelling by household type of the 2001/2002 British Crime Survey…………...………….. 145 Table 7.5: Propensity of household victims of burglary dwelling in Leeds………………. 147 Table 7.6: ‘Recorded burglary dwelling’ estimation…………...………………..……….. 150 Table 7.7: High risk areas of burglary dwelling in Leeds…………………………...……. 152 Table 8.1: Average distance travelled of known offender………………………………… 159 Table 8.2: Percentage of offender committed crime………………………………………. 160 Table 8.3: Crime type areas for overall crime…………………………………………….. 163 Table 8.4: Crime type areas for burglary dwelling……………………………………...... 163 Table 8.5: The National Classification of Census Output Areas (Super-Group) aggregated to ward level………………………………………………….…… 165 Table 8.6: Distance between origin ward (i) to destination ward (j)……………………… 167 Table 8.7: Offender flows of all crime in Leeds…………………………………………... 168 Table 8.8: Offender flows of burglary dwelling in Leeds………………...………………. 169 Table 8.9: Self-containment of crime in Leeds by ward………………………………….. 173 Table 8.10: Correlations between potential attractiveness factors and inflow of burglary dwelling…………………………………………………………...... 176 Table 8.11: The values calculated for 180

γj

for 33 wards…………………………………….

Table 8.12: Model summary………………………………………………………………. 187 Table 8.13: Predicted burglary dwelling flows…………………………………………….188 Table 8.14: Predicted inflows burglary dwelling by ward………………………...……… 189 Table 9.1: SimCrime attributes……………………………………………………………. 192 Table 9.2: Economic activity in Gipton NRA and in Leeds………………………………. 198 Table 9.3: Scenario 1……………………………………………………………………… 198 Table 9.4: Economic activity in Harehills NRA and in Leeds……………………………. 201 Table 9.5: Scenario 2……………………………………………………………………… 201 Table 9.7: Estimated burglary dwelling flows under Scenario 3………………………… 206 Table 9.8: Scenario 4……………………………………………………………………… 208 Table 9.8: Estimated burglary dwelling flows under Scenario 4…………………………. 209 Table 9.9: Recorded burglary dwelling (2001/02 and 2003/04)………………………….. 212 Table 9.10: Trends of ‘percentage share’ of recorded burglary dwelling by ward

xii (between 2001/02 and 2003/04)………………………………………………. 214

Abbreviations ACORN

A Classification of Residential Neighbourhoods

ACPO

Association of Chief Police Officer

BCS

British Crime Survey

BHPS

British Household Panel Survey

CAS

Census Area Statistic

CASWEB

Census Area Statistics Website

CCD

Census Collection District

CCG

Centre for Computational Geography

CPUs

Central Processing Units

DETR

Department of the Environment Transport and the Regions

DYNASIM

Dynamic Simulation of Income Model

ED

Enumeration District

GAM

Geographic Analysis Machine

GIS

Geographic Information System

HIAs

High Intensity Crime Areas

HMP

Her Majesty's Prison

IDS

Income Distribution Survey

IMD

Index of Multiple Deprivations

IPF

Iterative Proportional Fitting

IW

Index of Wealth

JSA

Job Seekers Allowance

LSP

Local Strategic Partnership

MicroMaPPAS Micro-simulation Modelling and Predictive Policy Analysis System NATSEM

National Centre for Social and Economic Modelling

NCRS

National Crime Recording Standard

NRA

Neighbourhood Renewal Area

NRU

Neighbourhood Renewal Unit

NS-SEC

National Statistics Socio-Economic Classification

OA

Output Area

ODPM

Office of the Deputy Prime Minister

ONS

Office for National Statistics

SAE

Standardised Absolute Error

SAR

Sample of Anonymised Records

SAS

Statistical Analysis System

xiii SAS

Small Area Statistics

SDSS

Spatial Decision Support System

SEG

Socio-Economic Group

SLA

Statistical Local Area

SMILE

Simulation Model for the Irish Local Economy

SPSS

Statistical Package for the Social Sciences

SRMSE

Standardised Root Mean Square Error

STAC

Spatial and Temporal Analysis of Crime

STINMOD

Static Income Model

SVERIGE

System for Visualising Economic and Regional Influences in Governing the Environment

SYNAGI

Synthetic Australian Geo-demographic Information

TAE

Total Absolute Error

TOPSWING

Total Population of Sweden Individual and Geographical database

UK

United Kingdom

WYP

West Yorkshire Police

Chapter 1- Introduction

1

Chapter 1 Introduction 1.1 Background 1.2 Aims and Objectives 1.3 Thesis Structure

1.1 Background Crime is one of the most important problems facing the United Kingdom today. Crime can result in economic, physical and emotional suffering for victims and can have wider social and economic impacts on areas. Consequently crime has become a major issue of public policy. Although crime has been studied for a long time, most research centres on patterns of offences and crime prevention rather than offenders and victims. Crime is a social phenomenon involving people (offenders and victims) and places, yet a long-standing criticism of the official statistics (see for example, McClintock and Avison, 1968) has been that they do not give a clear picture of the social or situational context of crimes, or of the likelihood of different kinds of people becoming victims (Maguire, 2002). Crime victimisation is unevenly distributed across populations, time and space. Who is at greatest risk of becoming a victim? How is victimisation distributed across space and what are the demographic dimensions? The British Crime Survey (BCS), organised annually by the UK’s Home Office, provides rich information about levels of crime and crime victimisation; however, it cannot presently be used to explain crime victimisation for small geographical units. The BCS now provides limited information at the police force area level, but not for smaller geographical areas.

“Macro studies have been criticised for their failure to explain or accurately forecast more recent trends in crime and also for the lack of practical policy implications that can be drawn form their findings.”

(Hansen and Machin, 2003: 4) Typical approaches to modelling crime have involved the use of regression models. These techniques are unable to model policy impacts effectively. Moreover, the spatial scales used in most studies are at the national or local authority district level. In the UK, the smallest area of crime modelling is generally at the police force area level (there are 52 police forces


2

in the UK) and there has only been a limited amount of work carried out so far at this scale (Hansen and Machin, 2003). The work of Hansen and Machin (2003) shows the importance of taking geographical areas into consideration when looking at crime and offending models. Their work suggests that there is a much greater need to provide a geographical breakdown of crime. This is both very plausible because of the large variation in demographic and socio-economic patterns in various parts of the country, and necessary because, as Maguire (2002) comments, there are growing demands for more detailed information at the local area level from multi-agency crime and disorder partnerships. In addition, greater understanding at the smallest level, that of the geo-located individual, would enhance behavioural studies, such as those on offender movements, by allowing comparison with larger scale crime statistics.

‘Microsimulation’ is a methodology aimed at building large-area datasets of individual units such as persons, households or firms (Clarke, 1996; Ballas and Clarke, 2000) and can be used to simulate the effect of changes in policy or other changes on these microunits. It essentially creates individual-level data from example individuals and aggregate statistics, matching the two together and allowing the merging of additional datasets. The microsimulation approach dates back to the work of Orcutt (1957) and Orcutt et al. (1961). It has been increasingly adopted to study the impacts of social and economic policies on individual units (Merz, 1991; Ballas et al., 2005c), mainly for predicting the future effects of changing public policies (Clarke, 1996; Ballas and Clarke, 2001a, b). Spatial microsimulation combines the advantages of aspatial micro-analytical approaches with those of geographical models that take space into account. The key advantage of the spatial microsimulation approach is that it contains geographical information that can be used to investigate the local area impacts of policy changes. Spatial microsimulation is useful for modelling the socio-economic and spatial effects of policy changes at different geographical scales. Due to the advantages that it offers over traditional approaches, spatial microsimulation has become increasingly popular and a powerful tool within applications that have a geographical aspect.

A major challenge in modelling crime in the UK is adding geographical detail into the British Crime Survey, which is not available for small geographical areas. One solution is to attach the information from the BCS to the more geographically disaggregated UK national census data using spatial microsimulation techniques. Blending geo-referenced data into the BCS microdata makes it more valuable and the spatial aspect is capable of providing geographical detail for different scales.


3

Spatial detail is of utmost importance when studying crime. When it comes to offenders, victims, and offences it is important to take account, not only of the number of events, but also their locations. Spatial microsimulation offers a potentially powerful framework for modelling crime victimisation at the small area levels. This also can be linked with spatial interaction models to explain offender flows (Figure 1.1). Spatial interaction modelling is a technique which allows the predictions of flows based on the attractiveness of areas and the constraints on travel between them.

Although social scientists have developed numerous theoretical and empirical models of crime, currently there are no published examples of spatial microsimulation models being applied to study crime. Moreover, there are no examples of crime modelling being linked to spatial interaction models to explain offender flows. This study aims to address these shortcomings.

Figure 1.1: Framework for modelling crime at the small area level

1.2 Aims and Objectives of the Research The principal aim of this thesis is to investigate the potential of spatial microsimulation for modelling crime. In order to achieve this aim the following research objectives were formulated. 1) To review the geography of crime and crime modelling. 2) To review microsimulation models and the procedures involved in creating a synthetic population microdata dataset. 3) To investigate and review the available data for modelling crime in Leeds. 4) To explore the geography of crime in Leeds.


4

5) To study the relationship between crime and its related determinants in the Leeds context. 6) To use the knowledge that has been gained in objectives 1-5 to build SimCrime, a static spatial microsimulation model for crime in Leeds. 7) To use this model to estimate crime victimisation at ward level. 8) To investigate movements of offenders 9) To explore the interaction between the location of offence and the location of offender by linking the microsimulation model with a spatial interaction model. 10) To enable the what-if analysis of a range of policy scenarios. 11) To evaluate the success of the research and propose the possibilities for future work.

1.3 Thesis Structure In order to achieve the research objectives set out in § 1.2, the thesis is organised into ten chapters as outlined in Table 1.1. Each chapter relates to one or more of the research objectives.

Table 1.1: Thesis outline Chapter

Objective

Chapter 2: Geography of Crime and Modelling Crime: A Literature Review

1

Chapter 3: Microsimulation Modelling: A Literature Review

2

Chapter 4: Modelling Crime: Data Sources and Issues

3

Chapter 5: Geography and Determinants of Crime in Leeds

4&5

Chapter 6: SimCrime: A Spatial Microsimulation Model for Crime in Leeds

6

Chapter 7: Modelling Crime at the Small Area Level

7

Chapter 8: Movement of Offenders and Spatial Interaction Modelling

8&9

Chapter 9: What-if Analyses

10

Chapter 10: Conclusions

11


5

Chapter 2 (Geography of Crime and Modelling Crime: A Literature Review) reviews the geographical approaches to crime and modelling crime. Crime victimisation is reviewed and crime attributes are summarised in relation to three main categories: demographic characteristics, socio-economic characteristics, and neighbourhood characteristics for a given offence location. Different types of regression model for modelling crime are presented showing the method, variables used, and findings. The chapter concludes with a table of crime attributes relating to offenders, victims and offence areas derived from the literature reviewed. Chapter 3 (Microsimulation Modelling: A Literature Review) reviews different types of microsimulation model. A number of major microsimulation models are briefly reviewed to show their variables and application areas. The advantages and disadvantages of such models are also summarised. The procedures involved in creating a synthetic population microdata dataset, which can be seen as the most important part of the microsimulation, are also reviewed and compared to select the best method to use for this research (which turns out to be combinatorial optimisation using simulated annealing method). Chapter 4 (Modelling Crime: Data Sources and Issues) provides detail on the data that will be used throughout the thesis, including the 2001 Census, the 2001/2002 British Crime Survey, police recorded crime datasets (2000/01, 2001/02, 2002/03, 2003/04), and a known offender dataset (2000-2004). Details and limitations of the datasets are described and discussed. Chapter 5 (Geography and Determinants of Crime in Leeds) describes Leeds crime figures and trends and the geographical variation at police division and ward levels using the four years worth of recorded crime data detailed in Chapter 4. Findings by crime type are presented. The chapter contains maps and graphs for a better understanding of the spatial patterns. The relationship between crime and its related determinants are explored. The chapter concludes with the variables that need to be included in the spatial microsimulation model to provide reasonable predictions. Chapter 6 (SimCrime: A Spatial Microsimulation for Crime in Leeds) presents SimCrime, a spatial microsimulation for crime in Leeds. The chapter describes in detail the creation of a synthetic microdata dataset which comprises 514,523 individuals aged 16-74 in households in Leeds. The chapter runs through the SimCrime model specification by explaining the inputs, model execution process, and model outputs. The chapter also describes a method to tackle the problem of discrepancies in census counts between tables. The evaluation of the synthetic microdata dataset is also described.


6

Chapter 7 (Modelling Crime at the Small Area Level) demonstrates how SimCrime can be used for modelling crime victimisation rates. The synthetic microdata created by the process described in Chapter 6 provides an estimation of the likelihood of being a victim of burglary dwelling at ward level under the assumption that if the synthetic population from SimCrime has the same characteristics as the population from the BCS, they will have the same propensity to be a victim of crime. The chapter also gives recorded burglary dwelling estimates based on the victim estimations. The chapter proposes a method for comparing the relative wealth of proximal areas using the ‘number of cars in the areas’. It then gives an ‘Index of Wealth’ score by ward which will be used in following chapters to help model criminal flows. The chapter also provides the household characteristics indicating a risk of becoming a victim of burglary dwelling. The reporting rate of each crime type is also estimated and compared with the rate for England and Wales.

Chapter 8 (Movement of Offenders and Spatial Interaction Modelling) explains the movement of offenders including crime travel patterns, crime type areas, offender flows, inflow/outflow ratios, and self-containment rates for wards. It then presents a spatial interaction model for burglary dwelling in Leeds. The chapter details the model formulation, attractiveness factors, model calibration and goodness-of-fit statistics.

Chapter 9 (What-if Analyses) shows how SimCrime (from Chapter 6) and the spatial interaction model (from Chapter 8) can be used together for what-if analysis. The chapter focuses on factors that are most likely to have a major impact on crime. The scenarios presented in this chapter are based on real policies from the Leeds Neighbourhood Renewal Strategy. SimCrime is used to estimate the impact of changing the socio-economic structure of the Gipton and Harehills Neighbourhood Renewal Areas in Leeds: specifically, how this is likely to change the number and location of victims of burglary dwelling. The chapter also gives examples of analyses in which the spatial impacts of a change in the offender flows are investigated using the synthetic populations and the spatial interaction model.

Chapter 10 (Conclusions) provides a conclusion to the study by summarising the findings of the research. The chapter reviews how well the aims and objectives have been fulfilled. It then moves on to limitations of the research. The chapter finally looks to the possibilities for future research and ends with some concluding statements.

The way in which each chapter links to others is summarised in Figure 1.2.


7

Chapter 1: Introduction

Chapter 2 Review Geography of crime and modelling crime

Chapter 3

Chapter 4

Available data The 2001 Census The 2001/2002 BCS Recorded crime datasets Known offender dataset

Review Microsimulation Models and the creation of Synthetic Microdata

Chapter 5 Studying geography of crime in Leeds Exploring relationship between crime and its related determinants Finding variables to be included into the microsimulation model

Chapter 6 Building SimCrime, a spatial microsimulation for crime in Leeds

Chapter 7 Estimate ‘victim of burglary dwelling’ at ward level Estimate ‘recorded burglary dwelling’ at ward level Provide household characteristics most at risk of becoming a victim of burglary dwelling Estimate ‘reporting rate’ by crime type

Chapter 8 Studying movement of offender and building Spatial Interaction Model

Chapter 9 What-if analyses

Chapter 10: Conclusions

Figure 1.2: Chapter linkage

Chapter 2- Geography of Crime and Modelling Crime: A Literature Review

8

Chapter 2 Geography of Crime and Modelling Crime: A Literature Review 2.1 Introduction 2.2 The Geographical Approaches to Crime 2.3 Crime Attributes 2.3.1 Demographic Characteristics 2.3.2 Socio-economic Characteristics 2.3.2.1 Employment 2.3.2.2 Tenure Type 2.3.2.3 Poverty and Deprivation 2.3.3 Neighbourhood/Area Characteristics and Offence Locations

2.4 Movement of Offenders: Journey to Crime 2.5 Crime Victimisation 2.6 Modelling Crime 2.6.1 Multivariate Regression Model 2.6.2 Poisson and Negative Binomial Regression Model 2.6.3 Logistic Regression Model 2.7 Concluding Comments

2.1 Introduction Crime is a violation of law (Mannhiem, 1965), associated with offenders and victims. It is a problem which can be identified and acknowledged by most people who experience it directly and indirectly. It adds stress to peoples lives and impairs the quality of life of individuals and communities (Smith, 1989). Crime has been studied for almost two centuries (Herbert, 1983) by a variety of different academic disciplines, including criminology, biology, psychology, economics, psychiatry, anthropology, sociology, and, of course, geography. There have been many theoretical and methodological advances that have allowed crime to be studied from a number of different geographical perspectives (Yarwood, 2001). This chapter starts with consideration of the geographical approaches to crime in § 2.2, followed by the exploration of crime attributes such as demographic, socio-economic, and area characteristics and offence locations in § 2.3. Movement of offenders and crime victimisation are also described in § 2.4 and § 2.5 respectively. Crime modelling using regression approaches are reviewed in § 2.6. Finally, § 2.7 is concluding comments.

2.2 The Geographical Approaches to Crime The study of the geography of crime can be traced back to the beginnings of the Cartographic School of criminology in France in 1830, and its subsequent spread to England and other European countries. The Cartographic School was mainly concerned with regional patterns of crime, focusing on where crime occurs. They used maps to show crime rate variations. In the period 1830-1880, most studies on the geography of crime examined the relationship between crime rates and other indicators of social condition along with climatic factors.


9

The spatial distribution of crime has been extensively studied over the decades, particularly in respect to the socio-economic characteristics of the neighbourhoods in which the offender lives or a crime occurs (Harries, 1974; Shaw and McKay, 1969): so-called ecological analysis. Ecological analysis has focused on mapping with modelling often being used to link the distribution of crime to other socio-economic and environmental variables. This approach was first developed in Chicago by Shaw and McKay (1942), who made major contributions to the methodology. They correlated variables such as substandard housing/housing quality, poverty, foreign-born population, unemployment, ethnic minority group and mobility, with high delinquency rates. They demonstrated how the development and persistence of delinquent behaviour is associated with social deprivation, disorganisation, and disadvantage (Shaw and McKay, 1969). Basing their approach on Burgess’s city model, they found that juvenile delinquency was highest in the innermost ring of the city and declined steadily outwards. The Chicago School had a great influence over criminological studies, not only in the United States but also in European countries in the 1950s-1970s.

The fundamental relationship between crime and socio-economic factors remains the core of criminological study. Following Shaw and McKay’s ecological approach, much of the continuing research has focused on the relationship between low socio-economic status, residential instability, and resulting crime. Witt et al. (1999) examined the relationship between economic factors and property crime using aggregated data from 42 police force areas. They were mainly concerned with the impact of unemployment on crime. They found that high crime is associated with increases in male unemployment, high growth in the number of available properties and high wage inequalities. Entorf and Spengler (2000) studied the socio-economic and demographic factors of crime using panel data from the German States. They found socio-economic and demographic factors are important and significant influences. Being young and unemployed increases the probability of committing crime substantially. The concept of ‘place’ is fundamental to ecology analysis. Most of the studies about crime using the spatial ecology approach have found a strong relationship between crime and ‘poor’ environments.

It can be argued that mapping is a valuable tool for revealing patterns of crime. In particular crime mapping in recent years has become increasingly sophisticated by the arrival of Geographic Information System (GIS). The use of GIS has become more prevalent in recent years for capturing, storing, managing and displaying spatial data (Haining, 2003). It has proved useful for crime analysis. With GIS, we can obtain a better understanding of crime


10

from a geographical perspective by using computers to represent and analyse spatially related crime incidents. They have two important functions: 1) to display maps or geographic features such as crime location (points), flows (lines), or areas (polygons); and 2) to use a digital map database to link spatial data to descriptive information. As an example, suppose two datasets are available for an area: number of burglary dwelling incidents, and number of households. These two datasets can be either mapped individually or combined to show a burglary dwelling rate (for example per 1000 households). Moreover, GIS can combine data using either spatial or tabular file attributes and allows the combination of information associated with different geographical boundaries. One of the advantages of using the GIS is the ability to use and display a large amount of data. Information shown on a map can often be better understood than if only displayed in tabular form. In addition, there have been significant advances in the analytical tools such as those for the identification of hotspot areas, where concentrations of crime exist (Read and Oldfield 1995; Longley et al., 2001). Important to such analyses is geographically referenced crime incidents. Hotspot maps are very useful for visualising areas of high crime and they are now widely used by the police. Most police departments have limited resources including manpower; therefore the identification of hotspots helps them to prioritise the needs of areas. There are some well known hotspot identification methods. For example, the Geographic Analysis Machine (GAM) developed by Openshaw and colleagues (Openshaw et al., 1987) can be seen as an early attempt at automated exploratory spatial data analysis (Turton and Openshaw, 2001). It has been used for looking for clusters of events in heterogeneous spatial populations. The Spatial and Temporal Analysis of Crime (STAC) package developed by the Illinois Criminal Justice Information Authority is also very popular. It is one of the most widely used point-based methods for crime analysis. The STAC package was completed in 1988, with support from the U.S. Department of Justice, Bureau of Justice Statistics. Another program developed in the US is CrimeStat, a spatial statistics program for the analysis of crime incident locations and concentration (Levine, 2006). In the UK Hirschfield et al. (1995) examined relationships between the spatial concentration of disadvantaged residents and levels of crime on Merseyside and identified hotspot areas by using the STAC Software. Groff and La Vigne (2001) applied opportunity theories to the study of crime for residential burglary. This research examined the utility of raster–based mapping software for predicting likely and unlikely locations for burglaries, as well as likely locations for crime displacement or diffusion. Craglia et al. (2001) used a GIS-based spatial analysis system to try and model the location of high-intensity crime areas (HIAs) in English cities using census data (for more detail see § 2.6.3 this chapter). GIS has also been used to


11

analyse the ‘journey to crime’ patterns based upon police recorded crime data (Costello and Wiles, 2001). It can be used to examine the location of offences and the location of victims and to help analyse the relationships between these locations. The book ‘GIS and Crime Mapping’ by Chainey and Ratcliffe (2005) provides GIS application use in three broad areas (operational, tactical and strategic) including case studies at relevant areas.

2.3 Crime Attributes There are a large number of studies that suggest some basic demographic and socioeconomic attributes and neighbourhood/area characteristics are related to crime. A strong association has been found between the level of property crime and some key economic and demographic factors (Shaw and McKay, 1942; Schmid, 1960; Herbert, 1982; Kongmuang, 1995). The next sections will describe the key attributes, including demographic, socioeconomic, and area characteristics which are associated with crime.

2.3.1

Demographic Characteristics

Crime is not evenly distributed throughout the population. The associations between demographic characteristics and crime have long been observed. For as long as observation of offending has been made, it has been noted that men and women have different offence rates and patterns, and experiences of victimisation (Heidensohn, 2002). In particular, the demographic variable that appears to most effect crime is the number of male young adults. It has been found in many studies that males and young adults are more prone to commit crimes than females and older adults (Kongmuang, 1995; Rephann and Öhman, 1999; Croall, 1998). Schmid (1960) found high proportions of males to be the main indicator of crime. Nelson et al. (1996) studied the spatial distribution of shoplifting using crime data reported to the police for Cardiff City Centre in 1993. Offender characteristics indicated that shoplifting is predominantly carried out by young males. In a study of crime in Thailand, Kongmuang (1995) also found that most offenders are males under 25 years old. The single and divorced have higher propensities to commit crime than the married. It is interesting to note that single parent families, which might be seen as broken homes or weak families, and large families (a large number of children in the family) are significant features of delinquent groups (Ellis, 1988) and there is also above average offender rates among minority groups (Herbert, 1976). Offenders from ethnic minorities are also over-represented in the prison population (Croall, 1998). For young people, significant risk factors that may increase the likelihood of offending behaviour are low income, poverty, poor parental supervision, poor housing, poor school behaviour (attainment and attendance), and socially-disorganised communities (Raikes, 2002).


2.3.2

12

Socio-economic Characteristics

There is now widespread acceptance that there is a strong relationship between criminality and socio-economic factors. Socio-economic status has been suggested as a key indicator by several studies and unemployment appears to have special significance (Schmid, 1960; Becker, 1968; Herbert, 1983; Darden, 2001; Edmark, 2005). A large number of studies have shown a consistent and inverse relationship between crime, delinquency, and social class or socio-economic status (Herbert, 1983). Field (1998) pointed out that economic factors have a major influence on trends in both property and personal crime. He found that personal crime (sexual offences and violence against the person) show a distinctive relation to personal consumption, which could be explicable in terms of a relationship between the economic status of individuals and their propensity to become involved in crime. People who are unemployed, poorly educated, and have a low income will have higher risks of committing crime (Becker, 1968; Darden, 2001, Kongmuang, 1995).

2.3.2.1 Employment There has been a long debate as to whether there is a relationship between crime and unemployment. Schmid (1960) found high rates of unemployment to be the main indicator of crime. Empirical studies trying to link unemployment and property crime have been undertaken in many different ways. Two recent and interesting examples are Hale (1997) and Witt et al. (1999). Hale found that there is a long-run relationship between trends in recorded burglary and theft and the structure of employment. By using aggregate data from 42 police force areas for the period 1986 to 1996 Witt et al. (1999) found a significant relationship between high property crime and both increases in male unemployment and high wage inequality (see, for more details, § 2.6.1 this chapter). Reilly and Witt (1992) examined the relationship between crime and unemployment in Scotland using regional data. They concluded that unemployment cannot be dismissed as one of the determinants of the crime rate. A study in Sweden that investigated the effects of unemployment on crime using county panel data for the period 1988-1999 also found that unemployment had a positive and significant effect on property crime especially burglary and car theft (Edmark, 2005). There are a lot of studies indicating that unemployment has a great influence on crime. Chamlin and Cochran (2000) studied how best to measure unemployment within the context of the relationship between unemployment and property crime. They suggested that the number of individuals unemployed for 15 weeks or more significantly affect the level of property crime. Long term or permanent unemployment, rather than temporary unemployment, is reflected in the production of higher levels of property crime.


13

2.3.2.2 Tenure Type In Britain, Herbert (1979) and Baldwin and Bottoms (1976) found that spatial variations in offender rates are associated with variations in household tenure. Areas dominated by owner-occupied dwellings show lower rates than would be predicted by their socioeconomic composition alone. The owner-occupied dwellings show the lowest crime rates. Public-sector housing display relatively high levels of criminal activity (Cater and Jones, 1989). Housing and criminality are related because social groups with a greater propensity to commit crime are concentrated in certain types of housing. Some areas have a large numbers of offenders residing within them because the type of housing in such areas is more available to those individuals at greater risk of offending. Examples include areas of privately rented accommodation with over-representations of young single males, and transient populations (Bottoms and Wiles, 1988). Individuals who rent a home may have a higher propensity to commit crime. Baldwin and Bottoms (1976) strongly suggested that the housing market might be relevant to the spatial distribution of offender rates over other variables, for example, social class distribution of households in the area. Wikström (1991) also found similar results, based on the path-model analysis for offender rates in different areas of Stockholm. 2.3.2.3 Poverty and Deprivation Poverty and crime are believed to be highly correlated (Kongmuang, 1995; Hirschfield and Bowers, 1997; Darden, 2001). However, Laycock (2003) argued that crime is not the result of poverty per se but of growing disparities in affluence. Indeed, from the end of the First World War, crime has risen steadily, in line with growing prosperity.

“Individuals, families and groups can be said to be in poverty if they lack the resources to obtain the types of diet, participate in the activities and have the living conditions and amenities which are customary, or at least widely encouraged or approved in the societies to which they belong”

(Townsend, 1979: 31)

Poverty, which is defined as low income or material resources, can be considered as narrower in scope than deprivation. Deprivation is a difficult concept to define. It involves anything from poverty to inequality. Deprivation overlaps, but is not the synonymous with, poverty (Carstairs and Morris, 1991). However, deprivation is narrower than both social exclusion and inequality (Senior, 2002). When people face difficult situations, they may


14

have greater propensities to commit crime. It has been found in many studies that deprivation has a high correlation with crime. Ratcliffe and McCullagh (2001) explored the relationship between repeat victimisation and deprivation using GIS techniques. They found that burglary increases linearly with increasing household deprivation. There have been a number of deprivation indices used for crime analysis. For example, the Townsend Index of Deprivation made up of the following four census variables:

Unemployment: unemployed people aged over 16 years as a percentage of all economically active people aged over 16.

No car: households with no car as a percentage of all households.

Overcrowded accommodation: households with more than one person per room as a percentage of all households.

Not owner occupier: households not owning their own home as a percentage of all households.

The Townsend Index has been used to study crime pattern analysis in many studies (see for example, Craglia et al., 2000; and Craglia et al., 2001). The Index of Multiple Deprivation (IMD) developed by the Index Team at Oxford University for the Department of Transport, Local Government and the Regions is another index which has been accepted as useful for studying crime. The IMD 2000 was constructed by combining the six domain scores, using the following weights:

Income (25%)

Employment (25%)

Health Deprivation and Disability (15%)

Education, Skills and Training (15%)

Housing (10%)

Geographical Access to Services (10%)

However, it can be argued that the relationship between socio-economic attributes and crime cannot explain every type of crime, especially ‘white-collar crimes’. The ‘white-collar crime’ was first defined by Sutherland in 1941 as crime committed by persons of respectability and high social status in the course of their occupations (Croall, 1998). Currently, the definition of ‘white-collar crime’ is still unclear. However, it is generally associated with wealthy and powerful offenders. The most common ‘white-collar crimes’ include different kinds of fraud, for example credit card fraud, bankruptcy fraud, insurance fraud, and financial fraud. ‘White-collar’ offenders are motivated by two important factors:


15

economic difficulty and greed. ‘White-collar crime’ can be accepted as a ‘crime career’. Individuals and organisations commit these types of crime to obtain money; to avoid payments; or to secure personal or business advantage. The socio-economic status of ‘whitecollar’ offenders is different from common offenders because ‘white-collar’ offenders have higher socio-economic status.

2.3.3

Neighbourhood/Area Characteristics and Offence Locations

Crime also varies by the kind of area in which people live. It seems to be commonly accepted that urbanisation is an important factor when considering crime against property in modern western societies. Urbanisation is highly related to the extent of propensities to commit crime. In many cases relationships between propensity factors and crime rates have been found to be stronger with higher levels of urbanisation (Dahlbäck, 1998). Such analysis has a long history. Guerry (1833) used a series of annual reports to analyse French crime patterns. He found that crimes against property were higher in urban areas. If society moves from a rural to urban character, and particularly as large cities begin to emerge, official crime rates will rise (Scott, 1972). Urban areas not only have much higher crime rates than rural areas but crime rates are also positively correlated with city size and hierarchical rank (Harries, 1974). Several parts of the world also show a link between urbanisation and crime. In Thailand for example, it has been found that violent crime and property crime are clustered in commercial and residential areas in the city centre rather than the surrounding suburbs. Crime has also been related to population density in that areas with high population density tend to have high crime rates (Kongmuang, 1995). In general, as detailed elsewhere (Bottoms and Wiles, 2002) there are considerable differences in both offending and victimisation between neighbourhoods. These differences have some relationship to social class composition, degrees of deprivation, types of housing, and features of the physical environment (Smith, 2002). Schmid

(1960)

concluded

that

urban crime areas (areas where offenders live and where crimes are committed) are normally characterised by all or most of the following factors: low social cohesion, weak family life, low socio-economic status, physical deterioration, high rates of population mobility and personal disorganisation. According to research findings by Shaw and McKay, the inner city stands out as a high crime area and area of offender residence even when demographic and socio-economic characteristics are taken account of. This can imply that crime is to a significant degree the product of a neighbourhood and not only the residents’ characteristics. The same people tend to behave differently in different places/locations and environments (Cater and Jones, 1989).


16

Baldwin and Bottoms (1976) found that lower-class individuals were more likely to commit crimes if they lived in lower-class neighbourhoods than if they lived in higher-class or mixed neighbourhoods. The neighbourhood type may either expand or reduce the propensities of individuals to commit crime. Demographic and socio-economic characteristics of neighbourhoods could affect life-cycle events and may affect an individuals’ propensity to experience crime. However, Wikström (1991) argued that area of residence and offender rates might be related because of the distribution of more or less crime-prone individuals or groups. The social life of the area itself might not affect the criminality levels of the residents. Community change can also produce crime. The study of juvenile offender rates in different areas of Los Angeles for the period 1950-1970 by Schuerman and Kobrin (1986) found that there is a three-stage process to making particular areas develop high offender rates. First, there were changes in land use such as an increase in rented accommodation. Secondly, there were changes in demographic features such as an increase in residential mobility. Third, there were changes in socio-economic characteristics such as more unskilled people and more unemployed people. These changes were cumulative and in the order specified for the areas examined, ultimately resulting in a shift from low to high offender rate areas. Urban land use changes, together with new patterns of mobility and new lifestyles, are inducing changes in offence patterns and the emergence of new geographical concentrations of offences (Ceccato et al., 2002). Wiles and Costello (2000) carried out an area-based analysis of residential neighbourhoods by using police recorded data. Residential neighbourhoods were categorised as high, medium, or low offence rate areas. They found that residential areas with a high offender rate but low offence rate did not exist, and few areas had a low offender rate but high offence rate. On a neighbourhood basis, crime tends to be highest in areas with low neighbourhood stability, high poverty, and a high minority ethnic population (Bottoms and Wiles, 2002). Bottoms and his colleagues argued that in Britain social segregation emerged as the intended and unintended consequence of policy decisions taken by local government departments responsible for housing. Housing allocation was an indirect effect of moral judgements about tenants that resulted in the concentration of criminal populations (Bottoms and Wiles, 1986; Bottoms et al. 1989). The lack of social cohesion and high disorganisation within a neighbourhood are possible variables explaining where high offence rates occur and where offenders come from (Shaw and McKay, 1942; Hirschfield and Bowers, 1997). Explanations of where offenders come from often focus on spatial attributes and emphasise housing type and neighbourhood


17

socialisation processes. Wikström and Loeber (2000) identified neighbourhood socioeconomic context as having a direct impact on offences for certain groups of young offenders. In terms of the proximity to opportunity, Waller and Okihiro (1978) examined burglary in Toronto, Canada and found that proximity of affluent housing to public sector housing was the highest correlate of high burglary rates. The study of relations between crime and disadvantage on Merseyside (the City of Liverpool and the four surrounding metropolitan districts) in north-west England by Hirschfield et al. (1995) suggested that the impact of bordering effects, particularly the proximity of poor or disadvantaged areas to affluent neighbourhoods, is that it is likely to have high offence rates (Hirschfield et al., 1995; Bowers and Hirschfield, 1999).

2.4 Movement of Offenders: Journey to Crime Offenders travel certain distances from their residence to the offence location. The trip from the point of origin to the actual crime scene is known as the ‘journey to crime’. ‘Movement of offenders’ or ‘journey to crime’ has long been studied. Most studies of the ‘journey to crime’ are empirical. The most general and consistent conclusion is that offenders do not travel very far from where they live. The majority of offender movements are relative short. For example, the Sheffield Crime Survey found most of the city’s burglaries to have occurred within two miles of the offender’s residence (Baldwin and Bottoms, 1976). In North Staffordshire, England, Evans (1989) found that almost half of the burglaries are committed within 0.8 km of the convicted burglar’s home. Most offender movements associated with crime appear to be based on opportunities. Offending appears to be concentrated around offenders’ homes, areas of work and recreation, and the pathways in between (Brantingham and Brantingham, 1981). Wikström (1991) tried to model where offences occur by looking at the forms of social interaction taking place within the urban area and their variation. He found that offences take place where criminal opportunities intersect with areas that are known to the offender because of their routine use of that space. A study of ‘travel to crime’ patterns in Sheffield by Wiles and Costello (2000) found that 1) the vast majority of offender movements are relative short: on average offenders travelled only 1.93 miles to commit a crime from their homes 2) much travel is not primarily driven by plans but opportunities and 3) although the travel distance can be different depending on the offence type, an offender’s travel to crime is local in nature.


18

2.5 Crime Victimisation The risk of victimisation is unevenly distributed across populations, time, and space. Who is at greatest risk from crime? How is victimisation distributed across space and what are the demographic dimensions? Studies have consistently shown that the likelihood of victimisation varies dramatically with demographic, socio-economic and area characteristics (Budd, 1999; Zedner, 2002). Individual and community attributes significantly affect household victimisation risk (Trickett et al., 1995). A person’s lifestyle may also affect the likelihood of victimisation (Gottfredson, 1984). Laub (1997) suggested that age is one of the variables most highly correlated with victimisation. Income is also related to risk of individual victimisation because if income goes up, risk goes down; those who live in urban areas are more prone to be victims than are residents of suburban or rural areas. Laub proposed this pattern can explain all individual and household crimes. A victim-focused study by Maguire and Bennett (1982) examined residential burglary by mapping recorded burglary and carrying out in-depth interviews with victims. They found that burglaries tended to be clustered either in poorer housing areas or in more expensive properties close to main roads. Houses were more likely to be burgled if located near a road junction, or if they offered good access to potential offenders. Zedner (2002) found that the risk of burglary is much higher in inner-city areas, particularly those with high levels of physical disorder, and in rented accommodation rather than owneroccupied households. The British Crime Survey (BCS) has indicated that both AfroCaribbeans and Asians suffer more victimisation than whites. The possible reasons are that minority groups tend to have lower household incomes and have a larger proportion of young people and higher rates of unemployment. In the study ‘university student safety’ Barberet et al. (2003) found that property crime including burglary, theft and criminal damage victimisation was more prevalent than violent crime. Over 70% of all crimes experienced by students are property crimes and nearly 12% of students in private accommodation were burglary victims (compared with 5% of students who live in university accommodation). In general terms, the British Crime Surveys reveal that urban areas have higher rates than rural, and inner city areas have higher rates than suburbs. However not all inner city areas have higher rates of crime. It can be argued that crime rates are related to a variety of factors including the kind of area in which people live which are not always reflected in the distinction between inner city and suburban areas. Generally, geographical differences are related to social differences, which are illustrated in the way that victimisation rates are


19

spread between different types of areas (Mayhew et al., 1993). There are ‘place’ variations with differences between types of areas, and ‘people’ variations with specific types of people at greater risk. For example, households with lower income, single-adult, young, or unemployed heads of households, are at greater risk (Kershaw et al. 2000; Budd 1999). In the 2001 British Crime Survey, it has been found that young households, single-parent households and those in areas of high physical disorder were particularly at risk for domestic burglary (Kershaw et al., 2001). Most research on the relationship between crime and opportunities has been strongly influenced by the ‘routine activity theory’ proposed by Cohen and Felson (1979). According to the routine activities theory, crimes can occur when three elements come together: a motivated offender, a suitable target such as unprotected property, and the absence of people who could prevent the offence from being committed. According to this theory, changes in routine activities can strongly affect crime opportunities and crime rates, even if individuals’ propensities to commit crime do not change. More activity away from home and families increases the opportunity for crime and thus generates higher crime rates (Cohen and Felson, 1979; Felson, 1994). For example, teenagers or young adults typically go to study/work 5 days a week and leave their rooms/houses empty. This can put these houses at greater risk of burglary say compared with the houses of elderly persons, who may spend considerable time at home. A good deal of previous research shows that offenders are very likely to select targets not far from their own residence (see § 2.4). For this reason, living close to motivated offenders produces a much higher victimisation risk and living in a high crime area increases individual’s risks of victimisation (Cornish and Clarke, 1986). Miethe and Meier (1990) found that people who lived in areas with higher levels of offenders had higher risks of burglary. In the UK, inner city residents or council housing occupiers have higher burglary risks and rates (Ellingworth et al., 1997; Osborn and Tseloni, 1998). In some cases the risk of crime differs according to demographic characteristics which are in turn related to lifestyle. One of the interesting factors has been discovered by the British Crime Surveys is that individuals who tend to have an increased risk of victimisation spend several evenings a week out, drink heavily, and may be more likely in turn, to assault others (Budd, 2003).


20

2.6 Modelling Crime There have been a number of attempts to develop crime models, especially in regard to property crime. For example, in 1998 the Home Office developed two models of property crime based on long-run aggregate relationships between recorded crime and macroeconomic and demographic factors. These have been updated and revised to incorporate new techniques and, for the first time, to investigate the implications of projecting trends in crime on the basis of the ‘macro-level’ relationships captured by econometric models (Dhiri et al., 1999). Earlier Home Office research by Field (1998) attempted to model historical trends in the level of recorded crime in England and Wales over the last half century. The research focused on a subset of recorded property crime– burglary and theft and handling, which accounts for around two-thirds of total recorded crime incidents. In terms of crime predictors, it is now well established that household and area characteristics play important roles. The British Crime Survey has been used to develop statistical models describing property crime victimisation at the household level (Tseloni et al., 2002). The statistical modelling studies of British Crime Survey data at the individual household level typically use the technique of logistic regression. Recent years have seen the development of models that try to explain property crime victimisation at the individual household level using data from the British Crime Survey. The resulting models tried to predict the victimisation risk of a particular household. The predictors generally include household characteristics and the characteristics of the area of residence (Trickett et al., 1995). Regression has been used in many studies to develop crime models. Examples of multivariate regression, poisson and negative binomial regression, and logistic regression models are shown in the following sections.

2.6.1

Multivariate Regression Model

Dahlbäck (1998) applied a non-linear longitudinal model when analysing the influence of crime opportunity and propensity factors on societies’ theft rates in Sweden. The variables used in his study were: -

Average population density

-

Business volume in restaurants

-

Relative size of agricultural population


-

Proportion of males 15-24 years

-

Proportion of divorced residents

-

Proportion of children in public leisure centres

-

Proportion of foreigners

-

Migration

21

He found that the theft rate is higher the denser the population and the weaker the social bonds. Family disintegration, cultural estrangement, and change in place of residence are three important factors that affect the strength of social bonds. Witt et al. (1999) attempted to ascertain the relationship between some important economic factors and crime. As mentioned in § 2.2, they used data from 42 police force areas in England and Wales to examine the impact of unemployment on crime. The New Earnings Survey (NES) (an annual sample survey of the earnings of employees in Great Britain) is also used to measure the extent to which rises in earning inequality can explain changes in crime. The linear dynamic specification below forms the basis for subsequent estimation.

Δyijt = β1Δyijt −1 + β 2 Δx'ijt +γ t + ϕ + Δυijt

Where yijt

(2.1)

is the number of crimes per capita in police force area i for crime category j in year t

xijt

is the vector of exogenous variables

γt

is a vector of year dummies

φ

is a vector of regional dummies

υijt

is the error term

Δ

is a first difference operator

As discussed in § 2.3, it has been found that high crime is associated with: increases in male unemployment, high growth in the number of properties which can be burgled, and high wage inequality associated with the distributions of weekly earnings of full-time manual men. The increase in the size of police force is negatively correlated with property crime.


22

Hope et al. (2001) tried to identify antecedents and correlates of multiple crime-type victimisations by fitting an appropriate multivariate statistical model to data taken from the 1992 British Crime Survey. “Multiple crime-type victimization is the extent to which some households or persons are victims of more than one kind of offence over a given period”

(Hope et al., 2001: 595)

Victim/target proneness Property crime victimization

Property crime victimization

Personal crime victimization

Personal crime victimization Event-dependency

Figure 2.1: A model of repeated, multiple crime victimisation (of Hope et al., 2001)

Figure 2.1 outlines their theoretical model. The model tried to explain the variation in two discrete random variables: being a present property crime victim and being a present personal crime victim. Thus, they used the bivariate probit model which is defined by: Y1* = β1x1 + ε1,

Y1 = 1 if Y1* > 0,

0 otherwise

(2.2)

Y2* = β2x2 + ε2,

Y2 = 1 if Y2* > 0,

0 otherwise

(2.3)

Where E(ε1) = E (ε2) = 0,

Var(ε1) = Var (ε2) 1 and Cov(ε1, ε2) = ρ

Y1 and Y2 denote the individual observed binary outcomes for property and personal crime victimisation during the reference period. The coefficients along with the correlation can be estimated using the maximum likelihood technique. If ρ = 0, any correlation between property and personal crime victimisation is well explained by the influences of the variables x1 and x2. These influences can be ascertained by estimating each of the equations separately.


23

If ρ ≠ 0, there is a correlation between two types of crime victimisation which is not captured by the explanatory variables (prior victimisation, selected characteristics of respondents, their households, and area of the household).

They found a highly significant positive association between the binary victimisation variables. The model shows young adults living in households where the head of household is a young adult, living in a household that have a number of children, renting tenure, and living in a below average car ownership area have higher proneness to be a victim. Both affluence and disadvantage are associated with the probability of being the victim of property crime. The risks of personal crime are associated with a number of variables such as being single/divorced, young adults, living with children, and rented dwelling.

Gaviria and Pagés (2002) focus mainly on how the relative socio-economic status of individuals, the population size of the city of residence, and the population growth affects the probability of being a victim. They found that the probability increases with socioeconomic status, city size, and urban growth. They also found that the victims of property crime in Latin America typically come from rich and middleclass households and tend to live in the bigger and faster growing cities. The ‘Latinobarometer’, a public opinion survey covering more than 50,000 households in 17 Latin American countries, was used as the main source of data. The survey includes information about demographic characteristics as well as crime victimisation at the household level. Major shortcomings of the ‘Latinobarometer’ are the absence of information about victimisation type and data on household income. In reality, Gaviria and Pagés used only two sets of data: ownership of durable goods and housing characteristics related to the socio-economic status of the households. They ranked households according to their socio-economic status. Their procedure has three main steps: 1) Use principal components to compute a weighted average of the relevant household attributes. 2) Rank all households on the basis of the average. 3) Use the corresponding ranking to compute quintiles of socio-economic status. They used the following equation to study the patterns of crime victimisation in Latin America


Yijct = c + X ijct β + Z jcθ + λc + ζ t + ε ijct Where Yijct Xijct Zjc λc ζt εijct

24

(2.4)

is a dummy variable showing whether a member of family i who lives in city j of country c was a victim of crime in year t is a vector of household characteristics (including education of the household head, relative socio-economic status, and house ownership) is a vector of the city characteristics (including population size and population growth). is a country effect. is a year effect is an individual error term.

The study by Martin (2002) used multivariate regression and spatial analysis to test the significance of several concepts as predictors of neighbourhood burglary rates. Three hundred and twenty census tracts were used as proxies for neighbourhoods in Detroit. He used official reported residential burglaries during 1995-1997 obtained from the Detroit Police Department. Because of a highly skewed distribution, the Freeman-Tukey transformation in the SpaceStat software was used to transform the burglary rate. Formally, the Freeman-Tukey transformation can be stated as

Zi = Ei / Pi + ( Ei + 1) / Pi Where Ei Pi

(2.5)

is the number of events in unit i is the population at risk

The Freeman-Tukey Transformation was computed in his study as

Burglary rate =

Where x n

1000 x n + 1000( x + 1) n

(2.6)

is the count of burglaries in the census tract is the number of housing units in the tract.

He modelled the burglary dwelling rate as a linear function of neighbourhood characteristics including poverty concentration, age composition, social capital, and residential stability. Age composition has the strongest effect on burglary dwelling rate. Poverty concentration and residential stability are also significant effects. Social capital is negatively related to burglary dwelling.


2.6.2

25

Poisson and Negative Binomial Regression Model

Poisson regression is often used to model the number/rate of occurrences of an event of interest such as crime, as a function of some independent variables. However, if evidence of underdispersion or overdispersion is shown, it indicates an inadequate fit of the poisson model. In 1998 Osborn and Tseloni studied the distribution of household property crimes (Osborn and Tseloni, 1998). They adopted the negative binomial generalisation of the poisson model because they found that a standard poisson model does not capture the distribution of victimisation (negative binomial is the test for overdispersion in poisson regression). Osborn and Tseloni examined how socio-economic and demographic features of the household and their neighbourhoods affect the probability distribution of the number of property crimes, including theft, burglary, and criminal damage. Because they argued that no previous study had used individual and area characteristics to model the entire crime probability distribution, socio-demographic attributes of the household and community-level characteristics were used in their study to predict victimisations, with the victimisation data from the 1992 British Crime Survey. The 1991 Census Small Area Statistics were used for variables at the household level, including the number of adults in the household, ethnicity and the age of the head of the household together with the socio-economic status of the head of the household, and also the number of cars in the household. The Poisson Model If the number of events (crimes) for the ith case (i = 1, 2,….N) is expressed by the random variable Yi, then the poisson model assumes that the mean number of events (crimes), λi = E(Yi), is related to a vector of interpretive variables xi , through

ln(λi ) = β T xi

(2.7)

where ln indicates the natural logarithm. The mean number of events is the expected property crime incidence for household i. The probability that Yi takes the specific value yi (yi = 0,1,….) is

Pr(Yi = yi ) =

exp(−λi )λi yi !

yi

(2.8)


26

The Negative Binomial Model The assumption of the poisson regression model of (2.7) and (2.8) are successive events occur independently and at a constant rate. However, in practice, overdispersion can be found. They model overdispersion with

ln(λi ) = β T xi + ε i

(2.9)

where exp(εi) follows a gamma distribution with mean 1 and variance α. Combining equation (2.9) with (2.8) results in one version of the negative binomial model for the number of events with

Γ( yi + v) v v μi i Pr(Yi = yi ) = yi !Γ(v) (v + μi )v + y i y

yi = 0,1,....

(2.10)

where v = 1/α is the precision parameter and Γ is the gamma function. As in the poisson case, the expected property crime incidence is E(Yi) = µI = exp(βT хi). The negative binomial specification has variance Var (Yi ) = μi + αμ I2

(2.11)

Osborn and Tseloni used maximum-likelihood estimation in the software package LIMDEP. They found that characteristics of the household affect the victimisation rate. The negative binomial regression allows the estimation of the probability distribution of crimes. From probability distributions, incidence can then be obtained as the estimated crime rate, whilst risk can be found as estimated probability of at least one victimisation.

2.6.3

Logistic Regression Model

Craglia et al. (2001) used a logistic model to predict police-defined, high intensity crime areas (HIAs) in English cities. HIAs are areas that produce special policing problems due to particularly violent crime where residents are either unwilling or afraid to cooperate with the police. To develop the model, they used three police force areas, Greater Manchester, Merseyside and Northumbria.


27

The response variable was a binary one, the value depending on whether each Enumeration District (ED) was in a police defined HIA or not. The initial model specification showed the link with many census variables that captured different aspects of socio-economic disadvantage, population instability and ethnic mix. These attributes have been found to be associated with levels of violent crime. At first, they started with all the possible predictor variables and then eliminated non-significant variables. The variable INDEX, the Department of the Environment Transport and the Regions (DETR) index of deprivation, was also included. The probability that an ED is in an HIA, according to the final model, is

given by exp( Χβ ) 1 + exp( Χβ ) Where exp( Χβ )

(2.12)

is the exponential function of

Χβ = -2.3465 + 0.2351(INDEX) + 0.0102(TERRACE) + 0.0229(TURNOVER) INDEX

(2.13)

is the Department of the Environment Transport and the Regions (DETR) index of local deprivation.

TERRACE

is proportion of people living in terraced housing

TURNOVER

is residents with a different address one year before the census

The model performed well in general because there are a few ‘false positives’ (EDs predicted by the model to be in HIAs but had not been claimed by the police). However, there is some evidence of positive spatial autocorrelation. Therefore the model was re-fitted with additional new variables that were the spatially averaged values (WTERRACE, WTURNOVER, WINDEX). The results of this model fitting are shown in Table 2.1. The model concludes that HIAs are characterised by populations who are deprived and live in high density areas with high levels of population turnover. Table 2.1: Best fitting final model for HIA data after including spatially averaged variables Variable INDEX TERRACE TURNOVER WINDEX WTERRACE Constant

Coefficient 0.1801 0.0091 0.0201 0.1447 0.0056 -2.9470

Percentage of HIA EDs classified Percentage of non-HIA ED classified

Source: Craglia et al. (2001) Note: Numbers in parentheses are the numbers of EDs.

Correctly (percentage) 53.68% (634) 52.9% (707) 81.49% (1554) 79.9% (1523)

Incorrectly (percentage) 46.32% (548) 47.1% (557) 18.51% (353) 20.1% (384)


28

Entorf and Spengler (2000) explored the socio-economic and demographic factors of crime in Germany based on the Becker-Ehrlich deterrence model. From the basic model of BeckerEhrlich:

Lnο = α + β ln D + γ ln Y + δ ln X Where O

(2.14)

is the crime rate

D

is deterrence

Y

is income

X

is other influences

They applied other influence variables such as unemployment, increasing income inequality, number of foreigners, urbanisation. They ran a static regression using the specification written as follow:

ln O = α + β ln p + γ 1 ln Y a + γ 2Y r + X 'δ Where O

(2.15)

is the crime rate (number of crime per 100,000 populations)

p

is the clear-up rate a

Y

is absolute income

X

is unemployment, age, foreigners, urbanisation, and east-west differentials

The second step was the implementation error correction model taken from dynamic timeseries analysis.

Δ ln O = c + g (ln O−1 − γ 1 ln Y−a1 − γ 2Y−r1 − X '−1 δ ) + βΔ ln P + γ 1Δ ln Y a + ΔX 'δ

(2.16)

Where ∆ is the difference operator and g should have a negative sign. The following variables were used: FOREIGN

=

percentage of foreigners in the population

Ya

=

GDP per capita in constant prices

M15-24

=

percentage of male aged 15-24

=

relative distance between states’ GDP and federal GDP

UNEMPL

=

unemployment rate

UNEMPL24

=

share of unemployed persons under 25 years of age out

Y

r

of all unemployed persons EAST

=

indicator variable for East Germany

CITY

=

indicator variable for the city-states


29

The results of this study support the notion that demographic and economic factors play important and significant influences. Being young and unemployed increases the propensity to commit crime.

2.7 Concluding Comments There has been almost two centuries of studies into the geography of crime. This chapter describes the geographical approaches to crime together with crime attributes and its related determinants. The geography of crime always carries its particulars in spatial structures, in environmental associations, and in the special qualities of place (Herbert, 1989). Crime is a very unevenly distributed phenomenon and cannot be understood outside its social context. The association of demographic and socio-economic characteristics with the locations of crimes can provide a clearer picture of crime. To understand crime patterns it is useful to examine the determinants of crime, such as demographic and socio-economic characteristics, by using GIS together with statistical analysis. Crime models are generally based on statistical regression approaches. In particular, for the estimation of small area crime rates, regression coefficients are typically estimated from survey data for large geographical areas and then applied to small areas using small area census data (Tanton et al., 2001). Moreover, as mentioned in § 1.1 these techniques are unable to model policy impacts. In the UK, the smallest area of modelling crime is generally at the police force area level. As can be seen from the literature reviewed in this chapter, there is considerable evidence that demographic, socio-economic, household and neighbourhood characteristics are related to crime, especially the likelihood of victimisation. Table 2.2 shows crime attributes relating to offenders, victims, and offence areas derived from the literature reviewed. In Chapter 5 these attributes will be explored in relationship to crime to find out: whether or not they are good predictors of Leeds crime, and how these can be possibly linked in a spatial microsimulation context.


Table 2.2: Crime attributes Categories

Indicators

High propensity

Demographic characteristics of offender

Age Sex Marital status Family status Family size

Ö Young adult Ö Male Ö Single Ö Weak family life (divorce, single parent families) Ö Large

Socio-economic characteristics of offender

Income Employment Education Deprivation

Ö Low income Ö Unemployed Ö Less Ö High level of deprivation

Density of living Type of tenure

Ö High density (substandard) Ö Rented

Age Sex Ethnicity Lifestyle Tenure

Ö Young adult Ö Male Ö Minority group Ö Away from home Ö Rented, not owner-occupied

Urbanisation Population density No of offender Social cohesion Ethnicity Proximity

Ö High Ö High Ö High Ö Lack of social cohesion and disorganisation Ö High minority ethnic population Ö Inner city, proximity to disadvantage areas

Household characteristics Victim characteristics

Neighbourhood characteristics

30

Chapter 3- Microsimulation Modelling: A Literature Review

31

Chapter 3 Microsimulation Modelling: A Literature Review 3.1 Microsimulation: An Introduction 3.2 Types of Microsimulation 3.2.1 Static and Dynamic 3.2.2 Spatial and Aspatial 3.3 A Review of Selected Microsimulation Models 3.4 Advantages and Disadvantages

3.5 The Creation of Synthetic Microdata 3.5.1 Synthetic Reconstruction 3.5.2 Combinatorial Optimisation 3.6 Combinatorial Optimisation using Simulated Annealing Method 3.7 Concluding Comments

3.1 Microsimulation: An Introduction The microsimulation technique was first used to study socio-economic systems in the United States by Guy Orcutt, in the late 1950s and early 1960s (Orcutt et al., 1961). It is a methodology aimed at building large-scale datasets at the micro scale for individual units such as persons, households or firms (Clarke, 1996; Ballas and Clarke, 2000) and simulating the effect of changes in policy or other changes on these microunits. In essence it is the distribution of a sample of individuals to meet a set of aggregate real statistics or probabilities. Simulation entails conducting a baseline simulation using a given initial population sample and later changing either the sample characteristics or parameters within the model in order to measure the effects of policy or structural changes (Rephann and Öhman, 1999). Microsimulation models have been increasingly adopted to study the impacts of social and economic policies especially for income and tax policy on individual units (Merz, 1991; Clarke, 1996; Ballas et al., 2005a, c). In the past two decades, microsimulation models have become very powerful tools and have been used widely in North America, European countries and Australia.

Section 3.2 describes the types of microsimulation. Selected microsimulation models are reviewed in § 3.3. The advantages and disadvantages are summarised in § 3.4. Synthetic reconstruction and combinatorial optimisation are compared in § 3.5 while combinatorial optimisation using simulated annealing method is described in detail in § 3.6. Section 3.7 is concluding comments.

3.2 Types of Microsimulation Microsimulation models are generally divided into two main types: static and dynamic. However, in this study, there are two key characteristics to be considered: static and dynamic, spatial and aspatial.


3.2.1

32

Static and Dynamic

Microsimulation models can be static or dynamic. The distinction between static and dynamic models is an important one, because they have evolved along different lines, and have very different areas of application (Martini and Trivellato, 1997). The difference between static and dynamic depends on the particular method that is used (Mitton et al., 2000). Most crucial is the method for ‘ageing’ the microunits (Merz, 1991).

Early microsimulation models were often static and generally used when only crosssectional information is required to answer a policy question. The microdata database of static models is typically comprised of cross-section information at a certain point of time. Such models are mostly designed for answering questions about short-term effects or the immediate distribution impact of policy changes, for example, the effects on income distribution if welfare systems are changed. Generally, these models produce output that shows the gains or losses from such policy changes. Static microsimulation models (especially tax-benefit models), have been developed for the majority of industrialised countries. Merz (1991) summarises the requirements for a static microsimulation model (Table 3.1) as the provision of an appropriate microdata database with implements for merging data if they come from several sources; a construction that computes the characteristics of microunits; the simulation part of the model (which is the process of imitating the behaviour of system patterns); the adjustment of microdata (before or after the simulations), and the evaluation of the executed simulation. In addition, each of these should be considered for efficiency and ease of use.

Dynamic microsimulation models are more complicated and involve updating each attribute for each microunit for each of a set of time intervals. Dynamic microsimulation models often start exactly the same as static microsimulation models. The difference is that the dynamic microsimulation models project a sample of the population forward through time by simulating the major life events that individuals would be expected to experience in the real world. These could be birth, death, marriage, divorce, education, labour force participation, health, retirement etc. The probabilities of these life events occurring are estimated using data concerning the rates at which each particular event happens, either in the populations or for a given individual. Therefore, within a dynamic microsimulation model, the characteristics of each microunit are recalculated for each time period. This allows the original population to be projected forward in time, while maintaining detailed information on the individuals within the simulation.


33

Dynamic microsimulation models usually include cross-sectional analysis, with static microsimulation being one dimension of these models. This view might be justified on the basis of the attempts to extend existing static models to dynamic models by including behavioural functions. An example of combining a dynamic model with an existing static system is the dynamic spatial microsimulation model developed by Ballas et al. (2005a).

Table 3.1: Requirement profile for a static microsimulation model

Requirement profile for a static microsimulation model 1.

2.

3.

4.

5.

6. 7.

Initial data- Preparation and construction 1.1 Microdata processing 1.2 Macrodata processing 1.3 Modifications of initial data 1.4 Statistical methods for matching (merging microdata) 1.5 Construction of the initial file 1.6 Extraction of subfiles 1.7 Documentation: Initial database, subfile(s) Module construction 2.1 Construction of micro modules 2.2 Construction of macro modules 2.3 Econometric and statistical methods for hypotheses testing and formulation 2.4 Documentation: Module construction Modifications of model parameters- Model operations 3.1 Scenario formulation 3.2 Parameter changes 3.3 Module changes 3.4 Handling of module sequence 3.5 Linkage micro to macro or other models 3.6 Testing 3.7 Documentation: Model operation Adjustment of microdata 4.1 Demographic adjustment ‘static aging’ 4.2 Economic aging 4.3 Stochastic changes and ‘alignment’ 4.4 Sensitivity analyse and changing aggregate control data 4.5 Statistical adjustment methods 4.6 Documentation: Adjustment Evaluation of simulation 5.1 Results of single simulation 5.2 Results of several simulation runs 5.3 Statistical methods for data analyses 5.4 Documentation: total and particular evaluation Efficiency in processing Ease of use

Source: Merz (1991)


34

Static and dynamic microsimulation models each have their own strengths. Dynamic models are viewed as better in terms of producing realistic long-range estimations. The main advantage of static versus dynamic microsimulation models is that static models are less computationally expensive (Merz, 1991; Fredrikson, 1998) and they also provide the possibility of more detailed representation (Fredrikson, 1998). It is less expensive because time-consuming simulations of demographic processes (with interactions among members of different microunit associations) are not included (Merz, 1991).

3.2.2

Spatial and Aspatial

Until recently, most static and dynamic microsimulation models were aspatial (Clarke, 1996), concerned with ‘who is affected’ not ‘where these people live’. Most of the microsimulation models developed so far do not take spatial scale into account (Birkin et al., 1996). As Clarke (1996) points out, very few microsimulation studies to date have had a spatial dimension. The most important limitation of such models is that results are only available at the national level (Brown and Harding, 2002). It is plainly not possible using aspatial microsimulation models to predict the spatial impact of policy changes upon the household sector.

Spatial microsimulation is a recent development, starting mainly in the late 1970s and 1980s (much at the University of Leeds) (Clarke 1996; Caldwell, Clarke, and Keister 1998; Ballas and Clarke 2000). For example, Birkin and Clarke (1988) used static spatial microsimulation techniques to create a synthetic microdata database for Leeds (UK) and used this to generate incomes at the individual level. Williamson developed a model for the spatial analysis of community care policies for older people (Williamson, 1992) and a microsimulation model for water demand estimation for small areas (Williamson et al., 1996; Williamson, 2001). Ballas (2001) developed SimLeeds, which is a spatial microsimulation model for Leeds’ local labour market.

The key advantage of spatial microsimulation models is that they contain geographical information that can be used to investigate the local area impacts of policy changes. Spatial microsimulation is useful for modelling the socio-economic and spatial effects of policy changes at different geographical scales (Ballas and Clarke, 2001a, b). It also helps policy makers to think more geographically about the possible effects of policy options they may consider (Ballas et al, 2005c). The availability of geo-coded microdata and new techniques for merging geographical, population, and socio-economic data make it likely that spatial models will become more prevalent (Clarke, 1996).


35

As Ballas et al. (2005a, c) point out; spatial microsimulation normally involves four major procedures:

The construction of a small area microdata dataset from samples and surveys.

Sampling from this dataset to generate a micro population for individuals for small areas who match the known data on those areas.

Static what-if simulations, in which the impacts of alternative scenarios on the population are estimated.

Dynamic modelling (updating a base microdata dataset) and future-oriented what-if simulations.

It has been noted that spatial dynamic microsimulation is an extremely difficult task involving the behavioural modelling of individuals over time and at various geographical scales (Ballas et al., 2005b).

3.3 A Review of Selected Microsimulation Models A large number of dynamic and static microsimulation models have been developed since the 1960s covering the following topics: tax policy analysis, income tax, social welfare, urban housing market, distributional impact of energy policies, national health insurance, state unemployment insurance, land-use forecasting, energy demand, health benefit, pension analyses, health insurance, labour supply shortening of working hours, distributional impacts of child allowance changes, effects of transfer wage and social policies, economic and social policy market and non-market activities, the shadow economy, and effect of tax regulations on agricultural firms (Merz, 1991). A number of the major models will now be briefly reviewed.

DYNASIM (Dynamic Simulation of Income Model) is a dynamic model which simulates the economic and social behaviour of households in the United States. It was among the first micro models to adopt the dynamic microanalytic simulation approach (Zaidi and Rake, 2001). It was developed by Guy Orcutt in the early 1970s. The model consisted of behavioural relationships for birth, death, marriage/remarriage, divorce, leaving home, disability, education, location, wage rate, labour force participation, hours in the labour force,

unemployment,

earnings,

social

security,

other

pensions,

unemployment

compensations and welfare programmes (Orcutt et al., 1986). Its uses included forecasts of the population to 2030 employing different assumptions about demographic and economic scenarios, and analysis of the cost of teenage childbearing to the public sector under different policy scenarios (O’Donoghue, 2001a). DYNASIM 2, a second version was


36

developed between 1979-1983. The time horizon extends from 1973 to 2030. DYNASIM is organised in three sub-models: the family and earning history model, the jobs and benefits history model, and the cross-sectional imputation model (Spielauer, 2002). Some of the ideas and experiences of DYNASIM are brought to CORSIM, a dynamic population model, developed at Cornell University in 1987. CORSIM is accepted as the first PC-based simulation model of the United States population. It was much influenced by DYNASIM (Zaidi and Rake, 2001). It was designed to model individuals and their families, basic demographic characteristics of birth, death, marriage and divorce, emigration and immigration, as well as levels of education, economic patterns of work and earnings, the accumulation of assets and debts, and contributions to pensions. It can be run through the recent past and carried on into the future making the program a robust tool for both basic social science research and policy analysis. The 1960 Public Use Microdata Survey is the base microdata for CORSIM. This database is a one-per-thousand representative sample drawn from the 1960 US Census which consists of about 180,000 persons (70,000 families) (Zaidi and Rake, 2001). The core CORSIM modules were also widely adapted by other models for example, DYNACAN and SVERIGE. In 1995 CORSIM was chosen to be used as a template for building a dynamic microsimulation model for Canada. It replaced US data, equations, regulations, etc. with Canadian counterparts. This model, called DYNACAN, was developed by the Office of the Chief Actuary of the Canadian Pension Plan. Therefore the model aims at projecting and evaluating the financial impacts on individuals and families of alternative policy options for the Canadian Pension Plan (Spielauer, 2002). The base population is the one per cent (213,000 person) public-use sample of the population from the 1971 Canadian Census then aged annually through to 2100. ESPASIM is a static microsimulation model of taxes and benefits for Spain. It is used to study the effect shortly before or immediately after a reform, before the agents adjust their behaviour as a consequence of the policy change. It is also static in that it takes only one period of time into consideration, assuming the demographic and socio-economic structures remain constant. ESPASIM uses microdata from representative samples of the Spanish population. It is set up to work with two different databases: the Household Budget Survey 1990-91, and the third wave of the Spanish sample of the European Community Household Panel. These two databases provide information on the income, housing and other demographic and socio-economic characteristics of the individuals and households (Levy et al., 2001).


37

The National Centre for Social and Economic Modelling (NATSEM) was established at the University of Canberra in 1993 and STINMOD (Static Incomes Model) is NATSEM’s first static microsimulation model of the Australian tax and transfer systems. The first version of STINMOD was launched in 1994. It has been used to analyse income distribution, poverty, and inequality (Lambert et al., 1994). It was developed in Statistical Analysis System (SAS). The base population of STINMOD is generated from the Income Distribution Survey (IDS) by a reweighting method. NATSEM also have a dynamic microsimulation model, called DYNAMOD, designed to project population characteristics over a period of up to 50 years. The model operates with a 1 per cent sample of the Australian population (Zaidi and Rake, 2001; King et al., 1999) which is about 160,000 individuals (Brown and Harding, 2002). Major elements of the DYNAMOD include demographics, international migration, education, the labour market and earnings (Zaidi and Rake, 2001; Spielauer, 2002). PENSIM is a dynamic population microsimulation model developed by Hancock, Mallender and Pudney in1992. It was built to project incomes of future pensioners, in order to enlighten policies on income security in old age in the UK (Hancock et al., 1992). It was used to project certain characteristics of the distribution of pensioner’s incomes up to 40 years in the future. The initial database for PENSIM came from three different sources: The 1988 Survey of Retirement and Retirement Plans, the 1988 Family Expenditure Survey, and the Social Change and Economic Life Initiative (O’Donoghue, 2001a). MOSART is a dynamic microsimulation model for Norway developed by Statistics Norway to examine policy options with regard to financing public expenditure (Andreassen et al., 1994; Fredriksen, 1998). The base dataset of the model consists of 12 per cent of the Norwegian population in 1993. This initial population is derived from a combination of several registers from Statistics Norway and the National Insurance Administration (Zaidi and Rake, 2001). The first version of MOSART was developed between 1988-1990. It focused on demographic behaviour, education and labour force participation in order to study the impact of demographic change on the labour force and educational achievement. The second version was used for pension modelling. Currently MOSART is in the third version. It includes more detailed behavioural modules concerning household formation and disability (Zaidi and Rake, 2001). LIFEMOD is a dynamic cohort microsimulation model developed by the London School of Economics. LIFEMOD models the life histories of a cohort of 4,000 persons, comprised of 2,000 males and 2,000 females. It was built to model the lifetime impact of a welfare state and to estimate the degree to which income is redistributed among people over time or across life-cycles (Falkingham and Hills, 1995a, b).


38

The SAGE model is a dynamic microsimulation model developed by the Simulating Social Policy in an Ageing Society research group at London School of Economics. It aims to assess the impact of different social policy options on the future demand for pensions, health and personal social services, and long term care (Evandrou et al., 2001). The base population contains data for each individual on gender, date of birth, current marital status, current labour market status, educational attainment, and current health and disability status (Zaidi and Scott, 2001).

In 1997 the Spatial Modelling Centre in Kiruna, Sweden, developed a spatial dynamic microsimulation model, called SVERIGE or System for Visualising Economic and Regional Influences in Governing the Environment. It is the first national-level spatial microsimulation model (Rephann et al., 2005). It has been noted that the initial structure of the SVERIGE was a replication of CORSIM (Caldwell et al., 1998), which itself is a modification of Guy Orcutt’s DYNASIM, the first dynamic microsimulation model (Caldwell and Keister, 1996). CORSIM was used as a template in the same manner as DYNACAN. It was replaced by Sweden data, module by module, with estimation equations based on the ‘Total Population of Sweden Individual and Geographical’ database (TOPSWING) (Holm et al., 2002). The main difference which can be seen as the greatest advantage of the TOPSWING database (over any other microsimulation model) is that it contains individual, longitudinal information for each person living in Sweden, on demography, family, work, income, employment, location etc. The locations of the individuals are given in coordinates accurate to the scale of 100 metres. Therefore, there is enormous potential to explore the spatial aspects of any policy. SVERIGE is based on households. SVERIGE is different from CORSIM as it is a spatial model.

SimLeeds is a spatial microsimulation model that has been used to explore the potential spatial impact of a factory closure in Leeds, and to estimate the geographical impact of changing national social policies (Ballas, 2001; Ballas and Clarke, 2001a, b). It was developed by Ballas in 1999 in the University of Leeds. SimLeeds used a simulated annealing-based reweighting method to create spatially disaggregated population microdata at the Enumeration District (ED) level. SimLeeds variables include:

Location (place of residence) at the ED level Location (workplace) at the ward level Age Sex Marital status Tenure Employment status Industry (SIC)


39

Socio-economic group Earned income Job seekers allowance (JSA)

MicroMaPPAS (Micro-simulation Modelling and Predictive Policy Analysis System) is a Spatial Decision Support System (SDSS) developed for Leeds City Council. It can be seen as the first attempt to link spatial microsimulation modelling frameworks to Spatial Decision Support Systems. The system is based on a spatial microsimulation model which links data from different sources including the 2001 Census data for output areas and sample data from the 10th wave of the British Household Panel Survey (BHPS) (Ballas et al., 2004). The simulated annealing method has been adopted to create a population microdata dataset involving reweighting the microdata sample from the BHPS so that it fits small area statistics for Leeds from the census. In fact the MicroMaPPAS model builds on SimLeeds. Some simulated results from the MicroMaPPAS can be found in Stillwell et al., (2004).

In recent years NATSEM has developed a spatial microsimulation model called SYNAGI (Synthetic Australian Geo-demographic Information) which seems to be influenced by the works of Williamson (especially in the method to create the synthetic microdata). The SYNAGI uses a reweighting method to combine the census and Australian Bureau of Statistics sample survey data together to create a synthetic unit record file for every Statistical Local Area (SLA) in Australia (Lloyd and Harding, 2004). Reweighting is undertaken using an optimisation method to iteratively generate a set of weights that best fits each Census Collection District (CCD), which is the smallest geographical area in the Australia Census (Melhuish et al., 2002). The variables used in SYNAGI include:

Total household income Age Marital status Country of birth Labour force status by sex Occupation Family type Student status High income segments by age Housing type Housing tenure Household size Number of motor vehicles Mortgage repayments Rent payments


40

SimBritain is a spatial dynamic microsimulation that uses the 1991 Census Small Area Statistics (SAS) and the British Household Panel Survey (BHPS) to dynamically simulate the entire population of Britain up to the year 2021 at the small area level. SimYork was used as the pilot project to test different methodologies and combinations of datasets. SimYork is the model for the city of York, whereas SimBritain is the model for Britain. The main difference between these two models is that the SimYork was implemented at the ward level, whereas SimBritain was implemented at the parliamentary constituency level. SimBritain is used for what-if policy scenarios and to assess the impact of different social policy options on the future demand for pensions, health, and personal social service, and long-term care (Ballas et al., 2005a).

SMILE (Simulation Model for the Irish Local Economy) is a dynamic spatial microsimulation model designed to examine the impact of policy change and economic development on rural areas in Ireland. Same as SimLeeds and SimBritain, SMILE has the advantage of providing spatially disaggregated microdata that can be aggregated to any spatial scale. SMILE is model of population simulating the basic components of population change such as mortality, fertility and internal migration at the small area level (Ballas et al., 2005b).

3.4 Advantage and Disadvantages There are many advantages of microsimulation (see Ballas and Clarke, 2000; Clarke, 1996). The first advantage is the ability to link data from different sources. This enables the provision of estimates of new population cross-classifications unavailable from published sources to be created. The second advantage is spatial flexibility. The microunit-base is not only a characteristic of such models but also the main advantage because they produce results which can be analysed at the individual level. This makes it possible to assess the impact of policy changes across different geographical scales through aggregation (Lambert et al., 1994). The third advantage is efficiency of storage because in the microsimulation framework variables are specified as lists rather than as matrices. The fourth advantage is the ability to update and forecast.

However, there are some drawbacks. First, is the difficulty of validating the model outputs, because microsimulation models estimate distributions of variables which were previously unknown (Ballas and Clarke, 2000; Clarke, 1996). However, as Ballas (2001) pointed out, the validation of microsimulation models can be done by re-aggregating the estimated datasets to levels at which observed datasets exist (by comparing the estimated distributions


41

with the observed). Another drawback is the large requirements of computational power. Traditionally, computer storage and computational speed were main barriers to microsimulation but nowadays these issues are getting less problematic because of the dramatic technological advances in computer hardware. Developments in computing allow microsimulation models to analyse more sophisticated problems.

3.5 The Creation of Synthetic Microdata Although many countries, for example Sweden, have a microdata database, because of confidentiality problems, in the UK we do not have a microdata database on individuals and households. Thus, it is useful to create synthetic microdata. Synthetic reconstruction and combinatorial optimisation are the two main approaches used to create small area population microdata which comprise lists of individuals along with an associated set of individual characteristics. (Williamson et al., 1998; Williamson, 2002).

3.5.1

Synthetic Reconstruction

Synthetic reconstruction, a well-established technique, has been used in many studies when suitable microdata have not been available (see for example, Birkin and Clarke, 1988; Williamson, 1992). This approach requires the construction of a set of synthetic individuals or households whose characteristics match aggregate characteristics for the small area. It normally involves a method such as Iterative Proportional Fitting (IPF) using contingency tables or conditional probability analysis to estimate chain conditional probabilities. The method proceeds in a sequential manner. Conditional probabilities, calculated from available known data, are used to reconstruct detailed micro-level populations by repeating Monte Carlo Sampling from a chain of conditional probabilities. For example, from the census data we can get the number of household heads by age, sex, and marital status in each small area. Given employment probabilities, the next step of the IPF procedure involves the estimation of the probabilities of economic activities given age, sex, and marital status of household head (Figure 3.1). Such a procedure is carried out for all the variables we wish to include in our synthetic microdata. The variables such as age, sex, marital status, tenure, and socioeconomic activity can be estimated using census data. However some variables are not available from the census. Using IPF procedure, data from different sources may be linked together. For more details on using IPF to estimate conditional probabilities see Birkin and Clarke (1988). The main advantage of the synthetic reconstruction approach is that the use of conditional probabilities allows data to be integrated from the widest possible range of sources (Huang and Williamson, 2001)


42

Household head (hh) Steps

1. Age, sex, and marital status (M) of household head (From SAS Table 39)

2. Probability of employment status of household head, given age, sex, and marital status (From SAS Table 34)

1st

Age: 19 Sex: Male M: Married

2nd

Age: 25 Sex: Male M: Married

Last

Age: 65 Sex: Female M: SWD

0.6

0.6

0.0

3. Random number (Computer generated)

0.43

0.38

0.27

4. Employment status assigned on the basis of random sampling

Employed

Employed

Unemployed

5. Next household head (repeat until all household heads assigned an employment status)

Figure 3.1: Microsimulation procedure for the allocation of employment status (after Clarke, 1996)

3.5.2

Combinatorial Optimisation

An alternative approach to generate synthetic microdata dataset is the combinatorial optimisation approach (Figure 3.2). The process involves selecting the combination of household records from available microdata which offers the best fit for known constraints in the selected small area. Williamson et al. (1998) describe this process in more detail and explore various techniques of combinatorial optimisation including the hill climbing approach, the generic algorithm approach, and the simulated annealing approach (used here: see § 3.6 for details). They found that modified simulated annealing stands out as the best solution. They estimated small area populations by combining information contained in the Sample of Anonymised Records (SAR) and the census Small Area Statistics (SAS) tables from the 1991 Census. The process starts from an initial set of households chosen randomly


43

from the SAR. These are randomly allocated into SAS areas until the number of households matches the number reported by the SAS tables. The other SAS aggregate statistics are then generated (for example the gender distribution). One household is then randomly replaced with a new household from the SAR, and the aggregate statistics reassessed. If the replacement improves the fit, the households are swapped. Otherwise, the swap is made or not made on the basis of the simulated annealing algorithm (see § 3.6). The process is repeated with the aim of gradually improving the fit between the observed data and the selected combination of SAR households. Given computational time limits, the final combination is the best achievable rather than the guaranteed optimal solution (Huang and Williamson, 2001). Synthetic reconstruction and combinatorial optimisation methodologies for the creation of small area synthetic microdata have been examined by Huang and Williamson (2001). They found that outputs from both methods can produce synthetic microdata that fit constraining tables very well. However, the dispersion of the synthetic data has shown that the variability of datasets generated by combinatorial optimisation is much less than by synthetic reconstruction, at ED and ward levels. The main problem for the synthetic reconstruction is that a Monte Carlo solution is subject to sampling error which is likely to be more significant where the sample sizes are small. Ordering is also important in the generation of new characteristics (Clarke, 1996). The ordering of conditional probabilities can also be a problem as synthetic reconstruction is a sequential procedure. The degree of error will increase when we go further along the chain in the generation of characteristics. Another drawback of synthetic reconstruction is that it is more complex and time consuming to program. The outputs of separate combinatorial optimisation runs are much less variable and much more reliable. Moreover, combinatorial optimisation allows much greater flexibility in selecting small area constraints. They conclude that combinatorial optimisation is much better than synthetic reconstruction when used to generate a single set of synthetic microdata. Table 3.2 summarises the difference between synthetic reconstruction and combinatorial optimisation from the work of Huang and Williamson (2001)


44

Step 1: Obtain sample survey microdata and small area constraints Survey Microdata Household

Known small area constraints [Published small area census tabulations]

Characteristics size adults children

(a) (b) (c) (d) (e)

2 2 4 1 3

2 1 2 1 2

0 1 2 0 1

1. Household size (persons per household)

2. Age of occupants

Household Size 1 2 3 4 5+

Type of person adult child

Total

Step 2:

Frequency 1 0 0 1 0

Frequency 3 2

2

Randomly select two household from survey sample [ (a) & (e) ] to act as an initial small area microdata estimate

Step 3: Tabulate selected households and calculate (absolute) difference from known small area constraints Household Estimated Observed Size frequency frequency (i) (ii) 1 2 3 4 5+

0 1 1 0 0

1 0 0 1 0 Sub-total:

Step 4:

Absolute difference │(i) – (ii)│

Age

Estimated frequency (i)

Observed frequency (ii)

4 1

3 2

1 1

Sub-total:

2

1 1 1 1 0

adults

4

Total absolute difference


=4+2=6

Randomly select one of selected households (a or e). Replace with another household selected at random from the survey sample, provided this leads to a reduced total absolute difference. **(The simulated annealing algorithm introduce additional at this stage)**

Households selected: (d) & (e) [Household (a) replaced] Tabulate selection and calculate (absolute) difference from known constraints Household Estimated Observed size frequency frequency (i) (ii) 1 2 3 4 5+

1 0 1 0 0

1 0 0 1 0 Sub-total


Age




3 2

0 1

Sub-total:

1

0 0 1 1 0

adults

3 1

2


=2+1=3

Step 5: Repeat step 4 until no further reduction in total absolute difference is possible: Result: Final selected households: (c) & (d) Household Estimated size frequency (i) 1 2 3 4 5+

1 0 0 1 0



1 0 0 1 0

0 0 0 0 0

Sub-total

0

Age adults




3 2

3 2

0 0

Sub-total:

0


Figure 3.2: A simplified combinatorial optimisation process Source: Williamson (2002), page 237

=0+0=0


45

Table 3.2: Synthetic reconstruction versus combinatorial optimisation (Summarise from Huang and Williamson, 2001)

Synthetic Reconstruction

Combinatorial Optimisation

@ Step by step process The value of each household or individual’s characteristics is estimated by random sampling from a probability conditional upon previously generated attributes. @ Ordering matters Because of the step by step process, each value is created in a fixed order. @ More complex and time consuming

@ Iterative process With the aim of gradually improving the fit between actual data and the selected sample of microdata datasets, the process is therefore repeated many times. @ Flexibility of selecting the constraining tables We can select small area constraints to match our own requirements.

3.6 Combinatorial Method

Optimisation

using

Simulated

Annealing

As mentioned in the previous section, simulated annealing is one of the combinatorial optimisation methods that has been used successfully to generate a microdata dataset (Ballas, 2001; Williamson et al., 1998). It has been noted that the simulated annealing procedure can generate real people living in real households (in the sense that individuals are modelled and not synthetically reconstructed, not statistical entities) which is a key advantage over the IPF-based methods (Ballas, 2001). The term ‘simulated annealing’ derives from the physical process of heating and then slowly cooling a substance to obtain a strong crystalline structure (the annealing process) until no further changes occurs. The simulated annealing algorithm is based upon that of Metropolis et al. (1953), which was originally proposed as a means of finding the equilibrium configuration of a collection of atoms at a given temperature. Because it can be formulated as the problem of finding a solution among a potentially large number of solutions, Kirkpatrick et al. (1983) suggested that it forms the basis of an optimisation technique for combinatorial problems. Figure 3.3 shows a standard simulated annealing algorithm. It consists of a sequence of iterations. Each iteration consists of randomly changing the current solution to generate a new solution in the universe of possibilities. Once a new solution is generated a goodnessof-fit statistic is generated and the change is compared with previous combinations to decide whether the newly produced solution can be accepted as the current solution. If the change is negative (lower than the previous one) the newly produced solution is accepted


46

unconditionally and the system is updated. If not then it is accepted dependent upon Metropolis’s criterion (Metropolis et al., 1953) which is based on Boltzman’s probability (Pham and Karaboga, 2000). The option of whether or not to accept a ‘worse’ combination instead of a ‘better’ one is essentially determined by the laws of thermodynamics (Williamson et al., 1998). Each iteration has a simulated ‘temperature’, and ‘energy’ determining the likelihood of a worse solution being chosen. At a given temperature T, the probability of an increase in energy p(δE) is given by

p (δE ) = exp( −

δE kT

)

(3.1)

Where k is a constant, called Boltzmann’s constant.

Briefly, some changes are accepted even if they lead to a reduction in performance. This means the simulated annealing algorithm has an ability to avoid becoming trapped at local minima in the universe of solutions, a major advantage over many other methods. When the value of the current solution has not changed or improved within the last iteration, the search is terminated and the current solution kept.


47

Initial Solution

Evaluate the Solution

Accepted?

Update the Current Solution

Generate a New Solution

Change Temperature?

Decrease Temperature

Terminate the Search?

Final Solution

Figure 3.3: Flowchart of simulated annealing algorithm (after Pham and Karaboga, 2000)


48

3.7 Concluding Comments This chapter has introduced microsimulation models and described the difference between static and dynamic, spatial and aspatial models. Further, the characteristics and uses of selected microsimulation models have been reviewed. Advantages and disadvantages are also summarised. Moreover, the methods to create a population microdata dataset were reviewed. Finally, the simulated annealing procedure which has been used successfully to create a microdata dataset was described.

It should be noted that dynamic models need much greater resources to build and maintain than static models, having much greater data and modelling requirements. They are, therefore, currently more suitable for long term scientific research than immediate policy reactions. Spatial microsimulation can be seen as a technique that creates synthetic microdata for small areas. Spatial microsimulation combines the advantages of aspatial micro-analytical approaches with those of geographical models that take space into account. Adding geo-references into the microdata makes it more valuable and the spatial aspect is capable of providing geographical detail for different scales. The benefits of creating synthetic microdata are the construction of spatially disaggregated data from aggregate data such as surveys and the ability to create tables of census variables that do not exist in published sources.

The simulation of tax-benefits and social security policies is one of the fundamental applications both of static and dynamic microsimulation. Although there are a large number of microsimulation models, they have currently not been used for crime analysis. As mentioned earlier, the spatial microsimulation approach has been used successfully to examine changes in policies, and there is no reason why this should not apply to crime policies. It can be argued that spatial microsimulation is ideally suited to analyse crime, given the considerable associations between crime and individual demographic and socioeconomic characteristics, as well as place. In terms of policy analysis potential using what-if analysis, it will be possible to demonstrate how policy change may make a difference to crime rates.

Chapter 4- Modelling Crime: Data Sources and Issues

49

Chapter 4 Modelling Crime: Data Sources and Issues 4.1 Introduction 4.2 Data Sources and Issues 4.2.1 The 2001 Census 4.2.2 The 2001/2002 British Crime Survey 4.2.3 Police Recorded Crime Datasets 4.2.4 Offender Dataset 4.3 Concluding Comments

4.1 Introduction This study is mainly based on the UK 2001 Census and the 2001/2002 British Crime Survey (BCS). To model crime the microsimulation model creates an individual level microdata database of Leeds, estimated using multiple census datasets. This includes all the attributes from the British Crime Survey. The model provides small area estimates of being a victim of crime and thus pinpoints high risk areas. To validate this procedure results need to be compared with real figures from police recorded crime. Moreover, with detailed data on known offenders, spatial interactions between the locations of offenders and crime locations can also be simulated. Four main sources of data will thus be used in this study: the 2001 Census, the 2001/2002 BCS, police recorded crime datasets, and police known offender dataset. The next section provides general information together with the limitations of these datasets. Finally, § 4.3 provides some concluding comments.

4.2 Data Sources and Issues 4.2.1

The 2001 Census

The census is a survey of the whole UK population. It has been carried out every ten years since 1801. The latest census was held on 29th April, 2001. The data in the census describes the characteristics of the population of the UK including demography, households, families, housing, ethnicity, birthplace, migration, illness, economic status, occupation, industry, workplace, transport mode to work, cars, and language (Rees et al., 2002). The questions listed in Table 4.1 allow the generation of results in a cross-tabulation format which is available for academic use. It provides a comprehensive spatial coverage. However, the output is modified when small numbers are involved and raw microdata itself is not released because of respondent confidentially (Rees et al., 2002). Data are thus released for small


50

areas only, e.g. output areas (OAs) or wards, and are not available at the individual or household level. The aggregate outputs are counts of people or households broken down by demographic and socio-economic characteristics. These are contained in a series of tables on a specific topic or area of interest. The 2001 Census aggregate statistics datasets include:

Key Statistics: The Key Statistics datasets provide an overview and summary of the main topics which are the most important and generally used statistics in a series of straightforward tables. It is available for all 2001 Census geographies.

Standard Tables: The Standard Tables datasets provide the most detailed information in a large number of cross-tabulated tables. It is available down to ward level in England, Wales and Northern Ireland, and postcode sector level in Scotland. It is not available for output areas.

Standard Table Theme Tables: The Standard Tables Theme Tables are designed to contain information about ranges of subjects related to particular themes available down to ward level in England, Wales and Northern Ireland, and postcode sector level in Scotland. It is not available for output areas.

Census Area Statistics: Census Area Statistics provide the most detailed results possible for smaller areas. They are generally produced for the same areas as the Key Statistics.

Census Area Statistics Theme Tables: The Census Area Statistics dataset includes a subset of Theme Tables, designed to contain information about a range of subjects related to particular themes. They are available for the full range of 2001 Census geographies down to output areas.

Census Area Univariate Tables: Available for the full range of 2001 Census geographies down to output areas describing a single variable only.

Armed Forces Tables: Provide information on members of the Armed Forces available down to Local Authority District level for England and Wales only.

All of these datasets are available via Census Area Statistics Website (CASWEB)


51

Table 4.1: Topics in the 2001 Census _____________________________________________________________________ No Topics _____________________________________________________________________ For all properties occupied by households and all unoccupied Household accommodation: 1 The address, including postcode 2 Type of accommodation 3 Names of all residents Names and usual addresses of visitors on census night (optional) 4 Tenure of accommodation 5 Whether rented accommodation is furnished or unfurnished (in Scotland only) 6 Type of landlord (for households in rented accommodation)a 7 Number of room 8 Availability of bath and toilet 9 Self-containment of accommodation 10 Lowest floor level of accommodation 11 Number of floor levels in the accommodation (in Northern Ireland only) 12 Availability of central heating 13 Number of cars and vans owned and available For residents: 14 Name, sex, and, date of birth 15 Marital status 16 Relationship to others in household 17 Student status 18 Whether or not students live at enumerated address during term-time 19 Usual address one year ago 20 Country of birth 21 Knowledge of Gaelic (Scotland only), Welsh (Wales only), and Irish (Northern Ireland only)a 22 Ethnic groupa 23 Religion 24 Religion of upbringing (Scotland and Northern Ireland) 25 General health 26 Long-term illness 27 Provision of unpaid personal care 28 Educational and vocational qualifications 29 Economic activity in the week before the census 30 Time since last employment 31 Employment status 32 Supervisor status 33 Job title and description of occupation 34 Professional qualifications (England) 35 Size of workforce of employing organization at place of work 36 Nature of employer’s business at place of work (industry) 37 Hours usually worked weekly in main job 38 Name of employer 39 Address of place of worka 40 Means of travel to worka 41 Address of place of study (in Scotland only) 42 Means of travel to place of study (in Scotland only)

_________________________________________________________________ Source: Denham and Rees (2002), pp 311-312 Note: Bold indicates a new question (compared with the 1991 Census); Italic indicates a question to be used in only one part of the United Kingdom. a Response categories vary among parts of the United Kingdom.


52

The main dataset used in this study is the Census Area Statistics (CAS), which is equivalent to the Small Area Statistics (SAS) of the 1971, 1981, and 1991 Censuses. It is available for geographical levels down to output area (OA), the smallest unit of the 2001 Census geography. Each output area contains approximately 290 persons or 125 households. This is different from the 1991 Census when the smallest areas were Enumeration Districts (EDs) and electoral wards with an average size of about 180 and 2,000 households respectively (Dale and Teague, 2002). As mentioned above, the Census Area Statistics provide the most detailed results possible for smaller areas. In terms of data volume, it is the largest of the 2001 Census datasets, containing approximately 2 billion individual items of data. Table 4.2 shows the Census Area Statistics dataset tables available via Census Area Statistics Website (CASWEB), the academic web interface to census aggregate outputs and digital boundary data. Census Area Statistics dataset tables vary in size. The number of cells in a table range from 21 to 540 depending upon the number of variables involved and the number of categories. Larger tables provide more detailed information. However, the larger the tables are the greater the possible effect of data blurring as the likelihood of private data disclosure is greater as detail increases. “The 2001 CAS will differ from the 1991 SAS in a significant respect. To avoid even the perception of disclosure, counts in tables will not only be subject to imputation and record swapping, but will also be randomly perturbed and rounded to the nearest three”

(Denham and Rees, 2002: 305)

Such data blurring applied to the released census can lead to discrepancies in census counts between tables. The impact of data blurring may mean that there is no possible combination of households that would match every constraining table perfectly (Huang and Williamson, 2001), as each may have different totals. Figure 4.1 shows the number of people aged 16-74 by output area from different tables. As can be seen, there is a different number in each different table. This problem will be discussed in more detail in § 6.2.2.


Table 4.2: Census Area Statistics dataset tables available from CASWEB

Census Area Statistics dataset tables CS001

Age by Sex and Resident Type: All People

CS002

Age by Sex and Marital Status: All People

CS003

Age of Household Reference Person (HRP) by Sex and Marital Status (Headship): All Households

CS004

Age by Sex and Living Arrangements: All People in Households

CS011

Family Composition by Age of Family Reference Person (FRP): All Families

CS012

Schoolchildren and Students in Full-Time Education Living Away from Home in Term-Time by Age: All Schoolchildren and Students in Full-Time Education who Would Reside in the Area were they not Living Away from Home During Term-Time

CS013

Age of Household Reference Person (HRP) and Tenure by Economic Activity of HRP: All Households with HRP Aged 16 to 74

CS016

Sex and Age by General Health and Limiting Long-Term Illness (LLTI): All People in Households

CS017

Tenure and Age by General Health and Limiting Long-Term Illness (LLTI): All People in Households

CS019

General Health, Limiting Long-Term Illness (LLTI) and Occupancy Rating by Age: All People in Households

CS020

Limiting Long-Term Illness (LLTI) and Age by Accommodation Type and Lowest Floor Level of Accommodation: All People in Households

CS021

Economic Activity by Sex and Limiting Long-Term Illness (LLTI): All People Aged 16 to 74

CS023

Age and General Health by NS-SeC: All People Aged 16 to 74

CS025

Sex and Age by General Health and Provision of Unpaid Care: All People in Households

CS026

Sex and Economic Activity by General Health and Provision of Unpaid Care: All People Aged 16 to 74 in Households

CS027

Households with a Person with a Limiting Long-Term Illness (LLTI) and their Age by Number of Carers in Household and Economic Activity: All Households

CS029

Sex and Age by Hours Worked: All People Aged 16 to 74 in Employment the Week Before the Census

CS030

Sex and Economic Activity by Living Arrangements: All People Aged 16 to 74 in Households

CS032

Sex, Age and Level of Qualifications by Economic Activity: All People Aged 16 to 74

CS033

Sex and Occupation by Age: All People Aged 16 to 74 in Employment the Week Before the Census

CS034

Former Occupation by Age: All People Aged 16 to 64 not in Employment the Week Before the Census

CS035

Sex and Occupation by Employment Status and Hours Worked: All People Aged 16 to 74 in Employment the Week Before the Census

CS038

Sex and Industry by Employment Status and Hours Worked: All People Aged 16 to 74 in Employment the Week Before the Census

CS039

Occupation by Industry: All People Aged 16 to 74 in Employment the Week Before the Census

CS040

Sex and Occupation by Hours Worked:

53


CS041 CS042 CS043 CS044 CS045 CS046 CS047 CS048 CS049 CS050 CS051 CS052 CS053 CS055 CS056 CS059 CS060 CS061 CS066 CS067 CS068 CS103 CS105 CS113 CS114 CS118

CS119 CS122

54

All People Aged 16 to 74 in Employment the Week Before the Census Economic Activity and Time Since Last Worked by Age: All People Aged 16 to 74 NS-SeC by Age: All People Aged 16 to 74 Sex and NS-SeC by Economic Activity: All People Aged 16 to 74 NS-SeC of Household Reference Person (HRP) by Household Composition: All HRPs Aged 16 to 74 NS-SeC of Household Reference Person (HRP) by Age (of HRP): All HRPs Aged 16 to 74 NS-SeC of Household Reference Person (HRP) by Tenure: All HRPs Aged 16 to 74 NS-Sec by Tenure: All People in Households Aged 16 to 74 Dwelling Type and Accommodation Type by Household Space Type: All Household Spaces. All Dwellings Dwelling Type and Accommodation Type by Tenure (Households and Dwellings): All Occupied Household Spaces. All Occupied Dwellings Dwelling Type and Accommodation Type by Tenure (People): All People in Households Tenure and Household Size by Number of Rooms: All Households Tenure and Persons Per Room by Accommodation Type: All Households Household Composition by Tenure and Occupancy Rating: All Households Dwelling Type, Accommodation Type and Central Heating by Tenure: All Households Tenure and Amenities by Household Composition: All Households Accommodation Type and Car or Van Availability by Number of People Aged 17 Or Over in the Household: All Households Tenure and Car or Van Availability by Number of People Aged 17 Or Over in the Household: All Households Tenure and Car or Van Availability by Economic Activity: All People Aged 16 to 74 in Households Sex and Approximated Social Grade by Age: All People Aged 16 and Over in Households Age of Household Reference Person (HRP) and Dependent Children by Approximated Social Grade: All Households Age and Dependent Children by Household Type (Household Reference Persons): All HRPs Sex and Age by Religion: All People Age by Highest Level of Qualification: All People Aged 16 to 74 Occupation by Highest Level of Qualification: All People Aged 16 to 74 NS-SeC by Highest Level of Qualification: All People Aged 16 to 74 Number of Employed People and Method of Travel to Work by Number of Cars or Vans in Households: All Households with at Least One Person Working in the Week Before the Census Sex and Age by Method of Travel to Work: All People Aged 16 to 74 Working in the Week Before the Census NS-SeC by Method of Travel to Work: All People Aged 16 to 74 Working in the Week Before the Census

Source: 2001 Census Area Statistics Note: Release of Census Area Statistics tables 15, 62, 64, 65 and 133 has been postponed pending the resolution of concerns over data quality with, and the re-supply of corrected data from the Office for National Statistics. Tables 5, 7, 14, 18, 22, 24, 28, 31, 36, 37, 54, 57, 63, 64, 65, 126 and 133 have been withdrawn due to a concern regarding the corruption of data supplied by ONS. However, these tables will be re-issued.


55

Figure 4.1: Discrepancies in census counts between tables Source: 2001 Census Area Statistics Note: Each cell shows the number of people aged 16-74 living in households

4.2.2

The 2001/2002 British Crime Survey

The British Crime Survey (BCS) produced by the Home Office is one of the largest social research surveys conducted in England and Wales. The BCS was first carried out in 1982 and further surveys were carried out in 1984, 1988, 1992, 1994, 1996, 1998, 2000 and 2001 respectively. The surveys have been carried out on a continuous basis since April 2001 and results from that point have been reported by financial year. The BCS is primarily a victimisation survey and is a very important source of information about levels of crime and public attitudes to crime. People do not always report crimes to the police for a variety of reasons and those crimes are therefore excluded in police recorded crime statistics. In the BCS the respondents are asked about their experiences of property crimes of the household (e.g. burglary) and personal crimes (e.g. theft from a person), and whether or not they reported these incidents to the police. Moreover, it is a rich source of detailed micro-level information. The BCS covers a wide range of topics describing the demographic and socioeconomic characteristics of respondents and household references (Table 4.3) (e.g. age, sex,


56

marital status, ethnicity, economic activity, socio-economic group, household income, car ownership, number of adults/children in households, long-term illness) and area characteristics, all of which play an important role in this study. The 2001/2002 BCS had a target sample of 40,000 households in England and Wales. The respondents were randomly selected from the Post Office’s list of addresses in England and Wales. Therefore it has a good mix of people from different ages, backgrounds and situations. The 2001/2002 BCS represented two linked populations: households in England and Wales living in private residential accommodation, and adults aged 16 and over living in such households. It has been noted that the BCS does not count all crimes that occur in England and Wales, but it does provide a consistent measure of trends in crime from one year to the next. Moreover, the BCS gives a more accurate picture of crime levels and trends compared with the police recorded crime, because it asks people about their actual experiences (thus covering crimes that do not get reported to the police). However, there are some limitations with the BCS. Firstly, the BCS only surveys people aged 16 and over in private households. Therefore it does not include crime against people aged under 16 and it does not cover the population resident in student Halls of Residence, those in residential care, those in prison, or members of the armed forces. Secondly, it does not cover certain types of crime including: victimless crime (drug offences), fraud, sexual offences and homicide because the victims cannot be interviewed (while police recorded crime does) (Table 4.4). Thirdly, while the BCS provides a picture of crime at the national level, it cannot tell what is happening in the local authority or neighbourhood as the police recorded data can. Thus, the BCS cannot identify small area hotspots or high risk areas.

Table 4.3: Selected topics in the British Crime Survey _______________________________________________________________________ Selected topics in the British Crime Survey Area Characteristics: Inner city flag Area type: Inner-city/Urban/Rural Standard Region Government Office Region ACORN type: ACORN Group ACORN category ACORN change type ACORN change group Police Force Area ONS Ward Classification : Group ONS District Level Classification : Family ONS District Level Classification : Group


Council areas (based on ACORN type) Government Office Region (Grouped) Area type: Rural/Not rural Neighbourhood type Structure of household Respondent: Sex Age Marital status ONS harmonised marital status Whether respondent living in a couple Cohabiting status HRP status/Respondent status Ethnic status Respondent Socio-Economic Classification (NS-SEC) - Operational Categories Respondent Socio-Economic Classification (NS-SEC) - Analytic Categories Respondent Socio-Economic Group (SEG) Respondent employment status Are you a full-time student at college or university Respondent on government training scheme Respondent away from job Whether respondent full-time student Respondent working full-time or part-time Respondent working as employee or self-employed Respondent managerial status Whether respondent employs people or not Highest qualification Cultural background In which way do you occupy this accommodation Who is your landlord ONS Harmonised Tenure type ONS harmonised accommodation type Number of adults in household Number of children under 16 in household Number of cars Total household income in last year Personal earnings of respondent in last year Personal earnings of partner in last year Total household income (4 bands) Total household income (5 bands) Total household income (6 bands) Is respondent victim or not ONS harmonised long-standing illness Respondent Lifestyle: No. visits pub/wine bar evening last month How often have you visited a nightclub in last month How often do you drink alcohol How many units of alcohol do you drink How many hours spent away from home during day Household occupied during day Number of hours home left unoccupied on average Is home ever left unoccupied during weekdays How long home is left unoccupied on an average weekday

57


Household Reference Person: Age of Household Reference Person Sex of Household Reference Person Marital status Cohabiting status HRP social class Ethnic Group Disability/illness Number of cars HRP Socio-Economic Classification (NS-SEC) - Operational Categories HRP Socio-Economic Classification (NS-SEC) - Analytic Categories HRP Socio-Economic Group (SEG) Household reference person employment status HRP: On a government scheme for employment training Is HRP a full-time student at college or university Whether Household Reference Person full-time student Household Reference Person on government training scheme Household Reference Person away from job Household Reference Person working full-time or part-time Household Reference Person working as employee or self-employed Household Reference Person managerial status Whether HRP employs people or not Victim Experiences: If vehicle stolen or driven away without permission How many times has this happened (MotTheft)) If something stolen off or out of vehicle How many times has this happened (MotStole) If vehicle tampered with or damaged How many times has this happened (CarDamag) Owned a bicycle at any time in reference period How many bicycles does household own If bicycle stolen How many times has this happened (BikTheft) If anyone got into previous residence to steal/try to steal How many times has this happened (PrevThef) If anyone got into previous residence and caused damage How many times has this happened (PrevDam) If anyone tried to get into previous residence to steal/cause damage How many times has this happened (PrevTry) If anything was stolen out of previous residence How many times has this happened (PrevStol) If anything was stolen from outside previous residence How many times has this happened (PrOside) If anything was damaged outside previous residence How many times has this happened (PrDeface) If anyone got into current residence to steal/try to steal (Movers) How many times has this happened (HomeThef) If anyone got into current residence to steal/try to steal (Non-movers) How many times has this happened (YrHoThef) If anyone got into current residence and caused damage How many times has this happened (YrHoDam) If anyone tried to get into current residence to steal/cause damage How many times has this happened (YrHoTry) If anything was stolen out of current residence How many times has this happened (YrHoStol) If anything was stolen from outside current residence How many times has this happened (YrOside) If anything was damaged outside current residence How many times has this happened (YrDeface)

58


59

If anything was stolen out of hands pockets or bag How many times has this happened (PersThef) If anyone tried to steal anything from hands pockets or bag How many times has this happened (TryPers) If anything has been stolen from a cloakroom office etc How many times has this happened (OtheThef) If personal items have been deliberately damaged How many times has this happened (Delibdam) If anyone has deliberately used force/violence on respondent How many times has this happened (Delibvio) If anyone has threatened to damage things/use force or violence How many times has this happened (ThreViol) If respondent has been sexually assaulted or attacked How many times has this happened (SexAttak) If member of household has used force or violence on respondent How many times has this happened (HhldViol) Have you ever been victim of crime reported to police Have you been the victim of crime in last 2 years Have you ever been arrested by police Have you been arrested by police in last 2 years Have you ever been in court during a criminal case Have you been in court in last 2 years Have you ever been a juror in criminal case Have you been a juror in a criminal case in last 2 years Have you ever been in court as the accused Have you been in court as the accused in last 2 years Have you ever been in contact with probation service Have you been in contact with probation service in last 2 years Have you ever been inside a prison Have you been inside a prison in last 2 years Have you been the victim of a vehicle crime in last 5 years How many times have you been the victim of vehicle crime Have you been insulted pestered or intimidated How many times have you been insulted or intimidated How many people insulted or intimidated you How well did you know person insulting you Any fires in last 12 months How many fires in the last 12 months Was the Fire Brigade called

____________________________________________________________________ Source: The 2001/2002 British Crime Survey, Crown Copyright. Note: A Classification of Residential Neighbourhoods (ACORN) variables are not included in the dataset for copyright/royalty reasons.


60

Table 4.4: Comparing the British Crime Survey and police recorded crime The British Crime Survey

Starting in 1982, it measures both reported and unreported crime. As such it provides a measure of trends in crime not affected by changes in reporting, or changes in police recording rules or practices In recent years has measured crime every two years. From 2001 the BCS has moved to an annual cycle Measures based on estimates from a sample of the population. The estimates are therefore subject to sampling error and other methodological limitations Has not measured crime at the small area level well, but more reliable regional information will be available from 2001 onwards

Police recorded crime Collected since 1857. Provides measure of offences both reported to and recorded by the police. As such they are influenced by changes in reporting behaviour and recording rules and practices The police figures are published annually in Home Office statistical bulletins Only includes ‘notifable’ offences which the police have to notify to the Home Office for statistical purposes Provides an indicator of the workload of the police

Provides data at the level of police force areas and for Basic Command Units (similar in size to Local Authorities)

Does not include crimes against: Those under 16 Commercial and public sector establishments Those in institutions, and the homeless

Includes crime against: Those under 16 Commercial and public sector establishments Those in institutions, and the homeless

Does not measure: Victimless crimes Crimes where a victim is no longer available for interview Fraud Sexual offences (due to the small number of incidents reported to the survey and concerns about willingness of respondents to disclose such offences, estimates are not considered reliable)

Measures: Victimless crimes Murder and manslaughter Fraud Sexual offences where these have been reported to the police

Collects information on what happens in crime (e.g., when crimes occur, and effects in terms of injury and property loss)

Collects information about the number of arrests, who is arrested, the number of crimes detected, and by what method

Provides information about how the risks of crime vary for different groups

Does not show which groups of the population are most at risk of victimisation

Source: Kershaw et al. (2000)

4.2.3

Police Recorded Crime Datasets

There are two accepted ways of measuring crime in the UK. The first way is by police figures, which reflect recorded crimes. The second way is by the British Crime Survey as mentioned in the previous section. However, it has long been recognised that police recorded crime does not represent the total crime picture because not all offences are reported to the police. The 2001/2002 BCS suggests that only 42 % of offences were reported to the police


61

(Simmons, 2002). Levels of crime reporting tend to vary with the type and the seriousness of the offence. Generally speaking, less serious crimes have a lower probability of being reported to the police than more serious crimes. Moreover, not all crime reported to the police will be recorded. As Simmons (2002) note, the number of crimes that are recorded by the police are dependent on, firstly, the victim or a representative of the victim bringing that crime to the attention of the police or on the crime coming to the attention of the police through some other means (such as the police officer being present at the time), and then whether that incident is determined as being recordable.

Changes in the way police record crime also affects crime figures. In 1998 and in 2002 there were changes in the Home Office Counting Rules for the counting and classifying of notifiable offences recorded by the police forces in England and Wales (Home Office, 2006a). In particular, in 2002 the Counting Rules were revised to incorporate the National Crime Recording Standard (NCRS). In April 2002 the NCRS was introduced across police forces to improve consistency in how police record crime (Home Office, 2006b). In many cases the NCRS has led to an increase in police recorded crime figures. However, not all crime types are similarly affected. The impact of the recording changes varies considerably between different types of recorded crime. Information about this change can be found in Simmons et al. (2003a, b).

However, information on crime as reported to the police is the principal database enabling us to study the geography and determinants of crime in Leeds in Chapter 5 and to validate the model outputs in Chapter 7 and 9. Four fiscal years of recorded crime datasets were obtained from the West Yorkshire Police:

April 2000- March 2001




Problems related to the quality of recorded crime data, from a geographical perspective, are widely known (Shepherd et al., 2004). For example, some data have no geo-referencing and others contain geo-referencing errors. Therefore, it is important to audit the quality of crime data before using them (Bowers and Hirschfield, 2001). However, these datasets were cleaned to produce geographically more reliable data. This was done by the West Yorkshire Police divisional intelligence analysts. This has improved the spatial accuracy and reliability of the crime data quite considerably. Table 4.6 shows the variables in the recorded crime


62

datasets. The advantage of this geo-coded data is that it can be plotted in a Geographic Information System (GIS) which could be supplemented with socio-economic data. These can then be aggregated up to any spatial zone such as police beat, police division, output area, community area, or ward. It can be argued that geographical referencing provides a better understanding, not only of geographical variations in certain types of offending, but also of the relationship between the number of crimes/crime rate and other variables from the census data. It should be noted, however, that victim and offender information in these datasets are not complete, and that the accuracy varies by crime type. There are 14 crime types recorded by the police including: 1) Burglary dwelling 2) Burglary elsewhere 3) Criminal damage 4) Drug offences 5) Fraud and forgery 6) Handling 7) Homicide 8) Other crime 9) Other theft 10) Robbery 11) Sexual offences 12) Theft from motor vehicle 13) Theft of motor vehicle 14) Violent crime


63

Table 4.6: Details on recorded crime Variables CRIMENUMBE DATEENTERE HOCLASS OFFENCE STATUS CRIMETYPE DATEFROM TIMEFROMH TIMEFROMM DATETO TIMETOH TIMETOM DIVISION BEAT FEATURE HOUSENUMBE STREETNAME AREA TOWNCITY POSTCODE OSREFERENC EASTING NORTHING IMPROVED VICTIMAGE VICTIMGEND VICTIMETHN NOMINALNUM OFFENDERAG OFFENDERGE OFFENDERET OFFVICRELA POLICESTN

Detail Crime number Date entered Sub-group of offence type Offence types Status (detected/undetected) Crime type Date from Time from (hour) Time from (minute) Date to Time to (hour) Time to (minute) Police division Police beat Detailed on where the crime occurs (ex. Roadside, garage etc.) House number Street name Area Town city Postcode OS reference Easting Northing Improved Victim age Victim gender Victim ethnicity Nominal number Offender age Offender gender Offender ethnicity Offender and victim relationship Crime committed in the police station

Source: West Yorkshire Police

4.2.4

Offender Dataset

An offender dataset was also derived from the West Yorkshire Police for the period 20002004 giving 70,645 records in total. It should be noted that in fact this dataset provides a list of people linked with offences (which is a so-called ‘nominal’ dataset). Such individuals might be wanted for questioning, they might be suspects, they might have committed the crime and/or they might eventually be prosecuted for the crime. However, for this study it has not been possible to disaggregate the ‘nominals’ into these categories. The ‘nominal’ is replaced by the word ‘offender’ for the analysis in chapter 7, 8 and 9. Some of these people will turn out not to have been involved in the crime to which they have been linked. The relationship of crimes to ‘nominals’ is ‘many to many’ which means one crime may have been committed by one or many people and one person may be linked with one or more than one crime.


64

From detected offences, the postcode of the offence and the postcode of the offender were recorded (Table 4.7). However, it has been noted that the police may not have charged all the offenders associated with a crime. In this dataset (in some cases) the offender’s postcode and ward may not be their home address at the time of the offence. It can be the case that the postcode and ward reflects the address they were arrested from.

As with the police recorded crime data the x y coordinates of known offenders can be used for geographical referencing and plotted in a Geographic Information System. This dataset provides information on ‘where crimes occur’ and ‘where the offenders live’. The movement of offenders can be explored and the travel distance of offenders can be calculated. It can also provide a picture of areas associated with offenders and their demographics through links with the census data.

Table 4.7: Detailed data on known offenders

Variables ID U_ID CRIMENO HOCLASS OFFENCE OFFENCEEAS OFFENCENOR NOMINAL NOMAGE NOMGENDER NOMETHNICI NOMPOSTCID NOMEAST NOMNORTH DRUGS

Detail Number of case ID number of each case Crime number Sub-classes of offences Offence type x y co-ordinate of offences Offender reference Age of offenders Gender of offenders Ethnicity of offenders Postcode of offenders x y co-ordinate of offenders For the drugs offences: supply or possession

Source: West Yorkshire Police

4.3 Concluding Comments This chapter has reviewed the available data usage for modelling crime in Leeds. The Census Area Statistics of the 2001 Census and the 2001/2002 British Crime Survey (BCS) are the main data sources identified for building a spatial microsimulation for crime analysis. They are used to build a micro-level population at the output area spatial scale in Chapter 6. Although the census provides a comprehensive spatial coverage for small geographical


65

areas, it has limitations. To protect the confidentiality of individuals, the output is sometimes modified and the raw microdata are not released. The limitation of the UK census is the discrepancies in census counts between tables. This problem can be tackled by a method described in § 6.2.2. The BCS provides a rich source of individual and household information together with crime victimisation data, but is not available for small geographical areas. One solution to this lack of detail is to combine the BCS with the geographically disaggregated census data to create synthetic microdata for small areas. In particular, data from the BCS is used in Chapter 6 to add victim of crime variables to the census data. It should be noted that this new microdata dataset includes attributes from all the original datasets and this microdata dataset can be re-aggregated to any spatial scale. Further, police recorded crime data and data on known offenders were also reviewed and their advantages and drawbacks were discussed. The police recorded crime datasets are used for two purposes in this study; to study the geography and determinants of crime in Leeds in Chapter 5 and to validate the model output in Chapter 7 and Chapter 9. The dataset on known offenders is used in Chapter 8 to study the movement of offenders using a spatial interaction model.

Chapter 5- Geography and Determinants of Crime in Leeds

66

Chapter 5 Geography and Determinants of Crime in Leeds 5.1 Introduction 5.2 Crime Figures and Trends 5.3 Geographical Variations 5.3.1 Police Divisions 5.3.2 Wards 5.4 Findings by Crime Type 5.4.1 Burglary Dwelling 5.4.2 Burglary Elsewhere 5.4.3 Criminal Damage 5.4.4 Drug Offences 5.4.5 Fraud and Forgery 5.4.6 Handling 5.4.7 Homicide 5.4.8 Other Crime 5.4.9 Other Theft 5.4.10 Robbery 5.4.11 Sexual Offences

5.4.12 Theft from Motor Vehicle 5.4.13 Theft of Motor Vehicle 5.5 Known Offenders and Victims Characteristics 5.5.1 Known Offender Characteristics 5.5.2 Known Victim Characteristics 5.6 The Relationship between Crime and its Related Determinants 5.6.1 Population/Household Density 5.6.2 Demographic Characteristics 5.6.3 Percentage of Students 5.6.4 Rented Tenure Type 5.6.5 Number of Car per Household 5.6.1 Unemployment 5.6.2 Deprivation 5.6.3 Number of Offenders 5.6.4 Multiple Regression Model 5.7 Concluding Comments

5.1 Introduction Within this chapter, information on crime as reported to the police is used to study the geography of crime in Leeds. Four fiscal years of recorded crime (2000/01, 2001/02, 2002/03, 2003/04) from West Yorkshire Police (WYP) are used for the analyses. Each period was recorded between the 1st April and 31st March. Information related to the recorded crime dataset has already been described in § 4.2.3. Plainly there are also crimes which are not recorded to/recorded by the police, and later chapters will attempt to deal with this issue – here only reported crimes are dealt with. Crime statistics in this chapter are presented in the form of ‘number’ and ‘rate’. Crime rate relates the incidence of crime to population totals, so that crime levels can be compared consistently without bias from population size. In this study ‘crime rate’ is calculated as per 1,000 population except burglary dwelling which is per 1,000 households. Note that most of the crime statistics presented in this chapter are similar to the work that the author has completed for ‘An Analysis of Crime and Disorder in Leeds 2000/01 to 2003/04’. This work was undertaken in the School of Geography, University of Leeds and was commissioned by the Leeds Community Safety Partnership.

This chapter starts with a review of crime statistics and overall trends at national, regional, and city levels. This is followed by an analysis of crime at police division and ward level in § 5.3 and findings by crime type in § 5.4. Section 5.5 describes known offenders’ and


67

victims’ characteristics. In § 5.6 the relationship between crime and its related determinants are explored. The final section gives some concluding comments.

5.2 Leeds Crime Figures and Trends The population of Leeds, as measured in the 2001 Census, was 715,402 of which 48% were male and 52% were female. The population of Leeds accounted for one third of the residents of the county of West Yorkshire. In 2003/04, 39.1% of all crime committed in West Yorkshire took place in Leeds with the rate of 177.95 crimes per 1,000 population. This was higher than the average crime rate in West Yorkshire, which was at 155.8 per 1,000 population and significantly in excess of the national picture at 114.04 per 1,000 population (Table 5.1). However, the number of crimes and the crime rate in Leeds dropped in 2003/04. There were 127,304 offences recorded in Leeds in 2003/04, which represents a 1.5% decrease in crimes over the previous year. However there has been a 20.32% increase overall since 2000/01.

Table 5.1: Number of crime and crime rate (all crime) in Leeds compared to West Yorkshire and England and Wales Leeds Year

West Yorkshire

Number

Rate*

Up/ down

2003/2004

127,304

177.95

2002/2003

129,254

180.67

2001/2002

120,623

168.61

2000/2001

105,803

147.89

England & Wales

Number

Rate*

Up/ down

Number

Rate*

Up/ down

ª

325,556

155.80

©

5,934,580

114.04

©

©

322,794

154.45

©

5,899,450

113.36

©

©

298,847

143.73

©

5,527,082

106.20

©

258,908

124.29

5,170,843

102.28

Source: West Yorkshire Police and Home Office Note: Rate* is per 1,000 population © Increase from the previous year ª Decrease from the previous year Its not surprising that there was a large increase across the country in the period of 2002/03 because in April 2002 the National Crime Recording Standard (NCRS), which aims to bring greater consistency in crime recording, was introduced across police forces. However, some police forces adopted the recording practices earlier, including West Yorkshire, where the NCRS was adopted in February 2002. As a result of the introduction of NCRS recorded crime rose considerably across the country. It is estimated that for 2002/03, the overall impact of the NCRS has been to increase recorded crime by 10% (Simmons and Dodd, 2003). This reflects a change in recording practice and is not a real increase in crime. It should be noted, however, that the impact has varied considerably by type of offence. The effects of NCRS on West Yorkshire are summarised in Table 5.2.


68

Table 5.2: The effects of NCRS on West Yorkshire Offence NCRS

Impact (%)

Violence against person Burglary dwelling Burglary other All burglary Robbery Vehicle theft Other theft All theft Criminal damage

47 1 9 3 -9 24 61 40 27

Total crime

25

Source: Simmons et al.(2003b)

0

5,000

10,000

15,000

20,000

25,000

30,000

Burglary Dw elling Burglary Elsew here Criminal Damage Drugs Off ences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual off ences Thef t From Motor Vehicle

2000/01 2001/02

Thef t Of Motor V ehicle Violent Crime

Figure 5.1: Leeds Crime. Source: Derived from West Yorkshire Police

2002/03 2003/04


69

Table 5.3: Leeds Crime figures and trends (2000/1-2003/04) 2000/01

2001/02

Crime Type Number Burglary Dwelling

Crime Rate per 1000 Pop

13,607

Burglary Elsewhere

45.11*

11,478 Burglary 25,085

Criminal Damage

15,693

52.03*

12,429 28,122

17.37 39.31

15.33% 8.29% 12.11%

Number 16,374

54.29*

12,179 28,553

2003/04

Change Crime Rate From the per 1000 Pop previous year

17.02 39.91

4.34% -2.01% 1.53%

Number


13,833

45.86*

10,849 24,682

15.16 34.50

-15.52% -10.92% -13.56%

18,304

25.59

21,646

30.26

18.26%

22,826

31.91

5.45%

24,952

34.88

9.31%

275

0.38

267

0.37

-2.91%

312

0.44

16.85%

310

0.43

-0.64%

Handling Other Theft

16.04 35.06

Number

2002/03


21,999

30.75

25,455

35.58

15.71%

27,421

38.33

7.72%

27,290

38.15

-0.48%

Robbery

2,285

3.19

3,307

4.62

44.73%

2,677

3.74

-19.05%

1,981

2.77

-26.00%

Fraud & Forgery

4,336

6.06

5,366

7.50

23.75%

7,460

10.43

39.02%

7,088

9.91

-4.99%

15,477

21.63

16,864

23.57

8.96%

16,064

22.45

-4.74%

14,145

19.77

-11.95%

9,093

12.71

9,624

13.45

5.84%

9,321

13.03

-3.15%

7,339

10.26

Theft from Motor Vehicle Theft of Motor Vehicle

Vehicle Crime 24,570 Total Property Crime

34.34

96,854

Homicide Sexual Offences

26,488

135.38

37.03

110,651

14

0.02

154.67

7.81%

25,385

14.25%

114,634

10

0.01

-28.57%

35.48 160.24

-4.16%

21,484

3.60%

107,787

14

0.02

40.00%

30.03

-21.26% -15.37%

150.67

-5.97%

15

0.02

7.14% 6.10%

488

0.68

538

0.75

10.25%

754

1.05

40.15%

800

1.12

Drug Offences

1,694

2.37

1,559

2.18

-7.97%

2,087

2.92

33.87%

2,212

3.09

5.99%

Violent Crime

5,813

8.13

6,825

9.54

17.41%

10,734

15.00

57.27%

15,308

21.40

42.61%

Other Crime Violent Crime & Other Total

940 8,949 105,803

1.31 12.51

1,040 9,972

147.89

120,623

1.45 13.94 168.61

Source: Derived from West Yorkshire Police Note: * Rate is expressed per 1000 households Total population = 715,402 and total households = 301,614

10.64% 11.43% 14.01%

1,031 14,620 129,254

1.44 20.44 180.67

-0.87% 46.61% 7.16%

1,182 19,517 127,304

1.65 27.28 177.95

14.65% 33.50% -1.51%


70

Table 5.4: Recorded crime in England and Wales by offence 2000/01 to 2003/04 and percentage change between 2002/03 and 2003/04 Crime Type

2000/01

2001/02

2002/03

2003/04

Change between 2002/03 and 2003/04

Burglary Dwelling

402,984

430,347

437,571

402,333

-8.05%

Burglary Elsewhere

433,043

448,162

451,256

416,309

-7.74%

Criminal Damage

960,087

1,064,495

1,109,258

1,205,576

8.68%

Drug Offences

113,458

121,393

141,101

141,060

-0.03%

Fraud & Forgery

319,324

314,859

330,104

317,949

-3.68%

850

891

1,043

853

-18.22%

63,188

65,683

72,461

74,193

2.39%

Homicide Other Crime Other Theft and Handling

1,176,925

1,283,688

1,389,359

1,378,972

-0.75%

Robbery

95,154

121,359

108,032

101,195

-6.33%

Sexual Offences

37,311

41,432

48,644

52,070

7.04%

Theft from Motor Vehicle

629,651

655,161

658,697

598,514

-9.14%

Theft of Motor Vehicle

328,037

316,321

305,618

279,111

-8.67%

Violent Crime

600,072

649,439

833,884

954,899

14.51%

Source: Dodd et al. (2004), Crime in England and Wales 2003/2004 (Home Office Statistical Bulletin 10/04)

Figure 5.1 shows Leeds crime figures between 2000/01 and 2003/04 while Table 5.3 shows more details on crime numbers, rates per 1,000 population, and changes from the previous period by crime type. Table 5.4 details recorded crime in England and Wales by offence for the period 2000/01 to 2003/04 and percentage change between 2002/03 and 2003/04 (which can be compared to the right column of Table 5.3 that shows ‘percentage change’ in Leeds for the same period).

The total number of recorded crimes in Leeds increased significantly from 105,803 in 2000/01 to 120,623 in 2001/02 or from 147.89 per 1,000 population to 168.61 per 1,000 population (a 14% increase). Then crime increased by 7.16% from 120,623 in 2001/02 to 129,254 in 2002/03. This rise partly reflects the implementation of the NCRS. However, the total number of recorded crimes has decreased slightly from 129,254 in 2002/03, to 127,304 in 2003/04 (by 1.5%) and some offences have fallen substantially, such as vehicle crime, while some offences such as violent crime continue to rise.

As can be seen from Table 5.3 the crime figures are dominated by ‘property crimes’ accounting for 84.67% of total crimes in Leeds for 2003/04. Burglary, criminal damage, other theft, and vehicle crime are the crimes accounting for the largest proportion. In terms of rising crime over the four year period, criminal damage and violent crime do not fair well


71

compared to other crime types. In percentage terms, violent crime is the only type that has increased significantly and continuously while burglary elsewhere and vehicle crime have decreased continuously in Leeds over the same period.

In 2003/04, burglary dwelling dropped to the previous levels of 2000/01 (down 15.52% from the previous year) with the rate 45.86 offences per 1,000 households. This is actually a greater improvement compared with the reductions being seen across England and Wales (down 8.1%) (Table 5.4). The rate of burglary elsewhere in 2003/04 was 15.16 per 1,000 populaion, down 10.92% from the previous year which is also better than the England and Wales rate (down 7.7%). Criminal damage has risen by 9.3%, quite similar to England and Wales. Although there was a small drop in fraud and forgery (5%) in 2003/04, there has been a 63.5% rise since 2000/01 whilst there was almost no change over the same period (2000/01 and 2003/04) in England and Wales. Other theft rate was 38.15 offences, down 0.5% on the previous year, quite similar to England and Wales. In the year 2003/04 robbery has fallen considerably, down 26% compared with England and Wales which fell just 6.3%. There was a small rise (up 6.1%) in sexual offences in 2003/04 which is quite similar to the national trend. The rate of theft from motor vehicle and theft of motor vehicle are almost double the England and Wales average. However, the theft from motor vehicle rate dropped by 11.95% compared to a national drop of 9.1% and the theft of motor vehicle dropped 21.3% compared to a national drop of only 8.7%. Note that in 2003/04 burglary elsewhere, robbery and vehicle crime were at their lowest level in 4 years.

Violent crime in Leeds is above the national average and rose 42.61% compared to a rise of 14.5% across England and Wales. Although violent crime has risen by 124.29% during 2001/02 and 2003/04, this figure should be noted with caution. Change in the way violent crime is recorded has affected rates, making it difficult to assess the true trend over the last three years (2001/02-2003/04). As mentioned in Simmons et al. (2003a), prior to the introduction of the NCRS many of these offences, particularly low-level violence offences, might have been dealt with by way of advice and not necessarily recorded as crimes, because they seldom involved injury.


72

Table 5.5 shows detection rates in Leeds compared to England and Wales. As mentioned in Thomas and Feist (2004), for any crime to be counted as ‘detected’ the following conditions must apply:

A notifiable offence has been committed and recorded;

A suspect has been identified (and interviewed, or at least informed that the crime has been cleared up);

There is sufficient evidence to charge the suspect;

The victim has been informed that the offence has been cleared up.

Note that detections are counts on the basis of crimes, rather than offenders. For example, if three offenders are involved in a robbery, and are all arrested and charged, this counts as one not three detections. If only one of them is identified and charged, while the other two remain unidentified and go free, this also counts as one detection.

Table 5.5: Detection rates in Leeds compared to England and Wales Leeds1

Crime Type Burglary dwelling Robbery Theft from motor vehicle Theft of motor vehicle Sexual offences Violent crime Total

England and Wales2

2002/03

2003/04

2002/03

2003/04

62.7% 50.4% 23.2% 13.1% 11.7% 3.5%

50.2% 36.8% 27.7% 11.8% 11.8% 5.4%

53.9% 43.3% 18.5% 14.6% 13.4% 6.1%

50.2% 39.2% 18.4% 15.0% 13.1% 6.3%

18.5%

15.4%

23.6%

23.5%

Source: 1 Derived from West Yorkshire Police 2 Home Office

As can be seen from Table 5.5, the detection rate in England and Wales in 2003/04 remained stable at around 23.5% (there was a very slight decrease of 0.1 percentage points between 2002/03 and 2003/04). While the detection rates in England and Wales have remained quite constant over the last two years there has been a marked drop in the detection rate in Leeds. The overall detection rate in Leeds in 2003/04 was 15.4%, much lower than the England and Wales average of 23.5%. However, detection rates have improved for Theft from motor vehicle and violent crime.


73

5.3 Geographical Variations 5.3.1

Police Divisions

There are four police divisions in Leeds (Figure 5.2): 1) Weetwood and Pudsey 2) Chapeltown 3) Killingbeck 4) City and Holbeck

D iv ision O ldw a rds D ivis io n s C h apleto w n C ity a nd H olb ec k K illing b ec k W e etw o o d a nd P ud sey

Figure 5.2: Leeds Police Division Source: West Yorkshire Police

The ‘Weetwood and Pudsey Division’ is based in the North and West of Leeds. It covers the University of Leeds and Leeds Metropolitan University and large student areas including Headingley, Burley, and Woodhouse. It also houses one of the largest prisons in the country at Her Majesty’s Prison (HMP) Leeds (in the Armley ward). ‘Chapeltown Division’ incorporates the North Eastern area of Leeds including inner city and rural areas as well as prosperous suburbs. ‘Killingbeck Division’ covers the Eastern area of Leeds. ‘City and Holbeck Division’ covers the city centre and stretches to the district’s southern boundary incorporating Morley, Rothwell and surrounding villages. The city centre has an enormous transient population made up in the daytime of thousands of people travelling to the city to work and shop.


74

Table 5.6: Number of crime by division (2000/01 to 2003/04) Offence Types Burglary Dwelling Burglary Elsewhere Criminal Damage Drugs Offences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime Total

Weetwood & Pudsey 2000/01 2001/02 2002/03 2003/04 5,834 6,802 7,159 7,103 3,594 3,817 3,659 3,688 6,526 7,190 8,003 8,882 528 511 636 651 1,226 1,707 2,435 2,624 73 63 77 92 5 3 6 6 260 309 342 420 5,952 6,767 7,480 7,574 734 1,162 984 790 148 208 209 278 5,408 5,632 5,736 5,104 3,376 3,393 3,084 2,584 1,792 2,074 3,420 4,934 35,456

39,638

43,230

44,730

2000/01 3,367 2,155 3,355 360 679 45 4 248 3,109 686 104 2,815 1,987 1,032 19,946

Chapeltown 2001/02 2002/03 3,709 3,354 2,099 1,940 3,741 3,585 381 597 843 1,043 46 52 1 2 213 198 3,134 3,286 870 666 113 218 2,614 2,564 1,520 1,444 1,181 1,707 20,465

20,656

2003/04 2,443 1,611 3,769 590 1,108 58 4 234 3,559 538 199 2,345 1,238 2,659

2000/01 2,506 1,945 4,132 253 667 58 2 131 2,726 288 93 1,867 1,463 852

20,355

16,983

2003/04 39.90 11.07 25.90 4.05 7.61 0.40 0.03 1.61 24.46 3.70 1.37 16.12 8.51 18.27

2000/01 45.32 14.57 30.94 1.89 5.00 0.43 0.01 0.98 20.42 2.16 0.70 13.98 10.96 6.38

139.89

127.19

Killingbeck 2001/02 2002/03 2,954 3,048 2,280 2,430 4,785 5,009 212 291 841 1,248 57 48 2 1 151 130 3,477 3,764 334 347 63 104 1,935 1,913 1,799 1,919 1,098 1,705

2003/04 2,424 1,990 5,948 329 922 67 2 159 3,408 260 141 1,633 1,576 2,995

2000/01 1,900 3,784 4,291 553 1,764 99 3 301 10,212 577 143 5,387 2,267 2,137

21,957

21,854

33,418

Killingbeck 2001/02 2002/03 53.42 55.12 17.08 18.20 35.84 37.51 1.59 2.18 6.30 9.35 0.43 0.36 0.01 0.01 1.13 0.97 26.04 28.19 2.50 2.60 0.47 0.78 14.49 14.33 13.47 14.37 8.22 12.77

2003/04 43.84 14.90 44.54 2.46 6.90 0.50 0.01 1.19 25.52 1.95 1.06 12.23 11.80 22.43

2000/01 29.35 25.07 28.43 3.66 11.69 0.66 0.02 1.99 67.65 3.82 0.95 35.69 15.02 14.16

163.67

221.39

19,988

City & Holbeck 2001/02 2002/03 2,228 2,813 4,233 4,150 5,930 6,229 455 563 1,975 2,734 101 135 4 5 367 361 12,077 12,891 941 680 154 223 6,683 5,851 2,912 2,874 2,472 3,902 40,532

2003/04 1,863 3,560 6,353 642 2,434 93 3 369 12,749 393 182 5,063 1,941 4,720

43,411

40,365

City & Holbeck 2001/02 2002/03 34.42 43.46 28.04 27.49 39.29 41.27 3.01 3.73 13.08 18.11 0.67 0.89 0.03 0.03 2.43 2.39 80.01 85.40 6.23 4.50 1.02 1.48 44.27 38.76 19.29 19.04 16.38 25.85

2003/04 28.78 23.58 42.09 4.25 16.13 0.62 0.02 2.44 84.46 2.60 1.21 33.54 12.86 31.27

Table 5.7: Crime rates (per 1,000 population) by division (2000/01 to 2003/04) Offence Types Burglary Dwelling* Burglary Elsewhere Criminal Damage Drugs Offences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime Total

Weetwood & Pudsey 2000/01 2001/02 2002/03 2003/04 48.52 56.57 59.54 59.07 12.60 13.38 12.83 12.93 22.88 25.21 28.06 31.14 1.85 1.79 2.23 2.28 4.30 5.99 8.54 9.20 0.26 0.22 0.27 0.32 0.02 0.01 0.02 0.02 0.91 1.08 1.20 1.47 20.87 23.73 26.23 26.56 2.57 4.07 3.45 2.77 0.52 0.73 0.73 0.97 18.96 19.75 20.11 17.90 11.84 11.90 10.81 9.06 6.28 7.27 11.99 17.30 124.32

138.98

151.57

156.83

2000/01 54.99 14.81 23.06 2.47 4.67 0.31 0.03 1.70 21.37 4.71 0.71 19.35 13.66 7.09 137.07

Chapeltown 2001/02 2002/03 60.57 54.78 14.42 13.33 25.71 24.64 2.62 4.10 5.79 7.17 0.32 0.36 0.01 0.01 1.46 1.36 21.54 22.58 5.98 4.58 0.78 1.50 17.96 17.62 10.45 9.92 8.12 11.73 140.64

141.95

149.69

164.44

268.52

287.60

267.42

Source: Derived from West Yorkshire Police Note: Burglary dwelling rate is per 1,000 households. ‘Number of households and population’ calculated for Table 5.7 are from local base statistics produced by Leeds City Council. Source: Leeds Statistics (2004).


75

Table 5.6 shows the ‘number of crimes’ by police division yearly (2000/01-2003/04) and Table 5.7 show ‘crime rates’ for the same period. The rates may be different slightly from other published sources; in this case the number of households and population totals are from local base statistics produced by Leeds City Council.

In 2003/04, only the ‘Weetwood and Pudsey division’ had a rise (up 3.5%) in overall crime on the previous year while the rest had seen a decrease. ‘City and Holbeck’ was down 7% while ‘Killingbeck’ and ‘Chapeltown’ decreased by 0.5% and 1.5% respectively. Overall crime in ‘Weetwood and Pudsey’ rose steadily from 2000/01, particularly for criminal damage, fraud and forgery, other crime, other theft, sexual offences, and violent crime. In ‘Chapeltown’, fraud and forgery, Handling, other theft, and violent crime rose steadily while vehicle crimes declined steadily from 2000/01. In ‘Killingbeck’, it has been found that there has been a steady increase in criminal damage, and violent crime. ‘City and Holbeck’ also has seen a steady increase in violent crime (Table 5.6 and Table 5.7).

Crime Rate per 1,000 Population 350.00

Crime Rate per 1,000 Population

300.00

250.00

200.00

150.00

100.00 Weetw ood and Pudsey Chapeltow n

50.00

Killingbeck City and Holbeck

0.00 2000/01

2001/02

2002/03

2003/04

Ye ar

Figure 5.3: Crime rates per 1,000 population by police division Source: Derived from West Yorkshire Police

Figure 5.3 shows crime rates per 1,000 population by division over the four year period. As can be seen the crime rate in the ‘City and Holbeck’ division was higher than the other three divisions with a rate of 267.42 crimes per 1,000 population. ‘City and Holbeck’ has the highest rate for every crime type except for burglary dwelling, which is highest in ‘Weetwood and Pudsey’ (Table 5.7).


-60%

-40%

-20%

0%

20%

40%

76

60%

80%

100%

Weetw ood and Pudsey

la Bu r g

ry Dw

ellin g

Chapeltow n Killingbeck City and Holbeck

la Bur g

ew h e r y Els

C r im

F ra u

inal D

d&F

O the

a ma

or ge

m Mo

Of T heft

ry

er y

e hic tor V

M oto

hic r Ve

Viole

ge

ft r The

R o bb

F ro T heft

re

n t C ri

le

le

me

Figure 5.4: Percentage changes (between 2002/03 and 2003/04) of selected crime types Source: Derived from West Yorkshire Police

Figure 5.4 shows changes (between 2002/03 and 2003/04) for selected crime types by police division. Over the period of 2003/04 there have been large falls in the burglary dwelling rate in all divisions except ‘Weetwood and Pudsey’ (a slight decrease). There is more variation in burglary dwelling rates at ward level as described in the next section. Burglary elsewhere decreased in every division except ‘Weetwood and Pudsey’ (a slight increase). By contrast, criminal damage increased in all divisions. This reflects the impact of the National Crime Recording Standard (NCRS) on criminal damage which means that many more minor offences are being recorded than before. Fraud and forgery decreased in ‘Killingbeck’ but increased in ‘Weetwood and Pudsey’ and ‘Chapeltown’. Other theft remained stable in ‘Weetwood and Pudsey’ and ‘City and Holbeck’. There was a large decrease in robbery in every division especially in ‘City and Holbeck’ (down 42.2%). Vehicle crimes also had a decrease in every division. However, violent crime has risen substantially in every division, especially ‘Killingbeck’ (up 75.7%).


77

Wetherby Otley and Wharfedale North

Aireborough

Cookridge Horsforth

Moortown

Roundhay

Weetwood Pudsey North

Whinmoor

Chapel Allerton KirkstallHeadingley Harehills

Bramley

Armley

University

Seacroft Barwick and Kippax

Burmantofts Halton

Pudsey South

City and Holbeck

Richmond Hill

Wortley Beeston

Hunslet Garforth and Swillington

Morley North Middleton

Morley South

Figure 5.5: Leeds wards

Rothwell


5.3.2

78

Wards

The city of Leeds is comprised of 33 wards as shown in Figure 5.5. Note that the ward boundaries used throughout this thesis are those defined at the time of the 2001 Census (which were used officially up to the June Election 2004). After this date the boundary lines were changed.

Details on the number of recorded crime for each crime type by ward (2000/01 to 2003/04) are given in Appendix A. In 2003/04, the wards with the highest percentage of recorded crime were City and Holbeck (16.53%), University (5.66%), Armley (4.4%), and Burmantofts (4.14%). Relative to the previous year, the largest increases (in 2003/04) occurred in Wetherby (up 32.7%), Armley (up 18.03%), Pudsey South (up 17.5%), and Otley and Wharfedale (up 15.34%). In contrast, the wards with the largest decreases in crime were Halton (down 18.99%), Morley North (down 16.5%), North (down 15.81%), and Headingley (down 14.61%).

In relation to population size, crime rates vary widely within each ward (Table 5.8). Crime rate per 1,000 population was the highest in the City and Holbeck ward (1,017.51 per 1,000 population). The small number of residents living in City and Holbeck produces ‘artificially’ high crime rates as most crimes are committed against the transient not inhabiting population. The next highest rate is University, followed by Burmantofts, Richmond Hill, Armley, Chapel Allerton, Hunslet, Seacroft, and Kirkstall. Note that all of these wards have crime rates higher than the Leeds average (183.43 crimes per 1,000 population). The lowest crime rates are recorded in Wetherby with the rate 71.28 crimes per 1,000 population, and Barwick and Kippax (75.77 per 1,000 population).

From the four years recorded crime, it has been found that 2,231 of the records (2000/01: 470 records; 2001/02: 563 records, 2002/03: 578 records; and 2003/04: 620 records) pertain to offences committed inside police stations, prisons and young offender institutions (especially drug offences and violent crime). This number seems to be a very small percentage of all crime committed in Leeds. However, it is likely that the presence of a police station or prison could bias statistics at the ward level, especially in Armley, City and Holbeck, Chapel Allerton, and Seacroft. Figure 5.6 depicts crime rates by ward. As can be seen the highest crime rates in Leeds are found in close proximity to the city centre.


79

Table 5.8: Leeds Crime rates by ward (2003/04)

BELOVW

ABOVE

Ward City and Holbeck University Burmantofts Richmond Hill Armley Chapel Allerton Hunslet Seacroft Kirkstall Average by ward

Crime Rate per 1000 Population 1,017.51 337.11 284.65 259.32 254.33 228.88 223.40 209.53 207.57 183.43

Headingley Beeston Bramley Harehills Wortley Weetwood Middleton Pudsey South Roundhay Morley South Morley North Halton Pudsey North Rothwell Whinmoor Moortown Horsforth Cookridge Aireborough Otley and Wharfedale Garforth and Swillington North Barwick and Kippax Wetherby

182.19 180.87 178.45 174.52 164.27 151.92 146.13 133.20 132.38 132.12 130.30 129.58 123.71 123.42 121.75 113.44 112.05 97.78 96.85 88.42 88.15 82.40 75.77 71.28

Source: Derived from West Yorkshire Police

All C rim e 71.2 8 - 97.7 8 97.7 8 - 151 .92 151 .92 - 22 8.88 228 .88 - 33 7.11 337 .11 - 10 17 .5 1

Figure 5.6: Crime rate per 1,000 population (2003/04). Source: Derived from West Yorkshire Police


80

5.4 Findings by Crime Type Note that crime rates presented in this section are based on the period 2003/04 and the changes are based on the difference between 2002/03 and 2003/04. Details on the rate for each crime type by ward (2002/03 and 2003/04) can be found in Appendix B. All figures shown in this section are produced from the police recorded crime datasets from West Yorkshire Police.

5.4.1

Burglary Dwelling

% C ha ng e -4 6.2 7 to -30 .0 -2 9.9 9 to -15 .0 -1 4.9 9 to 0 1.0 0 to 15 15.0 1 to 3 0 .0 30.0 1 to 4 0 .67

B u rgla ry D w ellin g 18.53 - 2 2.3 22.3 - 30 .5 3 30.53 - 4 1.97 41.97 - 7 2.08 72.08 - 1 04.54

a) Burglary dwelling rate per 1,000 households (2003/04)

b) Change, 2002/03 to 2003/04

Figure 5.7: Burglary dwelling

Recorded burglary dwelling rates vary across the region. Generally speaking, the highest rates are in the inner, urban part and the lowest rates in the most rural (Figure 5.7a). Rates in 2003/04 were the highest in Headingley, a large student population area which presents a relatively easy target for burglars, with the rate of 104.54 offences per 1,000 households. This is much higher than more affluent suburban wards such as Morley North (18.53 offences per 1,000 households), Barwick and Kippax (18.76 offences per 1,000 households), Morley South (19.32 offences per 1,000 households), Wetherby (21.75 offences per 1,000 households), Rothwell (22.22 offences per 1,000 households) and Garforth and Swillington (22.3 offences per 1,000 households).

Figure 5.7b shows the percentage changes between burglary dwelling rates in 2002/03 and 2003/04. By comparing the two maps above it is clear that while the wards in central and east Leeds had higher rates the hotspot areas of Headingley, City and Holbeck, Harehills, and Roundhay wards appeared to show declining number of offences. However, an increase in burglary dwelling offences is noted for Armley and Bramley.


5.4.2

81

Burglary Elsewhere

% Change -54 .4 1 to -30 -29 .9 9 to -15 -14 .9 9 to 0 0.01 to 15 .0 15.01 to 3 0.0 30.01 to 8 3.53

B u r glary E ls ew h ere 5.8 9 - 9.2 9.2 - 12.98 12.98 - 1 7.2 4 17.24 - 2 2.3 1 22.31 - 6 0.0 2

a) Burglary elsewhere rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.8: Burglary elsewhere Although Headingley has a very high burglary dwelling rate it has a very low burglary elsewhere rate (7.25 per 1,000 population), that is, burglaries against shops, warehouses etc. City and Holbeck has the highest rate for burglary elsewhere (Figure 5.8a) due to the large number of businesses in this area. However, it dropped 7% from the previous year. In percentage terms the most increases in 2003/04 were in suburban wards in North-West Leeds (relatively affluent) such as Cookridge (up 83.53% or a rise from 170 offences in 2002/03 to 312 offences in 2003/04). Those wards seeing the greatest falls in burglary elsewhere include Harehills (down 54.41%), North (down 40.52%), Hunslet (down 33.21%), and Moortown (down 30.46%) (Figure 5.8b). 5.4.3

Criminal Damage

% C ha ng e -1 7.1 2 to -10 .0 -9 .9 9 to 5 .0 -4 .9 9 to 0 0.0 1 to 15 15.0 1 to 2 5 25.0 1 to 3 3 .21

C rim in al D a m a g e 9.59 - 19 .2 5 19.25 - 3 0.15 30.15 - 4 2.54 42.54 - 6 8.72 68.72 - 1 14.72

a) Criminal damage rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.9: Criminal damage City and Holbeck ward has the highest rate for criminal damage with the rate of 114.72 offences per 1,000 population (Figure 5.9a) but this is mainly because of the use of


82

residential not day time population to calculate the rate. Most wards have increased and problems seem to be worst in Pudsey South (up 33.21%), Armley (up 31.29%), and Garforth and Swillington (30.75%) (Figure 5.9b). However, there was a drop in criminal damage in University (down 17.12%) and Rothwell (down 11.02%).

5.4.4

Drug Offences

% C ha n g e -6 3 .8 6 to -4 5 -4 9 .9 9 to -1 5 -1 4 .9 9 to 0 0.0 1 to 5 0 50 to 2 0 0 20 0 .0 1 to 33 0

Drug Offences 0.37 - 0.9 0.9 - 2.04 2.04 - 3.86 3.86 - 8.1 8.1 - 23.31

a) Drug offences rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04 Figure 5.10: Drug offences

The figures for drug offences comprise offences relating to supply and possession. The highest drug offences rates were in the City and Holbeck and Chapel Allerton wards with rates of 23.31 and 16.92 offences per 1,000 population respectively (Figure 5.10a). The largest decreases in drug offences were in Halton (down 63.86%), Barwick and Kippax (down 60.53), and Aireborough (down 48.15%) (Figure 5.10b). However, these decreases are based on very small numbers. 5.4.5

Fraud and Forgery

% C h an g e -55 .6 7 to -40 -39 .9 9 to -20 -19 .9 9 to 0 0.01 to 30 30.0 1 to 1 00 100 .01 to 33 5.71

Fraud and Forgery 1.36 - 3.76 3.76 - 7.19 7.19 - 12.43 12.4 3 - 20.9 2 20.9 2 - 46.2 9

a) Fraud and forgery rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.11: Fraud and forgery


83

From Figure 5.11a, fraud and forgery appears to show a different pattern from other crime types because there are concentrations across many parts of the city. The reason for this is that petrol stations bare the brunt of these crimes, with 62% of fraud and forgery offences in 2003/04 (Shepherd et al., 2004). Therefore, there are high records in those wards that include a petrol station. The highest rates of fraud and forgery were in City and Holbeck, Armley, and Horsforth. Armley and Horsforth have also seen increases since 2002/03 of 29.13% and 39.68% respectively while City and Holbeck has seen a slight decrease (Figure 5.11b). Barwick and Kippax and Wetherby did not have as many offences as many other wards but their fraud and forgery offences have risen 335.71% and 104% respectively since 2002/03. However, although the largest increases were in Barwick and Kippax and Wetherby, these are also based on very small numbers. 5.4.6

Handling

The number of handling offences is very low across the area with only 310 cases in 2003/04. The very small numbers make aggregation to wards inappropriate. However, the highest rate is found in the City and Holbeck ward with a rate of 2.85 per 1,000 population. The number of offences of handling in City and Holbeck accounted for 20% of all handling offences in 2003/04. 5.4.7

Homicide

#

#

# # ## #

# #

#

## #

# #

#

#

# #

# #

#

# ##

# # #

# # ### ## # ## # #

# # #

# # # #

#

#

# #

H o m ic id e _ 2 0 0 3 /0 4 H o m ic id e _ 2 0 0 2 /0 3 H o m ic id e _ 2 0 0 1 /0 2 H o m ic id e _ 2 0 0 0 /0 1 O ld w a rd s

# #

Figure 5.12: Location of homicides from 2000/01 to 2003/04.

Although the impact of Homicide is very large in human terms, in number terms it remains relatively small. Over the four year period there were only 53 cases in Leeds. Because of this small number it makes the aggregation to wards inappropriate. However, Figure 5.12 shows the location of the offences over the four year period. There is some evidence of clustering in the city centre area, but this is likely to be due to the population density effect.


5.4.8

84

Other Crime

% C h an g e -4 8 to -30 -2 9.9 9 to -10 -9 .99 to 0 0 .0 1 to 22 .73 2 0.01 to 5 0 5 0.01 to 1 45 .83

Other C rim e 0.21 - 0.92 0.92 - 1.36 1.36 - 2.54 2.54 - 4.31 4.31 - 11.46

a) Other crime rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04 Figure 5.13: Other crime

The range of crimes in this category is very varied. The sub-classifications of other crime that have been found in large numbers in Leeds include: other offences (against the state and public order), indecent exposure, dangerous driving, going equipped for stealing, and attempting to pervert the course of public justice. The highest rate of other crime is found in the City and Holbeck ward with a rate of 11.46 offences per 1,000 population (Figure 5.13a). Headingley did not have as many other crime offences as the City and Holbeck ward but its other crime offences have risen by 145.83% since 2002/03 (Figure 5.13b). 5.4.9

Other Theft

O th er T h eft 13.85 20.47 31.31 46.25 63.24 -

a) Other theft rate (2003/04) per 1,000 population

% Change -34 .7 to -2 0 -19 .9 9 to -10 -9.99 to 0 0.01 to 15 15.01 to 3 0 30.01 to 5 2.58

2 0.47 3 1.31 4 6.25 6 3.24 4 35.92

b) Change, 2002/03 to 2003/04

Figure 5.14: Other theft The City and Holbeck ward is the main hotspot for other theft with a rate of 435.92 offences per 1,000 population (Figure 5.14a). It is slightly down (1.87%) on 2003/04 (Figure 5.14b).


85

5.4.10 Robbery

R o b b e ry 0.1 1 0.7 2 1.5 2 3.4 9 5.6 3

-

% C h an g e -5 0 to -40 -3 9.99 to -20 -1 9.99 to 0 0.01 to 20 20.0 1 to 60 60.0 1 to 13 0

0.72 1.52 3.49 5.63 12.91

a) Robbery rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.15: Robbery Generally, robbery rates in Leeds are low for the most part. The highest rates are in the City and Holbeck (12.91 per 1,000 population) and University wards (11.42 per 1,000 population) (Figure 5.15a) but these are the areas that have also seen the largest decreases (Figure 5.15b). Ten wards have seen a rise in robbery. However, all rises are very small numbers. 5.4.11 Sexual Offences

% C ha n g e -5 0 to -3 0 -2 9 .9 9 to -1 5 -1 4 .9 9 to 0 0.0 1 to 7 5 75 .0 1 to 1 00 10 0 .0 1 to 30 0

S e xu a l O ffe n ce s 0.1 9 - 0 .4 2 0.4 2 - 0 .8 6 0.8 6 - 1 .5 2 1.5 2 - 2 .9 5 2.9 5 - 5 .0 3

a) Sexual offences rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.16: Sexual offences The City and Holbeck and Chapel Allerton wards accounted for 24% of sexual offences in 2003/04 with rates of 5.03 and 4.83 offences per 1,000 population respectively (Figure 5.16a). For the most part, those wards that saw high percentage rises (Figure 5.16b) only had a small number of offences. For example, a 300% increase in Horsforth is accounted for by a rise from 4 to 16 offences.


86

5.4.12 Theft from Motor Vehicle

% Change -51 .1 6 to -30 -29 .9 9 to -15 -14 .9 9 to 0 0.01 to 15 15.01 to 3 0 30.01 to 4 7.02

Th eft fro m M oto r V eh icle 4.91 - 9 .5 9 9.59 - 1 4.69 14.6 9 - 33.6 8 33.6 8 - 60.1 60.1 - 1 16.2 7

a) Theft from motor vehicle rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.17: Theft from motor vehicle Although overall crime rates in South Leeds (Morley North and Morley South) were low as indicated in Figures 5.6, theft from motor vehicle in these areas were as high as those in the inner city areas. The rate of theft from motor vehicle per 1,000 population is highest in City and Holbeck ward (116.27 offences per 1,000 population) (Figure 5.17a). The most common items stolen were CD players, CD discs and mobile phones (Shepherd et al., 2004). In terms of percentages, Cookridge, Horsforth, Headingley, and Seacroft are the wards that have seen the biggest decreases (Figure 5.17b). The most considerable rise appears to be in Wortley (up 47%). 5.4.13 Theft of Motor Vehicle

% C h an g e -4 4.93 to -30 -2 9.99 to -15 -1 4.99 to 0 0.01 to 10 10.0 1 to 20 20.0 1 to 35 .4

Th eft o f M o to r V eh icle 3.02 - 5 .9 7 5.97 - 9 .5 6 9.56 - 1 3.67 13.6 7 - 18.4 8 18.4 8 - 29.7 4

a) Theft of motor vehicle rate (2003/04) per 1,000 population

b) Change, 2002/03 to 2003/04

Figure 5.18: Theft of motor vehicle City and Holbeck, University, and Richmond Hill wards have the highest rates per 1,000 population of theft of motor vehicle (Figure 5.18a). However the rate has decreased by 32.19% (Figure 5.18b) in the City and Holbeck ward between 200/03 and 2003/04.


87

5.4.14 Violent Crime

% Change 16.07 22.28 37.23 47.99 62.99

Violent C rim e 7.1 - 11.34 11.34 - 16.75 16.75 - 31.65 31.65 - 45.39 45.39 - 129.96

a) Violent crime rate (2003/04) per 1,000 population

to to to to to

2 2.28 3 7.23 4 7.99 6 2.99 9 1.78

b) Change, 2002/03 to 2003/04

Figure 5.19: Violent crime The highest rate of violent crime is in the City and Holbeck ward with the rate 129.96 per 1,000 population, followed by Burmantofts (45.39 offences per 1,000 population) and University (43.62 offences per 1,000 population) (Figure 5.19a). It is interesting to note that violent crime has increased between 16% and 92% in every ward (Figure 5.19b). Although City and Holbeck ward has the highest rate, it has seen the smallest increase. The highest increase is found in Burmantofts (up 91.78%).

5.5 Known Offenders and Victims Characteristics 5.5.1

Known Offenders Characteristics

In Leeds, offenders aged between 16 and 44 make up almost 85% of recorded crimes (where the age of the offender is known). Burglary dwelling, burglary elsewhere, criminal damage, drug offences, handling, homicide, other crime, robbery, theft of motor vehicle were mainly committed by young adults or people aged between 16 and 24, while fraud and forgery, other theft, sexual offences, theft from motor vehicle, and violent crime were mainly committed by people aged 25-44 (Table 5.9).

The age structure of offenders’ committing sexual offences is different from most other crime types, with a higher proportion of people from the older age groups. As can be seen 21.5% of offenders committing sexual offences are in the age range 45-64 (compared to only 4.9% committed by this age group in total). Criminal damage has a much higher proportion of young people aged less than 16 compared with other crime types.


88

Table 5.9: Age group of known offenders (2000-2004) Age Group of Offender

Crime Type 0-15 Burglary Dwelling Burglary Elsewhere Criminal Damage Drugs Offences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime Total Known Offender

16-24

25-44

45-64

65-74

75+

10.1% 12.4% 23.5% 4.0% 1.7% 7.2% 0.0% 5.4% 10.2% 19.3% 9.7% 7.3% 11.7% 10.2%

51.9% 53.7% 40.1% 52.3% 37.8% 46.4% 47.8% 48.4% 36.1% 56.8% 20.1% 40.1% 66.3% 35.3%

36.7% 33.1% 31.8% 41.1% 55.2% 42.1% 39.1% 40.6% 48.0% 23.7% 42.0% 52.4% 21.3% 45.8%

1.4% 0.8% 4.4% 2.7% 5.0% 4.0% 8.7% 5.3% 5.3% 0.2% 21.5% 0.2% 0.6% 8.0%

0.0% 0.0% 0.2% 0.0% 0.2% 0.3% 4.3% 0.4% 0.3% 0.0% 4.8% 0.0% 0.0% 0.5%

0.0% 0.0% 0.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.1% 0.0% 1.9% 0.0% 0.0% 0.2%

10.8%

42.4%

41.6%

4.9%

0.3%

0.1%

Source: Derived from West Yorkshire Police Note: Bold figures are the highest values for each crime type and highlight figures (in gray) are the highest values for each age group. Table 5.10: Gender of known offenders (2000-2004) Gender of Offender Crime Type FEMALE Burglary Dwelling Burglary Elsewhere Criminal Damage Drugs Offences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual Offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime Total known offender

Ratio Male: Female

MALE

7.36% 4.29% 14.03% 13.44% 28.32% 21.73% 8.70% 9.52% 30.38% 12.76% 2.02% 3.70% 6.75% 18.19%

92.64% 95.71% 85.97% 86.56% 71.68% 78.27% 91.30% 90.48% 69.62% 87.24% 97.98% 96.30% 93.25% 81.81%

12.6 22.3 6.1 6.4 2.5 3.6 10.5 9.5 2.3 6.8 48.5 26.1 13.8 4.5

17.44%

82.56%

4.7


Females are generally less likely than males to commit crimes. Offenders in Leeds are predominantly male for every crime type (Table 5.10). Overall, the ratio of male offenders to female offenders is 4.7 to 1. The ratio is the highest for sexual offences (48.5 to 1) and is the lowest for other theft (2.3 to 1). Females are most likely to commit other theft with a higher proportion (30.38%) compared to other crime types.


89

DARK EUROPEAN

ORIENTAL

WHITE EUROPEAN

UNKNOWN

Total

ASIAN

Burglary Dwelling Burglary Elsewhere Criminal Damage Drugs Offences Fraud & Forgery Handling Homicide Other Crime Other Theft Robbery Sexual offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime

ARAB

Crime Type

AFRO-CARIBBEAN

Table 5.11: Ethnicity of known offenders (2000-2004)

8.7% 2.8% 6.0% 15.6% 5.8% 7.5% 15.2% 10.8% 5.7% 21.5% 7.4% 2.4% 4.8% 10.2%

0.0% 0.0% 0.3% 0.2% 0.4% 0.0% 0.0% 0.6% 0.4% 0.1% 2.6% 0.1% 0.2% 0.5%

2.2% 1.2% 3.1% 7.6% 10.1% 6.0% 2.2% 7.0% 3.6% 5.7% 8.9% 1.0% 2.9% 5.7%

0.8% 0.6% 0.9% 1.0% 0.9% 0.7% 0.0% 1.1% 1.4% 0.8% 1.3% 1.4% 0.5% 1.0%

0.0% 0.0% 0.1% 0.2% 0.2% 0.1% 0.0% 0.2% 0.3% 0.1% 0.2% 0.1% 0.1% 0.1%

87.7% 95.1% 89.0% 75.1% 82.1% 85.6% 80.4% 80.0% 88.2% 71.4% 78.6% 94.9% 91.3% 81.7%

0.5% 0.4% 0.7% 0.3% 0.5% 0.1% 2.2% 0.3% 0.4% 0.3% 1.0% 0.2% 0.2% 0.8%

8.53%

0.35%

4.72%

1.01%

0.15%

84.7%

0.5%

Number Of Known Offender

5,478 2,714 6,986 5,695 3,095 949 46 2,986 15,217 2,405 896 1,734 3,572 18,861 70,645

Source: Derived from West Yorkshire Police. Note: Overrepresentation in bold.

Table 5.11 gives details of the ethnic breakdown of known offenders. Offenders in Leeds are predominantly ‘white’ (84.7% of total known offenders). This is not surprisingly as 91.85% of Leeds inhabitants are ‘white’. The ethnic breakdown shows that ‘Afro-Caribbean’ (‘black’) people are over-represented in homicide (15.2%), although the reliability of this statistic is affected by a relatively small sample size (n= 46). ‘Black’ people are also overrepresented in robbery to a considerable degree, making up 21.5% of the offence robbery but only 1.4% of the Leeds population as a whole. ‘Asian’ people account for 4.51% of the Leeds population as a whole but they are over-represented making up 10.1% for fraud and forgery, and 8.9% for sexual offences.


90

Table 5.12: Age groups of known victims (2003/04) a) Age Groups

Crime Type 0-15 Burglary Dwelling Burglary Elsewhere Criminal Damage Other Theft Robbery Sexual Offences Theft From Motor Vehicle Theft Of Motor Vehicle Violent Crime

0.2% 0.2% 0.2% 3.3% 18.3% 37.3% 0.1% 0.2% 12.9%

Total Crime

19.7% 3.6% 11.4% 21.4% 36.1% 28.2% 18.9% 17.9% 24.6%

3.0%

Leeds

16.6%

20%

0-15

Crime Type Male Burglary Dwelling Burglary Elsewhere Criminal Damage Other Theft Robbery Sexual Offence Theft from motor vehicle Theft of motor vehicle Violent Crime

16-24

0.1% 0.2% 0.2% 6.1% 22.3% 54.2% 0.0% 0.2% 17.3%

16-24

Female 0.1% 0.1% 0.2% 3.1% 11.3% 38.4% 0.1% 0.0% 11.7%

Male 19.8% 5.0% 12.7% 26.9% 37.6% 19.3% 20.3% 19.0% 28.6%

13.6%

25-44

45-64

41.6% 32.7% 38.5% 28.8% 26.3% 21.8% 45.8% 47.9% 37.3% 34.8% 29.1%

22.2% 24.1% 20.5% 14.5% 11.3% 4.8% 21.7% 22.3% 10.1% 17.1% 22%

65-74 5.2% 4.4% 4.2% 3.7% 2.5% 0.1% 2.6% 2.4% 1.2% 3.2% 8.1%

b) Distribution of Victims by Age and Gender 25-44 45-64

Female 21.5% 5.8% 16.9% 30.3% 36.2% 31.9% 22.4% 20.9% 28.4%

Male 44.7% 46.4% 47.1% 39.4% 27.9% 22.9% 50.1% 51.0% 39.3%

Female 42.3% 51.7% 52.8% 38.3% 25.0% 23.6% 52.2% 55.1% 47.3%

Male 25.4% 38.3% 31.0% 21.0% 9.7% 3.6% 25.2% 25.4% 12.8%

Female 21.0% 31.9% 21.9% 18.5% 15.7% 5.4% 22.6% 21.8% 10.6%

75+

Unknown

6.7% 2.4% 2.4% 2.6% 2.8% 0.4% 0.7% 1.1% 0.5%

4.3% 32.6% 22.8% 25.8% 2.8% 7.4% 10.2% 8.3% 13.3%

2.3%

23.0%

7.2%

65-74 Male 5.5% 6.9% 5.9% 4.5% 1.6% 0.0% 3.3% 3.1% 1.5%

75+

Female 5.5% 6.2% 4.9% 5.3% 4.5% 0.2% 2.3% 1.5% 1.3%

Male 4.5% 3.2% 3.0% 2.2% 0.8% 0.0% 1.0% 1.4% 0.4%

Female 9.6% 4.3% 3.3% 4.5% 7.3% 0.5% 0.4% 0.6% 0.7%

Note: proportions shown in Table 5.12b are percentage of each gender with different age groups. Bold numbers show the highest proportion of each gender across the age categories. Source: Derived from West Yorkshire Police


5.5.2

91

Known Victims Characteristics

Table 5.12a shows the age groups of the victims of different crimes types. Table 5.12b gives more detail on the gender of victims by age group for each crime type and for Leeds as a whole. The percentage shown in the table is the proportion of each gender across the age categories while Table 5.13 shows overall rates and ratios for each crime type.

In 2003/04, where the age of the victim (of all crime) was known, 34.8% were people aged between 25 and 44. However, victim’s age varies by crime type. Between 25 and 44 was the most common age of victims of burglary dwelling (41.6%), burglary elsewhere (32.7), criminal damage (38.5%), other theft (28.8%), theft from motor vehicle (45.8%), theft of motor vehicle (47.9%), and violent crime (37.3%). Victims younger than 25 are most likely to be victims of robbery and sexual offences. As can be seen from Table 5.12a, 36.1% of victims of robbery were people aged between 16 and 24 while 37.3% of victims of sexual offences were people younger than 16 years old. These were over-represented relative to the overall age structure in Leeds. More than half of male victims of sexual offences are younger than 16 (Table 5.12b).

Table 5.13: Gender of known victims (2003/04)

Crime Type

Gender of Victim MALE

FEMALE

Ratio Male: Female

Burglary Dwelling

49.7%

50.3%

1.0

Burglary Elsewhere

60.5%

39.5%

1.5

Criminal Damage

51.8%

48.2%

1.1

Other Theft

43.3%

56.7%

0.8

Robbery

68.5%

31.5%

2.2

Sexual offences

11.4%

88.6%

0.1

Theft From Motor Vehicle

62.4%

37.6%

1.7

Theft Of Motor Vehicle

69.3%

30.7%

2.3

Violent Crime Total

52.4%

47.6%

1.1

53.2%

46.8%

1.1

Source: Derived from West Yorkshire Police. The highest proportion in each crime type is in bold.

Table 5.13 shows the gender balance of known victims together with the ratio of males to females. The overall ratio of male to female victims is not much different. The ratio of male to female victims of burglary dwelling is 1 to 1. However, this means nothing if the house is occupied by a couple or more than one person. Males are more likely than females to be victims of burglary elsewhere, criminal damage, robbery, vehicle crimes and violent crime. As can be expected, females suffer from sexual offences far more than males.


92

Table 5.14: Victimisation by major crime types and ethnic group in Leeds (period 2001/02, 2002/03, and 2003/04) a) Victims of crime 2003/04 Crime Type Burglary Dwelling Burglary Elsewhre Criminal Damage Other Theft Robbery Sexual Offence Theft from a vehicle Theft of motor vehicle Violent Crime

Ethnicity Missing 19.0% 32.0% 32.6% 53.7% 15.5% 18.7% 38.3% 38.5% 18.5%

Unknown White European 1.5% 72.4% 6.5% 57.7% 6.1% 54.4% 5.0% 37.3% 0.6% 68.0% 1.1% 73.1% 9.2% 48.8% 10.0% 45.6% 1.5% 68.7%

Crime Type Burglary Dwelling Burglary Elsewhre Criminal Damage Other Theft Robbery Sexual Offence Theft from a vehicle Theft of motor vehicle Violent Crime



Crime Type Burglary Dwelling Burglary Elsewhre Criminal Damage Other Theft Robbery Sexual Offence Theft from a vehicle Theft of motor vehicle Violent Crime



Afro-Caribean 2.1% 0.6% 1.7% 1.0% 3.0% 3.3% 0.9% 1.5% 3.8%

Arab 0.2% 0.2% 0.3% 0.2% 0.9% 0.3% 0.1% 0.5% 0.6%

Asian Dark European 3.2% 0.4% 2.7% 0.2% 4.5% 0.2% 2.2% 0.2% 10.0% 0.8% 2.3% 0.7% 2.2% 0.1% 3.3% 0.3% 6.1% 0.5%

Oriental 1.2% 0.3% 0.3% 0.4% 1.2% 0.5% 0.3% 0.3% 0.3%

Arab 0.2% 0.1% 0.2% 0.1% 0.7% 0.2% 0.0% 0.2% 0.5%


Oriental 0.6% 0.4% 0.3% 0.5% 2.0% 0.2% 0.2% 0.1% 0.3%

Arab 0.3% 0.1% 0.1% 0.1% 0.8% 0.0% 0.1% 0.1% 0.3%


Oriental 0.9% 0.3% 0.3% 0.8% 1.8% 0.6% 0.3% 0.2% 0.2%

b) Victims of crime 2002/03 Afro-Caribean 2.2% 0.5% 1.4% 0.8% 2.1% 3.0% 0.5% 0.8% 3.9%

c) Victims of crime 2001/02


Afro-Caribean 2.3% 0.7% 1.4% 1.0% 1.8% 3.3% 0.6% 0.7% 3.8%


93

Table 5.14 shows victims’ ethnic status. Note that records where no victim is identifiable have not been included in the calculation. It is not surprising that victims’ ethnicity for all crimes is majority ‘white’. Nevertheless, when comparing each group, it has been found that ‘Asian’ suffered from victimisation relatively more than other groups.

5.6 The Relationship between Crime and its Related Determinants The x y coordinates of existing data provide a better understanding, not only of geographical variations in certain types of offending, but also of the relationship between crime and other variables from the census data. The association of demographic and socio-economic characteristics which can be correlated with the locations of crimes provide a clearer picture of crime. Patterns of crime vary considerably across the region depending upon a range of different factors.

In line with Table 2.2 shown in Chapter 2, the Pearson’s correlation coefficient is used in this section to investigate the possible relationships between demographic, socio-economic, and area characteristics and crime. The unit of analysis is the ward. The correlation coefficient always takes a value between -1 and 1 indicating the strength and direction of a linear relationship between two variables. A value of 1 or -1 indicates perfect correlation. A positive correlation indicates a positive relationship between the variables (increasing values in one variable correspond to increasing values in the other variable), while a negative correlation indicates a negative relationship between variables (increasing values in one variable correspond to decreasing values in the other variable). A correlation value close to 0 indicates no relationship between the variables: thus the higher the absolute correlation coefficient the better in terms of finding relationships. The Statistical Package for the Social Sciences (SPSS) was used for the correlation analysis. Table 5.15 shows the correlation coefficient of selected demographic and socio-economic variables for overall crime and each crime type. Note that handling, homicide, and sexual offences are not shown in the table due to the small number of offences.

The analysis started with Pearson’s correlation coefficient calculated for (overall) crime in Leeds for each crime type. Note that numbers of crimes used in this section are from the period 2001/02 which is the same period as the 2001 Census. The dataset can be found in Appendix C. The correlation coefficients (Pearson’s r) for each variable for each crime type are shown in Table 5.15. Some variables are more closely correlated with some types of crime than others. These correlation coefficients will be used to describe crime and its related determinants in § 5.6.1 to 5.6.8.


94

Household density

.228 .786**

-.038 .462**

Population density

.174 .766**

-.078 .393*

Percentage of student

.234 .826**

-.065

.236

.338

.152

.274 .204

Number of car per household

-.513**

-.506**

-.329 -.841**

-.557**

-.435*

Number of rented tenure

.586**

.723**

.316 .782**

.564**

.416*

.08

.081 .449**

.092

.04

.042 .378*

.118

.102

.057

.121 .492**

.276 .424*

-.371*

-.369*

.420*

.434*

.17 .355*

-.562**

-.401*

.729**

.543**

Number of young adult

.206 .768**

-.06

.198

.153

.072

.054

.115 .409*

Number of male young adult

.196 .758**

-.065

.178

.136

.071

.038

.11 .394*

Violent Crime



Robbery

Other Theft

Other Crime

Fraud and Forgery

Drug Offences

Criminal Damage

Burglary Elsewhere

Burglary Dwelling

Determinants of Crime

All Crime_2001/02

Table 5.15: Correlation coefficient of crime and its related determinants

.207 .276

.156 .117

-.650**

-.539**

.770**

.545**

.246 .359* .238

.098 .339

.084

Number of unemployed people

.642**

.442*

.489**

.897**

.688**

.491**

.548**

.519**

.679**

.536**

.709**

.688**

Number of male unemployed

.665**

.468**

.502**

.904**

.718**

.510**

.569**

.540**

.709**

.563**

.722**

.709**

Number of unemployed male young adult

.545**

.471**

.406*

.827**

.569**

.379*

.451**

.423*

.573**

.433*

.623**

.597**

Index of Multiple Deprivation

.480**

.32 .345*

.825**

.550**

.383*

.408*

.361*

.485**

.346*

.557**

.547**

Number of offender living in area

.658**

.915**

.730**

.510**

.600**

.551**

.636**

.537**

.673**

.726**

Note:

.411*

.500**

** Correlation is significant at the 0.01 level (2-tailed) * Correlation is significant at the 0.05 level (2-tailed) Demographic and socio-economic variables are derived from the 2001Census (See Appendix C). Number of offender is derived from the West Yorkshire Police. Significant coefficients are marked in bold and the highest are marked in colour.


5.6.1

95

Population/Household Density

Generally, crime statistics show a positive correlation between ‘population density’ and crime rates (Entorf and Spengler, 2000). However, in Leeds, ‘population density’ has a low correlation with overall crime but has a very high correlation with burglary dwelling (.766). This means an area with a high ‘population density’ tends to have higher burglary dwelling. ‘Household density’ also has a high correlation with burglary dwelling (.786), criminal damage (.462), and robbery (.449). Figure 5.20 and Figure 5.21 show that high ‘population density’ and high ‘household density’ areas such as Headingley, University and Harehills have higher burglary dwelling than low density areas. Density has two roles in causing crime: it increases the potential victims and it reduces the chance of being caught (Glaeser and Sacerdote, 1999; Kelly, 2000). In the lower ‘population density’ areas the number of potential contacts between offenders and victims (attractive targets) is low. However, this pattern is not uniform. For example, in the City and Holbeck and Roundhay wards this pattern does not hold. These two wards have much lower densities compared with Headingley but quite high rates for burglary dwelling offences.

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 Pop u la tio n D en s ity 28 7.97 - 8 23 .0 1 82 3.01 - 1 68 5.24 16 85 .2 4 - 29 26 .6 9 29 26 .6 9 - 44 04 .9 5 44 04 .9 5 - 94 34 .5 5

Figure 5.20: ‘Population density’ and burglary dwelling

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 H ou s eh old D en s ity 11 8 - 3 3 1 33 2 - 6 2 6 62 7 - 1 0 18 10 19 - 2 23 8 22 39 - 3 48 5

Figure 5.21: ‘Household density’ and burglary dwelling


5.6.2

96

Demographic Characteristics

From the previous studies described in Chapter 2, it can be seen that the ‘young adult’ group (especially males) are more prone to commit crimes and to be the victims of crimes than other age groups. Young people (aged 16 to 24 years old) are around three times more likely to be victims of burglary than people in other age groups (Home Office, 2006c). This also can be seen in Leeds. Although, ‘(male) young adult’ has a very low correlation with overall crime, it has a high correlation with burglary dwelling. Areas with a high number of ‘(male) young adult’ tend to have high levels of burglary dwelling. Figure 5.22 shows the distribution of ‘young adults’ (people who are 16-24 years) in Leeds while Figure 5.23 shows only males of this age group. As can be seen, Headingley has the highest number of ‘(male) young adults’ followed by University, and Kirkstall. These wards also have high numbers of burglary dwelling offences. This would suggest that areas with high concentrations of ‘(male) young adults’ would experience higher burglary dwelling than areas where this group does not make a significant contribution to the total population.

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 You n g A d ult 17 70 - 2 06 0 20 61 - 2 51 2 25 13 - 3 19 9 32 00 - 7 67 1 76 72 - 1 57 32

Figure 5.22: Distribution of ‘young adult’ and burglary dwelling

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 N um b er of M a le Yo u ng A du lt 81 3 - 1 0 22 10 23 - 1 49 0 14 91 - 2 29 9 23 00 - 3 38 9 33 90 - 7 43 1

Figure 5.23: Distribution of ‘male young adult’ and burglary dwelling


5.6.3

97

Percentage of Students

It has been found that ‘percentage of students’ living in an area has a very high correlation with burglary dwelling, with a correlation coefficient of .826. This is higher than any other variables correlated to this crime type. ‘Percentage of students’ also has a significant correlation with robbery (.492).

Figure 5.24 shows the distribution of students in Leeds. The highest percentage is found in Headingley, followed by University. As can be seen, areas with a high proportion of students also have high burglary dwelling. Generally, students are less likely to be at home during daytime. Therefore, these areas are more likely to be targets for burglary dwelling.

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 P e rce n tag e of s tud e nt 4.1 5 - 5.68 5.6 8 - 8.09 8.0 9 - 10 .3 7 10 .3 7 - 2 2 .89 22 .8 9 - 6 0 .42

Figure 5.24: ‘Percentage of students’ and burglary dwelling

Statistically, students are one of the most likely groups to fall victim to crime. One third of students become the victim of a crime each year (Home Office, 2006c). However, surprisingly, the majority of studies on ‘modelling crime’ undertaken to date have not included this student factor despite the very high correlation with burglary dwelling. Traditionally, such models are more likely to include variables such as unemployment, inequality, income, deprivation, poverty, and tenure type.


5.6.4

98

Rented Tenure Type

In 2001 within Leeds, 62% of households lived in owner-occupied accommodation, whilst 38% lived in rented housing. ‘Rented tenure’ is concentrated in inner city areas which have a high correspondence with burglary dwelling (Figure 5.25). There are significant correlations between ‘rented tenure’ and every type of crime except burglary elsewhere (Table 5.15). The highest correlations are with criminal damage (.782), theft of motor vehicle (.77), robbery (.729), and burglary dwelling (.723). Housing type and crimes are related, possibly because social groups with a greater propensity to commit crime and to be a victim of crime are concentrated in certain types of housing. As seen in Figure 5.25, University, Headingley and City and Holbeck have high numbers of ‘rented tenure’ housing. Therefore, burglary dwelling is higher in these areas, especially in University and Headingley, which are also the most popular areas for students.

R ec o rde d B u rglary D w e lling 2 00 1 /0 2 R ec o rde d bu rglary _0 1 02 R en te d Te nu re 10 28 - 1 93 9 19 40 - 2 99 9 30 00 - 4 11 3 41 14 - 5 34 7 53 48 - 9 07 8

Figure 5.25: ‘Rented tenure’ and burglary dwelling

5.6.5

Number of Cars per Household

‘Number of cars per household’ can be seen as one indicator of affluence. As can be seen from Figure 5.26, the outer parts of Leeds are more affluent than inner parts in terms of ‘number of cars per household’. According to Table 5.15, there is a significant negative correlation between ‘number of cars per household’ and every type of crime except burglary elsewhere (negative correlation but not significant). ‘Number of cars per household’ has the highest (negative) correlation with criminal damage (-.841). Figure 5.26 shows ‘number of cars per household’ and crime and Figure 5.27 shows ‘number of cars per household’ and criminal damage. Areas that have higher ‘number of cars per household’ such as Barwick and Kippax, Wetherby, Garforth and Swillington tend to have lower crime. In contrast, areas that have lower ‘number of cars per household’ such as City and Holbeck, University, Headingley, Burmantofts, and Harehills tend to have higher crime, especially criminal damage.


99

Crim e s 2 00 1 /0 2 Allc rim e0 10 2 Nu m b er o f c ar p er ho u se ho ld 0.4 2 - 0.6 1 0.6 1 - 0.7 8 0.7 8 - 1 1 - 1.1 8 1.1 8 - 1.4 3

Figure 5.26: ‘Number of cars per household’ and crime

Crim in al da m a g e Nu m b er o f c ar p er ho u se ho ld 0.4 2 - 0.6 1 0.6 1 - 0.7 8 0.7 8 - 1 1 - 1.1 8 1.1 8 - 1.4 3

Figure 5.27: ‘Number of cars per household’ and criminal damage

5.6.6

Unemployment

While ‘unemployment’ is low for Leeds as a whole, there are some areas in Leeds that experience unemployment that is more than double the average for the city. In Leeds, the rate of economically active unemployed aged 16-74 was 3.32% in 2001 (3.35% in England and Wales). The inner city wards of City and Holbeck, Harehills, and University have the highest amount of economically active unemployment for those people aged 16-74; 989, 889, 881 respectively. ‘Unemployment rates’ are highest in City and Holbeck (6.65%), Harehills (6.32%), Seacroft (6.32%), and Burmantofts (6.06%) (Figure 5.28). It is interesting to note that the ‘number of unemployed’ has a significant correlation with every crime type (Table 5.15), and highest for criminal damage (.897) (Figure 5.29).


100

% of Unem ploym ent 1.7 - 2.07 2.07 - 2.51 2.51 - 2.85 2.85 - 4.69 4.69 - 6.65

Figure 5.28: Distribution of ‘unemployment rate’ in Leeds

Criminal damage Number of unemployed 303 - 374 375 - 423 424 - 564 565 - 775 776 - 989

Figure 5.29: ‘Unemployment’ and criminal damage

Criminal damage Male unemployed 187 - 234 235 - 277 278 - 424 425 - 507 508 - 692

Figure 5.30: ‘Male-unemployed’ and criminal damage

Male-specific unemployment is more likely than any other groups to generate significant (at 0.01 level) results for every category of crime especially criminal damage (.904) (Table 5.15). The correlation is almost linear with the R2 value at 0.8175 (Figure 5.31).


101

900 R2 = 0.8175 800

Male unemployed

700 600 500 400 300 200 100 0 0

500

1,000

1,500

2,000

2,500

Criminal Damage

Figure 5.31: Correlation between ‘male unemployed’ and criminal damage

5.6.7

Deprivation

The view that ‘deprivation’ has a crucial effect on crime is supported by a large number of studies (see Chapter 2). There is a close correlation between areas of high deprivation and those experiencing the worst levels of crime. Around 150,000 people in Leeds, almost 20% of the population, live in areas officially rated as amongst the most deprived in the country (Leeds Community Safety, 2004). In this section the Index of Multiple Deprivation (IMD) is used for analysis. In Leeds, areas with high deprivation tend to have high crime, especially criminal damage (Figure 5.32). However, the correlation between deprivation indices and crime is not perfect because not all deprived areas have high crime, and demographically very similar areas can have markedly different rates. For example, Headingley has relatively high crime but it does not have a very high degree of deprivation (Table 5.16).

Crim in al d am ag e Ind ex of M ultip le Depr ivation 6.89 - 8.8 8.8 - 13.87 13.87 - 19.56 19.56 - 42.5 42.5 - 55.41

Figure 5.32: ‘Index of Multiple Deprivation’ and criminal damage


102

Table 5.16: Indices of deprivation for Leeds wards 2000 (ranking from high to low)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Ward

Index of Multiple Deprivation (IMD)

City and Holbeck Seacroft Harehills Burmantofts Richmond Hill Hunslet University Chapel Allerton Beeston Middleton Bramley Armley Wortley Whinmoor Kirkstall Pudsey South Morley South Barwick and Kippax Rothwell Moortown Headingley Weetwood Morley North Garforth and Swillington Cookridge Roundhay Pudsey North North Halton Aireborough Otley and Wharfedale Wetherby Horsforth

55.41 55.07 54.07 53.66 52.52 47.97 47.76 42.50 40.73 37.04 35.06 33.20 31.61 30.28 29.22 19.56 18.65 16.99 16.87 16.55 16.17 15.71 15.04 13.87 12.07 11.99 11.98 11.66 10.65 10.15 8.80 6.91 6.89

Source: ‘Indices of Deprivation for wards, 2000’, Office for National Statistics (2003)


5.6.8

103

Number of Offenders

In addition to variables from the census, ‘number of offenders’ i.e. those detected aggregate up to ward level (derived from West Yorkshire Police) can also be tested for correlation with crime. ‘Number of offenders’ living in an area has significant correlation with every crime type and has a strong positive relationship with criminal damage (R=9.15; R2=.837)

Criminal damage Number of O ffender 450 - 687 688 - 1028 1029 - 2210 2211 - 4001 4002 - 5285

Figure 5.33: ‘Number of offender’ and criminal damage

7,000 R 2 = 0.837 6,000

Number of Offender

5,000

4,000

3,000

2,000

1,000

0 0

500

1,000

1,500

2,000

2,500

Criminal Damage

Figure 5.34: Correlation between ‘number of offenders’ and criminal damage

In summary, ‘household/population density’, ‘percentage of students’ living in an area, ‘number of (male) young adults’, have no effect on (overall) crime but a strong impact on burglary dwelling with a significant correlation above 0.7. By contrast, ‘deprivation’ has a significant effect on criminal damage and overall crime but little or no significance on burglary dwelling.


5.6.9

104

Multiple Regression Model

A multiple regression was also employed to explore the relationship between crime and its related determinants. It can be used to describe the relationship between multiple variables precisely by means of an equation that has predictive value. The multiple regression modelling in this study was carried out using ‘stepwise variable selection’ which is a method of choosing the best predictors of a particular dependent variable on the basis of statistical criteria. Fundamentally, the statistical procedure decides which independent variable is the best predictor, the second best predictor, etc. It is a combination of forward and backward procedures. When each variable is added, variables which are entered in earlier steps are rechecked to see if they are still significant. If not, they will be removed.

As mentioned in Chapter 2 and discussed in the previous section, there is considerable evidence that demographic, socio-economic, household and neighbourhood characteristics are related to crime. Therefore recorded crime incidents for the period of April 2001- March 2002 and demographic and socio-economic characteristics of the people in those areas derived from the 2001 Census were analysed. Note that the model presented in this section is for burglary dwelling, the most important crime type in this study, chosen for its importance and because it is likely to represent the most predictable crimes-relating, as it does, to home locations for which census data is available. As with the correlation analysis in the previous section, SPSS was used for the statistical modelling.

The Model:

With p independent variables, the regression equation is

yˆ = a + b1 x1 + b2 x2 + ... + b p x p

Where ŷ

(5.1)

is the predicted value of the dependent variable.

b

is the regression coefficient

x

are independent variables

Number in sample = 33 (wards)

The number of burglary dwelling incidents used in this section is for the period 2001/02. For independent variables, all potential possible census variables are taken into account.


105

Variables entered as independent variables (details can be found in Appendix C):

Household density Population density Number of student Percentage of student Number of rented house Percentage of rented house Number of high-class people (a combination of large employers and higher managerial occupations, higher professional occupations and lower managerial and professional occupation) Percentage of high-class people Number of young adult people (aged 16-24) Number of male young adult (aged 16-24) Number of unemployed employed (aged 16-74) Percentage of unemployed people (aged 16-74) Number of male unemployed (aged 16-74) Number of male young adult unemployed (aged 16-74) Number of people who have no qualification (aged 16-74) Number of car in area Number of car per household Index of Multiple Deprivation Income deprivation Number of offender living in area (derived from West Yorkshire Police)

Table 5.17: Burglary dwelling model (model summary) Model

R

R Square

1 2 3 4 5 6 7 8

.826a .892b .948c .956d .965e .963f .969g .967h

.683 .795 .898 .914 .931 .928 .938 .934

a. b. c. d. e. f. g. h.

Adjusted R Square .673 .782 .887 .902 .918 .918 .927 .925

Std. Error of the Estimate 152.229 124.299 89.265 83.400 76.091 76.377 71.968 72.970

Predictors: (Constant), PerCentOfStudent Predictors: (Constant), PerCentOfStudent, PerCentOfUnEmp Predictors: (Constant), PerCentOfStudent, PerCentOfUnEmp, PerCentOfHighClass Predictors: (Constant), PerCentOfStudent, PerCentOfUnEmp, PerCentOfHighClass, MaleYoungAdult Predictors: (Constant), PerCentOfStudent, PerCentOfUnEmp, PerCentOfHighClass, MaleYoungAdult, AllCarInArea Predictors: (Constant), PerCentOfUnEmp, PerCentOfHighClass, MaleYoungAdult, AllCarInArea Predictors: (Constant), PerCentOfUnEmp, PerCentOfHighClass, MaleYoungAdult, AllCarInArea, NumberOfMaleYoungAdultUnEmployed Predictors: (Constant), PerCentOfHighClass, MaleYoungAdult, AllCarInArea, NumberOfMaleYoungAdultUnEmployed


106

The model summary (Table 5.17) can be used to assess the relative importance of each variable. The goodness-of-fit statistics displayed in the table are R, R2, adjusted R2, and the standard error of the estimate. The R (Pearson’s product-moment correlation coefficient), R2 (coefficient of determination), and adjusted R2 statistics can take on any value less than or equal to 1, with the values closer to 1 indicating a better fit. The key column ‘Adjusted R2’ can be used to assess which model is the best. The figure .673 means that 67.3% of the variation in burglary dwelling is explained by variations in the independent variables of Model 1. Correspondingly, 78.2%, 88.7%, 90.2%, 91.8%, 91.8%, 92.7%, and 92.5% is explained by variation in the independent variables of Models 2 to 8 respectively. Therefore, Model 7 is the best for explaining burglary dwelling in Leeds. The equation with the smallest Standard Error of the Estimate will most likely also have the highest Adjusted R2.

Table 5.18: Coefficients of the models Coefficientsa

Model 1 2

3

4

5

6

7

8

(Constant) PerCenOfFTStudent (Constant) PerCenOfFTStudent PerCentOfUnEmp (Constant) PerCenOfFTStudent PerCentOfUnEmp PerCenOfHighClass (Constant) PerCenOfFTStudent PerCentOfUnEmp PerCenOfHighClass MaleYoungAdult (Constant) PerCenOfFTStudent PerCentOfUnEmp PerCenOfHighClass MaleYoungAdult AllCarInArea (Constant) PerCentOfUnEmp PerCenOfHighClass MaleYoungAdult AllCarInArea (Constant) PerCentOfUnEmp PerCenOfHighClass MaleYoungAdult AllCarInArea NumberOfMaleYoung AdultUnEmployed (Constant) PerCenOfHighClass MaleYoungAdult AllCarInArea NumberOfMaleYoung AdultUnEmployed

Unstandardized Coefficients B Std. Error 292.501 34.700 19.080 2.335 94.403 56.406 19.243 1.907 56.258 13.851 -590.776 133.175 21.907 1.456 127.211 16.478 16.403 3.037 -795.578 153.341 11.531 4.740 146.925 17.648 19.622 3.168 .111 .049 -477.176 186.671 5.426 4.931 107.709 22.157 22.950 3.166 .156 .048 -.031 .012 -474.196 187.353 105.839 22.175 24.579 2.809 .207 .014 -.038 .011 -395.981 180.315 46.523 34.816 26.270 2.764 .186 .016 -.048 .011

Standardized Coefficients Beta

.277 .867 .828 -.575

t 8.429 8.171 1.674 10.090 4.062 -4.436 15.049 7.720 5.401 -5.188 2.433 8.325 6.194 2.285 -2.556 1.100 4.861 7.249 3.281 -2.576 -2.531 4.773 8.749 15.118 -3.527 -2.196 1.336 9.506 11.418 -4.296

Sig. .000 .000 .105 .000 .000 .000 .000 .000 .000 .000 .022 .000 .000 .030 .017 .281 .000 .000 .003 .016 .017 .000 .000 .000 .001 .037 .193 .000 .000 .000

.826 .833 .335 .949 .759 .541 .499 .876 .647 .495 .235 .642 .757 .698 -.374 .631 .811 .922 -.451

2.174

1.021

.313

2.130

.042

-219.937 26.364 .170 -.059

124.829 2.801 .011 .008

.870 .756 -.708

-1.762 9.412 15.216 -7.875

.089 .000 .000 .000

3.265

.621

.470

5.257

.000

a. Dependent Variable: BurglaryDwelling


107

The multiple regression model for burglary dwelling was carried out and is statistically significant as detailed in Table 5.17. According to the coefficients (Table 5.18) the best model for burglary dwelling is given by Burglary Dwelling = -395.981+46.523* %of UnEmp+26.27*%of HighClass +.186* MaleYoungAdult -.048*AllCar+2.174*MaleYoungAdultUnEmp

(5.2)

Age, sex, economic activity, socio-economic classification, and car ownership play important roles and are important predictors for burglary dwelling. Therefore, these variables will be included in the spatial microsimulation model for crime discussed in the next chapter.

5.7 Concluding Comments This chapter has presented the geography and determinants of crime in Leeds. It is clear that most of the crime types tend to be clustered in certain areas in the proximity of the city centre, while wards with the lowest rates are predominantly in outer areas. The City and Holbeck ward has the highest number of crimes and crime rates in every type except burglary dwelling. However, different types of crime tend to occur in different types of areas. For example, vehicle crimes occur not only in the inner part of the city. It should be noted that an apparent rise of crime in 2002/03 was mainly due to a change in the way crimes were recorded (National Crime Recording Standards). Known offenders’ and victims’ characteristics (age, sex, and ethnicity) were also described. Generally, most of the known offenders are European white males aged between 16 and 44. Victims’ age varies by crime type. However, one third were people aged between 25 and 44. The ethnicity of victims is predominantly ‘white’, reflecting the population of Leeds as a whole. Correlation coefficients provide a better understanding about the relationship between crime and its related determinants. It can be argued that the student factor is one of the most important for modelling crime. Finally, the multiple regression model provided a picture of which variables need to be included in the spatial microsimulation model. Correlation coefficients and the multiple regression model suggested that the tenure type, age, sex, economic activity, socio-economic classification, and car ownership variables play an important role in predicting burglary dwelling. Although, ‘tenure type’ did not appear to be important in the multiple regression model, it has very high correlation with burglary dwelling in single-variable regressions, and, more importantly, it is a useful variable for manipulating social structure in the final model. It is, therefore included, along with the other variables, in the spatial microsimulation model that will be described in the next chapter.

Chapter 6- SimCrime: A Spatial Microsimulation for Crime in Leeds

108

Chapter 6 SimCrime: A Spatial Microsimulation Model for Crime in Leeds 6.1 Introduction 6.2 SimCrime Model Specification 6.2.1 Input 6.2.2 Input Adjustment 6.2.3 Model Execution Process 6.2.4 Model Output 6.3 Evaluation of Synthetic Microdata 6.4 Concluding Comments

6.1

Introduction

SimCrime is a spatial microsimulation model that is designed to estimate the likelihood of being a victim of crime and crime rates at the small area level in Leeds and to answer whatif questions about the effects of changes in the demographic and socio-economic characteristics of the future population. The model is based on individual microdata. Specifically, SimCrime combines individual microdata from the British Crime Survey (BCS) for which location data is only at the scale of large areas, with census statistics for smaller areas to create synthetic microdata estimates for output areas (OAs) in Leeds using a simulated annealing method. The new microdata dataset includes all the attributes from the original datasets. This allows variables such as crime victimisation from the BCS to be directly estimated for OAs. Section 6.2 describes the SimCrime model specification, associated processes, and the creation of 514,523 individuals aged 16-74 in households in Leeds whose characteristics match as closely as possible the characteristics of the 514,523 actual individuals living in Leeds, as shown in the 2001 Census. The result is an individuallevel dataset constrained by the census statistics. The synthetic microdata is evaluated in § 6.3. The final section gives some concluding comments.

6.2

SimCrime Model Specification

As with most microsimulation models the first step is to generate population microdata, which comprises of a list of individuals along with an associated set of individual characteristics (Williamson et al., 1998; Williamson, 2002). The chief task in microsimulation is to select individuals from a microdata dataset to fill small census areas. Usually this procedure begins by using random individuals initially, and then swaps out poor


109

(badly fitting) individuals for others to improve the match with the census statistics for the area in question. Previous studies (Williamson et al., 1998; Ballas, 2001) have shown that the simulated annealing technique works effectively in terms of finding the combination of records which best fits known small area statistical constraints (more detail appears in § 3.6). Therefore, in this study, combinatorial optimisation is achieved by using simulated annealing.

The synthetic population microdata dataset was generated at the census output area for Leeds with the use of a Simulated Annealing-Based Reweighting Program 1. The latter was implemented in Java, an object-oriented programming language, which has been accepted as the most suitable type of programming language for spatial microsimulation modelling (Ballas, 2001). It can be operated on any computer system and platform without amending any code (i.e. it is platform independent). The program implements a combinatorial optimisation using simulated annealing approach to generate spatially disaggregated population microdata dataset at the small area level. Specifically, here, the implementation of the microsimulation approach for Leeds involves selecting the combination of individuals from the microdata (the 2001/2002 BCS) which best fits the known constraints in the selected small areas of the 2001 UK Census.

More specifically the 514,523 people aged 16-74 living in households found in Leeds in the 2001 Census were recreated. The procedure involves taking records of individuals from the 2001/2002 BCS, and redistributing them (multiple times) in areas until the aggregate statistics for each area match those found in the census. The end result is an individual-level dataset constrained by the census statistics. To recap, an individual-level estimation is necessary as the individual-level census data is not available because of confidentiality restrictions.

6.2.1

Input There are four important files needed to run the program: 1) Model File 2) Microdata File 3) Constraint Table Files 4) ‘Group Number’ File (number of people in small areas)

1

The program was first developed by Dr. Dimitris Ballas in 1999. It has been maintained and further developed at the Centre for Computational Geography (CCG), School of Geography, University of Leeds. Thanks to Mr. Jianhui Jin for providing the program to generate the base micro population and source code for modifications.


110

1) Model File Model file is a text file containing the path to the constraint tables’ files, microdata file, ‘Group Number’ file, and filter definitions. The filter definitions use logic operations and conditions to define the fitting conditions for each column (for more detail see Appendix D).

2) Microdata from the 2001/2002 British Crime Survey As Huang and Williamson (2001) pointed out, the quality of the synthetic microdata is likely to be affected by the size of the sample used as a parent population. The larger the sample size, the more possible combinations of individuals exists and the better the fit is likely to be. The 2001/2002 BCS used as a microdata database in this study has 32,824 records. To make the variables from the BCS compatible with the census, the following variables in the BCS were checked to determine whether an individual fits each column in the constraint tables from the census or matches the classifications that exist in the census. sex:

Respondent Sex 1 = Male 2 = Female

age:

Respondent Age

marst: Respondent Marital status 1 = Single, that is, never married 2 = Married and living with husband/wife 3 = Married and separated from husband/wife 4 = Divorced 5 = Widowed 8 = Refused 9 = Don’t know remploy: Respondent employment status 1.0 = Employed 2.0 = Unemployed 3.0 = Inactive infstudy: Are you a full-time student at college or university 1 = Yes 2 = No 8 = Refused 9 = Don’t know


111

respsec2: Respondent National Statistics Socio-Economic Classification (NS-SEC) 1.10 = Large employers and higher managerial occupations 1.20 = Higher professional occupations 2.00 = Lower managerial and professional occupations 3.00 = Intermediate occupations 4.00 = Small employers and own account workers 5.00 = Lower supervisory and technical occupations 6.00 = Semi-routine occupations 7.00 = Routine occupations 8.00 = Never worked 9.00 = Not classified numcars: Number of cars tenharm: ONS Harmonised Tenure type 1 = Owners 2 = Social rented sector 3 = Private rented sector

3) Constraint Tables Generally, the more constraint variables used the better the synthetic microdata dataset produced. However, the more constraint variables are added in, the more comparisons with the real data will be required which means more time will have to be spent running the model. It has been noted that using a different set of constraints would generate different results (Huang and Williamson, 2001). The constraint variables in this study were chosen as they are potential predictors for crime analysis. Specifically, stepwise multiple regression was used to identify the best predictors (see the detail in previous chapter, § 5.6.9). Three cross-tabulation tables from the 2001 Census Area Statistics (CAS) were used to cover the seven constraining variables (Table 6.1) including table CS004: Age by Sex and Living Arrangements, table CS047: National Statistics- Socio Economic Classification (NS-Sec) by Tenure, and table CS061: Tenure and Car or Van Availability by Economic Activity (Table 6.2). All data is at the output area level. There are 2,439 output areas in Leeds.


Table 6.1: SimCrime constraint variables SimCrime Constraint Variables Age

Sex Living Arrangement

Economic Activity

Tenure Type Car or Van availability

Socio-economic Classification

Categories Aged 16-24 Aged 25-34 Aged 35-49 Aged 50-74 Male Female Couple Not couple Employed Unemployed Inactive Full-time Student Owned Rented No Car One Car Two or more car Higher Managerial and professional occupations Lower Managerial and professional occupations Intermediate occupations Small employers and own account workers Lower supervisory and technical occupations Semi-routine occupations Routine occupations Never worked and long-term unemployed Not classified

112


113

Table 6.2: SimCrime constraint tables

CS004: Age by Sex and Living Arrangements: All People in Households (16 categories) 1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 11) 12) 13) 14) 15) 16) 17)

(Zone Code) Male_16-24_couple Male_16-24_not couple Female_16-24_couple Female_16-24_not couple Male_25-34_couple Male_25-34_not couple Female_25-34_couple Female_25-34_not couple Male_35-49_couple Male_35-49_not couple Female_35-49_couple Female_35-49_not couple Male_50-74_couple Male_50-74_not couple Female_50-74_couple Female_50-74_not couple

CS047: NS_Sec by Tenure: All people in Households Aged 16-74 (18 categories) 1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 11) 12) 13) 14) 15) 16) 17) 18) 19)

(Zone Code) Higher Managerial and professional occupations_Owned Higher Managerial and professional occupations_Rented Lower Managerial and professional occupations_Owned Lower Managerial and professional occupations_Rented Intermediate occupations_Owned Intermediate occupations_Rented Small employers and own account workers_Owned Small employers and own account workers_Rented Lower supervisory and technical occupations_Owned Lower supervisory and technical occupations_Rented Semi-routine occupations_Owned Semi-routine occupations_Rented Routine occupations_Owned Routine occupations_Rented Never worked and long-term unemployed_Owned Never worked and long-term unemployed_Rented Not classified_Owned Not classified_Rented

CS061: Tenure and Car or Van Availability by Economic Activity: All People Aged 16 to 74 in Households (24 categories) 1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 11) 12) 13) 14) 15) 16) 17) 18) 19) 20) 21) 22) 23) 24) 25)

(Zone Code) Owned_NoCar_Employed Owned_NoCar_Unemployed Owned_NoCar_Inactive Owned_NoCar_FTStudent Owned_1Car_Employed Owned_1Car_Unemployed Owned_1Car_Inactive Owned_1Car_FTStudent Owned_2 or MoreCar_Employed Owned_2 or MoreCar_Unemployed Owned_2 or MoreCar_Inactive Owned_2 or MoreCar_FTStudent Rented_NoCar_Employed Rented_NoCar_Unemployed Rented_NoCar_Inactive Rented_NoCar_FTStudent Rented_1Car_Employed Rented_1Car_Unemployed Rented_1Car_Inactive Rented_1Car_FTStudent Rented_2 or MoreCar_Employed Rented_2 or MoreCar_Unemployed Rented_2 or MoreCar_Inactive Rented_2 or MoreCar_FT Student


114

4) Group Number (Number of people in small area) ‘Group Number’ is the number of people in each small area (expected count). To run the program we need to specify how many people we want to populate in each small area. This is according to the census counts. However, as mentioned in § 4.2.1 there are inconsistencies between the constraint tables produced by official disclosure control measures. The unfortunate result of this process is that there can be different numbers of people in the different tables for a given output area. The impact is that the simulated annealing process may not find the combination of individuals that would match every constraining table perfectly (Huang and Williamson, 2001). In some cases it would be unlikely to achieve an absolute error of zero and will always run until the iteration limit is matched (Ballas, 2001). This can produce a high error for the synthetic population (when compared with the real population) in some areas.

6.2.2

Input Adjustment

Given the problems mentioned above, it is therefore necessary to adjust the ‘Group Number’ and the constraint tables before using them. It should be noted that there is no way of deriving a true estimate of the number of residents or households prior to the imposition of disclosure control. However it is possible to improve on the method by extending the search for the same variable totals to tables in different datasets.

There are two steps needed to adjust the constraint tables. First is to adjust the ‘Group Number’ (number of people in the small areas that we want to populate). To do this the mean value of all related tables is used to give the number of people aged 16-74 in households for each small area. Secondly, each table cell is adjusted such that the row totals match these means.

The number of people in each cell is given by

Number of people from the constraint table x Group Number Total Sum for each area of the constraint table

(6.1)


115

Figure 6.1 shows the (original) constraint table from the census on the left and the adjusted constraint table on the right. The number of people for output area 00DAFA0001 in the adjusted table is 115, which is derived from 116 divided by the total sum of people in that area (from the constraint table) and multiplied by the ‘Group Number’.

As can be seen we attempt to minimise discrepancies between the totals of the constraint tables using this method. Although the adjusted tables may not be more accurate than the original CAS table, the adjustment method ensures the constraint tables are more consistent or at least can be guaranteed to produce the smallest discrepancy. Despite this, a rounding error of up to + 5 can be expected.

Figure 6.1: Constraint table adjusted method


6.2.3

116

Model Execution Process

The algorithmic steps of the Simulated Annealing-Based Reweighting Program are as follows: Step 1:

Read in model file

Step 2:

Read in constraint tables and microdata records referenced in the model file.

Step 3:

Query the microdata according to the definitions in the model file

Step 4:

Select sufficient individuals at random to populate the tables.

Step 5:

Apply simulated annealing to find the best fitting set of individuals by the step 3 query result.

Step 6:

When error = 0 or iteration count is exceeded then write out the best set of records.

Clearly, the program starts by reading in the model file (see Appendix D) which contains the path to all input datasets. Then the constraint table files are read in followed by the microdata file and the ‘Group Number’ file. The first key part of the program is the ‘microdata filtering process’. During this process the algorithm goes through the entire microdata database and checks whether an individual potentially fits into each column of the constrainting tables for the current area. This operation essentially links variables in one dataset to similar, but not identical, variables in another dataset. The filter queries the microdata by using logic operations and conditions including: -

OR

-

AND

-

OR NOT

-

AND NOT

-

=

-

‘some value’ < ‘Variable’ < ‘some value’

-

‘some value’ < ‘Variable’ =< ‘some value’

-

‘some value’ =< ‘Variable’ < ‘some value’

-

‘some value’ =< ‘Variable’ =< ‘some value’


117

For example, for the column of ‘Rented_1Car_Employed’ (people who are employed living in rented house and have 1 car) the following variables were queried.

Column&Name,Rented_1Car_Employed OR,TENHARM,=,2,INDIVI#DUAL OR,TENHARM,=,3,INDIVI#DUAL AND,REMPLOY,=,1,INDIVI#DUAL AND,NUMCARS,=,1,INDIVI#DUAL

Where: TENHARM is tenure type (2=Social rented sector; 3=Private rented sector) REMPLOY is Employment status (1 = Employed) NUMCARS is Number of Cars INDIVI#DUAL is individual microdata

Another example is useful: for the column ‘Male_25-34_not couple’ the following variables were queried. Column&Name,Male_25-34_not couple OR,MARST,=,1,INDIVI#DUAL OR,MARST,=,3,INDIVI#DUAL OR,MARST,=,4,INDIVI#DUAL OR,MARST,=,5,INDIVI#DUAL AND,SEX,=,1,INDIVI#DUAL AND,25.0,= 10 km

Burglary Dwelling

34.35

14.89

11.65

7.63

6.63

17.86

6.99

Burglary Elsewhere

28.62

21.66

17.75

6.56

5.39

14.05

5.96

Criminal Damage

55.93

13.31

7.45

5.73

4.07

9.14

4.35

Drugs Offences

39.18

13.65

11.68

7.00

5.61

13.36

9.52 16.94

Fraud & Forgery

12.35

14.26

13.01

8.55

7.96

26.92

Handling

46.34

11.10

13.19

5.87

4.57

13.45

5.48

Homicide

32.56

18.60

9.30

6.98

4.65

20.93

6.98

Other Crime

37.23

15.08

10.77

9.21

7.00

13.53

7.19

Other Theft

18.49

17.97

18.10

9.92

6.86

18.56

10.11

Robbery

31.88

19.34

16.35

9.42

5.19

13.18

4.64

Sexual Offences

38.96

11.04

10.90

5.99

6.54

13.62

12.94


19.02

15.95

20.23

12.32

6.25

18.44

7.79

Theft Of Motor Vehicle

27.13

19.31

13.88

9.73

6.89

17.37

5.69

Violent Crime

50.79

13.62

9.49

6.34

4.23

10.04

5.48

Total Known Offender

36.29

15.53

12.78

7.84

5.65

14.45

7.46


Chapter 8- Movement of Offenders and Spatial Interaction Modelling

161

Dis tance De cay 4,000

Number of movements

3,500 3,000 2,500 2,000 1,500 1,000 500 0 1 to 5

6 to 10

11 to 15

16 to 20

21 to 25

26 to 30

31-35

Distance from origin to de s tination (km.)

Figure 8.1: Distance decay for burglary dwelling Source: Derived from West Yorkshire Police

For burglary dwelling, a potential offender may prefer to commit the crime somewhere close to where they live. However, this may not always be true. For instance, where they live may be too poor to provide the kinds of goods they are interested in stealing, or opportunities are limited. For this reason, offenders may need to travel further to commit crime. However, it has been found that:

Most burglaries occur in relatively close proximity to the home of the offender. The average distance is 3.157 km (1.962 miles). About 50% of burglary dwelling incidents were committed by offenders who live less than 2 km away (Table 8.2).

The number of offenders drops steadily with an increase in the distance between the place of their residence and where they commit burglary dwelling (Figure 8.1). The distance decay principle shows that offenders commit burglary dwelling near their homes with proportionally fewer burglary dwelling incidents occuring at longer distances from home.


8.2.2

162

Crime Type Areas

There are clearly relationships between offence rates and offender rates (see below) of different areas. Following the method of Wiles and Costello (2000), an area can be categorised in terms of whether they have a high/medium/low ‘crime rate’ and ‘offender rate’. ‘Crime rate’ is derived by comparing the number of crimes with total population (or with number of households for burglary dwelling) while the ‘offender rate’ is derived by comparing the number of offenders with the total population. By a combination of these categories there are, in theory, nine possible crime type areas. However, in practice some of these are very unlikely to exist. Tables 8.3 and 8.4 illustrate the nine crime type areas for Leeds in a simple trichotomous form where Table 8.3 shows overall crime and Table 8.4 shows burglary dwelling. As can be seen, both for overall crime and burglary dwelling, the majority is found in two types of area:

High ‘offender rate’ and high ‘crime rate’ (burglary dwelling rate).

Low ‘offender rate’ and low ‘crime rate’ (burglary dwelling rate).

This suggests that if an area has a high ‘offender rate’, it tends also to have a high ‘crime rate’. However, for burglary dwelling, an area with a low ‘offender rate’ may have a high burglary dwelling rate if it is attractive to burglars. As can be seen from Table 8.4, Roundhay and Moortown have low ‘offender rates’ but have high burglary dwelling rates. This shows that Roundhay and Moortown are very attractive to burglars living nearby (as also discussed in Chapter 7). This data can be further explored by reference to the ‘National Classification of the Census Output Areas’, a three level hierarchy consisting of 7 Super-groups, 21 Groups and 52 Subgroups. The classification was created from 41 census variables and classifies every output area in the UK based on its value for those variables (Vickers, 2006a). The output area classification was aggregated up to ward level. Table 8.5 shows the number of output areas and proportion of each Super-Group in Leeds by ward.


163

Table 8.3: Crime type areas for overall crime Crime Rate Offender Rate

High

Medium

High

Medium

Low

High/High (9) Burmantofts Chapel Allerton City and Holbeck Harehills Hunslet Kirkstall Richmond Hill Seacroft University Medium/High (0)

High/Medium (2) Armley Beeston

High/Low (1) Bramley

Medium/Medium (0)

Low/High (0)

Low/Medium (2) Headingley Roundhay

Medium/Low (3) Middleton Whinmoor Wortley Low/Low (16) Aireborough Barwick and Kippax Cookridge Garforth and Swillington Halton Horsforth Moortown Morley North Morley South North Otley and Wharfedale Pudsey North Pudsey South Rothwell Weetwood Wetherby

Low

Table 8.4: Crime type areas for burglary dwelling Burglary Dwelling Rate Offender Rate

High

Medium

Low

High

Medium

High/High (8) Burmantofts Chapel Allerton City and Holbeck Harehills Kirkstall Richmond Hill Seacroft University Medium/High (2) Headingley Weetwood

High/Medium (1) Bramley

Low/High (2) Moortown Roundhay

Low/Medium (2) Horsforth North

Medium/Medium (2) Cookridge Whinmoor

Low High/Low (2) Armley Hunslet

Medium/Low (3) Beeston Middleton Wortley Low/Low (11) Aireborough Barwick and Kippax Garforth and Swillington Halton Morley North Morley South Otley and Wharfedale Pudsey North Pudsey South Rothwell Wetherby


164

Roundhay and Moortown are dominated by the ‘prospering suburbs’ super-group (Figure 8.2) which is obviously seen as a potential target for burglary dwelling. Armley and Hunslet have high ‘offender rates’ but have low burglary dwelling rates. This is partly because Leeds prison is located in Armley. Hunslet is one of the most deprived areas in Leeds and it has a very high proportion of Super-group 5 (‘constrained by circumstances’), mostly Group 5c or public housing, groups likely to be associated with high crime: “It is apparent that council estates are exposed to the highest risks of property crime, that they are associated with high offender rates…” (Murie, 1997: 23)

Th e N atio n al C la ss ifica tion o f C e ns u s O u tp u t Area s (S u p er G ro up )

North

Moortown

Roundhay

Chapel Allerton

W a rds Cla ss ific atio n Blu e C ollar C om m u m itie s City L iv ing Co un trys id e Pro sp erin g S ub urb s Co ns train ed by C ircu m s tan ce s Typ ica l Traits M u ltic ultu ra l

Figure 8.2: Area classification of output areas: Super-group Source: Derived from Vickers (2006b)

N W

E S


165

Table: 8.5: The National Classification of Census Output Areas (Super-Group) aggregated to ward level The National Classification of Census Output Areas (Super-Group)

WARD Aireborough Armley Barwick and Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth and Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley and Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley Total

Number of Output Areas 89 74 79 56 76 66 64 80 74 80 76 64 77 71 56 70 69 71 84 99 79 83 77 75 62 70 70 59 87 74 87 58 83 2,439

1 Blue Collar Communities number % 16 17.98 11 14.86 18 22.78 5 8.93 15 19.74 18 27.27 5 7.81 5 6.25 4 5.41 19 23.75 2 2.63 4 6.25 2 2.60 7 9.86 15 26.79 4 5.71 27 39.13 0 0.00 10 11.90 20 20.20 2 2.53 11 13.25 9 11.69 16 21.33 24 38.71 13 18.57 0 0.00 17 28.81 0 0.00 3 4.05 10 11.49 21 36.21 16 19.28 349 14.31

Source: Derived from Vickers (2006b)

2 City Living number 2 2 0 0 2 0 11 5 3 0 1 0 60 4 2 21 1 9 0 1 11 1 2 1 1 1 11 0 34 25 0 0 0 211

% 2.25 2.70 0.00 0.00 2.63 0.00 17.19 6.25 4.05 0.00 1.32 0.00 77.92 5.63 3.57 30.00 1.45 12.68 0.00 1.01 13.92 1.20 2.60 1.33 1.61 1.43 15.71 0.00 39.08 33.78 0.00 0.00 0.00 8.65

3 Countryside number 2 0 13 0 0 0 0 0 1 4 1 0 0 2 0 0 1 0 1 2 2 7 0 1 0 2 2 0 0 0 18 1 2 62

% 2.25 0.00 16.46 0.00 0.00 0.00 0.00 0.00 1.35 5.00 1.32 0.00 0.00 2.82 0.00 0.00 1.45 0.00 1.19 2.02 2.53 8.43 0.00 1.33 0.00 2.86 2.86 0.00 0.00 0.00 20.69 1.72 2.41 2.54

4 Prospering Suburbs number 21 6 27 6 4 3 4 0 41 36 37 3 1 28 2 4 10 29 25 22 41 22 22 5 1 24 26 4 0 12 39 13 12 530

% 23.60 8.11 34.18 10.71 5.26 4.55 6.25 0.00 55.41 45.00 48.68 4.69 1.30 39.44 3.57 5.71 14.49 40.85 29.76 22.22 51.90 26.51 28.57 6.67 1.61 34.29 37.14 6.78 0.00 16.22 44.83 22.41 14.46 21.73

5 Constrained by Circumstances number % 9 10.11 20 27.03 4 5.06 13 23.21 31 40.79 24 36.36 4 6.25 26 32.50 17 22.97 6 7.50 3 3.95 4 6.25 0 0.00 5 7.04 31 55.36 14 20.00 16 23.19 13 18.31 9 10.71 15 15.15 9 11.39 8 9.64 5 6.49 14 18.67 25 40.32 14 20.00 2 2.86 36 61.02 7 8.05 11 14.86 4 4.60 17 29.31 27 32.53 443 18.16

6 Typical Traits number 39 22 17 20 22 5 5 12 7 15 32 2 1 25 4 14 13 13 39 39 8 34 38 38 8 16 18 2 0 16 16 6 25 571

% 43.82 29.73 21.52 35.71 28.95 7.58 7.81 15.00 9.46 18.75 42.11 3.13 1.30 35.21 7.14 20.00 18.84 18.31 46.43 39.39 10.13 40.96 49.35 50.67 12.90 22.86 25.71 3.39 0.00 21.62 18.39 10.34 30.12 23.41

7 Multicultural number 0 13 0 12 2 16 35 32 1 0 0 51 13 0 2 13 1 7 0 0 6 0 1 0 3 0 11 0 46 7 0 0 1 273

% 0.00 17.57 0.00 21.43 2.63 24.24 54.69 40.00 1.35 0.00 0.00 79.69 16.88 0.00 3.57 18.57 1.45 9.86 0.00 0.00 7.59 0.00 1.30 0.00 4.84 0.00 15.71 0.00 52.87 9.46 0.00 0.00 1.20 11.19


8.2.3

166

Offender Flows

The starting point of almost all spatial interaction analysis is an origin-destination matrix. Movements of offenders can be seen as spatial interactions, with origins (where offenders live) and destinations (where crimes occur). Each pair can be represented as a cell in a matrix where rows are related to the origins, while columns are related to the destinations. Such a matrix is commonly known as an origin/destination matrix or spatial interaction matrix. In this study this matrix has been constructed at the ward level. There are 33 wards in Leeds. Therefore the offender flow matrix dimensions are 33 × 33 (1,089 cells in total). By summing the observed interaction matrix across each row, we obtain the outflow from each origin and by summing the observed interaction matrix down each column we obtain the inflow into each destination. The sum of either the outflows or the inflows gives the total movements/total flow. In terms of mean distance travelled a problem arises whenever intra-zonal distances have to be calculated. If the origin and destination are the same, the calculated distance would be zero (while the average distance travelled from within the zone would be positive in reality). This is a problem caused by using aggregate data and assuming the population to be located at zone centroids rather than distributed continuously across the region. Table 8.6 shows the distance between ward centroids within Leeds. The distance of each pair was calculated as a straight-line from the centroid of the origin ward to the destination ward centroid. Table 8.7 shows offender flows for all crime types in Leeds. As mentioned earlier, there are 70,645 records of known offenders but there are only 57,124 offenders from within Leeds. As can be seen there is less interaction if the distance between two wards is high. Table 8.8 gives the offender flows for burglary dwelling in Leeds. There are 4,728 flows in total. It should be noted that there will be a very high interaction if the origin and destinations are the same. However, in some cases it might be the case that there are more outflows to other areas. For example, the number of offenders living in Beeston that commit crimes in City and Holbeck (701) is more than the number contained in Beeston itself (599) (Table 8.7). Similarly, the number of offenders living in Kirkstall committing burglary dwelling crimes in Headingley (130) is more than that contained in Kirkstall itself (114) (Table 8.8).


167

Table 8.6: Distance between origin ward (i) to destination ward (j) Destination ( j )

Distance between i to j

(km)

Origin ( i ) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth&Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

Note:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 reborouArmleyick & KBeeston Bramle rmantopel Alle& Holbookridgh & SwHaltonHarehilleadinglHorsfortHunsleKirkstaMiddleto Moortowrley Norley So North& Whadsey No dsey SohmondRothweoundha Seacrofniversi Weetwoo Wetherb WhinmoWortley 0.0 11.6 25.0 16.2 9.0 17.6 14.0 15.0 8.5 25.0 21.0 16.3 12.2 6.2 18.3 10.7 19.7 13.0 15.3 19.5 13.3 5.7 7.3 10.4 18.2 22.8 15.9 18.2 14.3 10.3 22.9 20.0 13.0

11.6 0.0 15.6 4.7 2.7 7.5 4.9 3.5 6.0 14.1 10.7 6.8 2.7 5.5 6.7 1.8 8.2 5.8 5.7 8.6 9.9 10.0 5.4 4.4 7.2 11.3 8.2 9.1 3.9 3.7 18.1 11.6 2.5

25.0 15.6 0.0 13.7 17.8 8.1 11.3 12.7 16.5 5.6 5.2 9.0 13.5 19.3 10.9 15.3 12.9 12.0 18.0 16.3 13.1 20.4 20.8 19.8 9.0 9.4 9.1 6.8 11.7 14.9 10.3 5.1 16.6

16.2 4.7 13.7 0.0 7.2 6.4 6.3 2.2 10.2 10.7 8.5 6.6 5.5 10.1 2.9 6.1 3.5 8.1 4.4 4.6 12.6 14.5 9.6 7.3 4.7 7.1 9.1 8.5 4.5 7.6 18.9 11.1 3.7

9.0 2.7 17.8 7.2 0.0 9.7 6.6 6.1 5.0 16.6 13.1 8.8 4.3 3.1 9.4 2.5 10.7 6.8 7.0 10.7 10.1 8.2 3.0 3.4 9.7 13.9 9.7 11.1 6.1 3.8 19.2 13.5 4.1

17.6 7.5 8.1 6.4 9.7 0.0 3.6 4.9 9.4 7.6 3.4 1.4 5.5 11.6 4.5 7.3 7.3 5.0 10.6 10.3 8.5 13.8 12.7 11.8 2.2 6.9 3.7 2.2 3.6 7.2 12.7 4.7 8.7

14.0 4.9 11.3 6.3 6.6 3.6 0.0 4.1 5.8 11.2 7.0 2.4 2.4 8.0 6.1 4.1 8.8 1.9 9.6 10.8 6.3 10.3 9.6 9.2 4.8 9.9 3.3 4.5 1.8 3.6 13.3 6.9 6.9

15.0 3.5 12.7 2.2 6.1 4.9 4.1 0.0 8.3 10.7 7.6 4.7 3.5 8.8 3.3 4.4 5.2 5.9 5.8 6.8 10.4 12.7 8.9 7.2 3.9 7.8 7.0 6.9 2.4 5.6 17.0 9.5 3.9

8.5 6.0 16.5 10.2 5.0 9.4 5.8 8.3 0.0 17.0 12.8 8.1 4.9 3.8 11.3 4.3 13.5 4.6 11.6 14.5 5.7 4.5 6.8 8.4 10.5 15.5 7.4 9.8 6.7 2.7 15.3 11.6 8.4

25.0 14.1 5.6 10.7 16.6 7.6 11.2 10.7 17.0 0.0 4.4 9.0 12.8 18.9 7.8 14.5 8.7 12.6 14.7 12.0 15.4 21.4 19.5 17.8 6.9 4.4 10.6 7.8 10.7 14.7 15.4 8.0 14.2

21.0 10.7 5.2 8.5 13.1 3.4 7.0 7.6 12.8 4.4 0.0 4.7 8.9 15.0 5.8 10.7 8.0 8.3 12.8 11.3 11.0 17.1 16.0 14.8 3.8 5.5 6.2 3.4 7.0 10.7 12.5 4.3 11.5

16.3 6.8 9.0 6.6 8.8 1.4 2.4 4.7 8.1 9.0 4.7 0.0 4.5 10.4 5.3 6.3 8.2 3.6 10.6 10.8 7.2 12.4 11.8 11.1 3.3 8.2 2.7 2.4 2.9 6.0 12.3 4.9 8.3

12.2 2.7 13.5 5.5 4.3 5.5 2.4 3.5 4.9 12.8 8.9 4.5 0.0 6.1 6.4 1.8 8.7 3.2 8.0 10.0 7.5 9.3 7.3 6.9 6.0 10.8 5.6 6.8 2.1 2.2 15.5 9.2 4.9

6.2 5.5 19.3 10.1 3.1 11.6 8.0 8.8 3.8 18.9 15.0 10.4 6.1 0.0 12.0 4.5 13.6 7.5 10.0 13.8 9.4 5.3 3.3 5.7 12.0 16.6 10.5 12.5 8.2 4.4 19.0 14.6 7.2

18.3 6.7 10.9 2.9 9.4 4.5 6.1 3.3 11.3 7.8 5.8 5.3 6.4 12.0 0.0 7.6 2.9 8.0 7.2 5.8 12.3 15.7 12.0 10.0 2.4 4.6 8.0 6.6 4.6 8.6 17.1 9.0 6.5

10.7 1.8 15.3 6.1 2.5 7.3 4.1 4.4 4.3 14.5 10.7 6.3 1.8 4.5 7.6 0.0 9.5 4.5 7.5 10.3 8.2 8.4 5.5 5.4 7.6 12.1 7.2 8.6 3.8 1.9 16.9 11.0 4.3

19.7 8.2 12.9 3.5 10.7 7.3 8.8 5.2 13.5 8.7 8.0 8.2 8.7 13.6 2.9 9.5 0.0 10.7 6.4 3.4 15.0 17.9 13.0 10.5 5.1 4.4 10.8 9.4 7.2 10.8 19.9 11.6 7.0

13.0 5.8 12.0 8.1 6.8 5.0 1.9 5.9 4.6 12.6 8.3 3.6 3.2 7.5 8.0 4.5 10.7 0.0 11.1 12.6 4.5 8.8 9.6 9.9 6.6 11.7 3.0 5.3 3.6 3.2 12.4 7.1 8.1

15.3 5.7 18.0 4.4 7.0 10.6 9.6 5.8 11.6 14.7 12.8 10.6 8.0 10.0 7.2 7.5 6.4 11.1 0.0 4.4 15.4 15.2 8.1 5.0 9.1 10.7 12.8 12.7 8.0 9.3 22.8 15.3 3.2

19.5 8.6 16.3 4.6 10.7 10.3 10.8 6.8 14.5 12.0 11.3 10.8 10.0 13.8 5.8 10.3 3.4 12.6 4.4 0.0 17.1 18.6 12.3 9.3 8.2 7.6 13.4 12.4 9.1 11.9 22.9 14.8 6.6

is i = j

The distance was calculated as straight-line from centroid of origin ward and destination ward Source: Derived from West Yorkshire Police

13.3 9.9 13.1 12.6 10.1 8.5 6.3 10.4 5.7 15.4 11.0 7.2 7.5 9.4 12.3 8.2 15.0 4.5 15.4 17.1 0.0 7.9 12.4 13.5 10.5 15.4 4.9 7.6 8.0 6.4 9.7 8.1 12.3

5.7 10.0 20.4 14.5 8.2 13.8 10.3 12.7 4.5 21.4 17.1 12.4 9.3 5.3 15.7 8.4 17.9 8.8 15.2 18.6 7.9 0.0 8.4 11.0 15.0 20.0 11.3 13.9 11.2 7.1 17.3 15.3 12.2

7.3 5.4 20.8 9.6 3.0 12.7 9.6 8.9 6.8 19.5 16.0 11.8 7.3 3.3 12.0 5.5 13.0 9.6 8.1 12.3 12.4 8.4 0.0 3.1 12.6 16.5 12.6 14.1 9.1 6.5 21.8 16.5 6.0

10.4 4.4 19.8 7.3 3.4 11.8 9.2 7.2 8.4 17.8 14.8 11.1 6.9 5.7 10.0 5.4 10.5 9.9 5.0 9.3 13.5 11.0 3.1 0.0 11.1 14.4 12.5 13.5 8.2 7.1 22.3 16.0 3.6

18.2 7.2 9.0 4.7 9.7 2.2 4.8 3.9 10.5 6.9 3.8 3.3 6.0 12.0 2.4 7.6 5.1 6.6 9.1 8.2 10.5 15.0 12.6 11.1 0.0 5.1 5.9 4.3 3.9 8.0 14.8 6.6 7.7

22.8 11.3 9.4 7.1 13.9 6.9 9.9 7.8 15.5 4.4 5.5 8.2 10.8 16.6 4.6 12.1 4.4 11.7 10.7 7.6 15.4 20.0 16.5 14.4 5.1 0.0 10.6 8.3 8.8 12.9 18.0 9.7 10.8

15.9 8.2 9.1 9.1 9.7 3.7 3.3 7.0 7.4 10.6 6.2 2.7 5.6 10.5 8.0 7.2 10.8 3.0 12.8 13.4 4.9 11.3 12.6 12.5 5.9 10.6 0.0 2.8 4.8 6.1 10.0 4.2 10.2

18.2 9.1 6.8 8.5 11.1 2.2 4.5 6.9 9.8 7.8 3.4 2.4 6.8 12.5 6.6 8.6 9.4 5.3 12.7 12.4 7.6 13.9 14.1 13.5 4.3 8.3 2.8 0.0 5.3 8.1 10.5 2.6 10.6

14.3 3.9 11.7 4.5 6.1 3.6 1.8 2.4 6.7 10.7 7.0 2.9 2.1 8.2 4.6 3.8 7.2 3.6 8.0 9.1 8.0 11.2 9.1 8.2 3.9 8.8 4.8 5.3 0.0 4.1 14.8 7.8 5.5

10.3 3.7 14.9 7.6 3.8 7.2 3.6 5.6 2.7 14.7 10.7 6.0 2.2 4.4 8.6 1.9 10.8 3.2 9.3 11.9 6.4 7.1 6.5 7.1 8.0 12.9 6.1 8.1 4.1 0.0 15.4 10.2 6.2

22.9 18.1 10.3 18.9 19.2 12.7 13.3 17.0 15.3 15.4 12.5 12.3 15.5 19.0 17.1 16.9 19.9 12.4 22.8 22.9 9.7 17.3 21.8 22.3 14.8 18.0 10.0 10.5 14.8 15.4 0.0 8.3 20.2

20.0 11.6 5.1 11.1 13.5 4.7 6.9 9.5 11.6 8.0 4.3 4.9 9.2 14.6 9.0 11.0 11.6 7.1 15.3 14.8 8.1 15.3 16.5 16.0 6.6 9.7 4.2 2.6 7.8 10.2 8.3 0.0 13.2

13.0 2.5 16.6 3.7 4.1 8.7 6.9 3.9 8.4 14.2 11.5 8.3 4.9 7.2 6.5 4.3 7.0 8.1 3.2 6.6 12.3 12.2 6.0 3.6 7.7 10.8 10.2 10.6 5.5 6.2 20.2 13.2 0.0


168

Table 8.7: Offender flows of all crime in Leeds Destination ( j )

All Crime 1 Origin ( i ) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth&Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

Inflow Note:

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

Out

eboro Armleyck and BeestonBramleyurmantoapel Alley & Holbeookrid & SwHaltonHarehillseadingleorsforHunsletKirkstalMiddletoMoortoworley Noorley So North& Whadsey Nosey SchmondRothwe RoundhaSeacrofUniversit Weetwoo WetherhinmoWortley flow

586 1 20 1,047 1 2 11 20 22 253 7 17 8 32 18 202 19 12 1 2 2 1 8 46 6 20 31 8 15 43 31 3 10 2 3 4 1 3 11 4 41 8 12 4 25 6 22 4 3 9 13 20 10 44 10 8

4 1 5 46 5 7 27 217 53 19 554 9 9 12 37 391 2 19 6 54 70 19 7 6 599 14 13 29 701 1 2 2 10 7 32 1,181 22 21 270 10 1 13 10 22 18 13 1,405 92 672 6 25 164 243 5 18 28 43 1,173 509 14 6 17 189 10 352 64 69 61 3,042 35 8 27 42 2 3 11 17 14 134 382 13 6 3 39 3 8 10 108 1 336 26 1 18 11 3 43 6 110 1 25 224 8 9 53 20 274 291 852 25 6 44 1,141 18 11 17 40 259 8 4 4 17 1 1 2 2 93 32 1 1 12 97 6 29 19 501 13 24 14 18 34 39 12 24 301 49 2 3 13 2 79 12 16 6 274 3 5 3 11 1 9 2 10 77 88 13 3 14 1 14 7 11 5 117 1 3 2 1 2 35 4 5 6 158 1 4 3 8 2 5 7 8 46 131 7 1 4 16 1 3 4 27 6 1 4 23 3 45 3 1 3 51 8 3 59 3 1 3 25 26 16 294 58 902 6 47 218 97 2 4 2 11 10 100 1 10 8 7 5 1 18 55 113 2 10 64 37 25 27 227 69 381 6 32 209 144 4 14 15 97 145 902 31 13 16 135 2 6 9 9 29 174 59 7 3 4 2 3 3 8 18 1 2 1 3 6 31 18 79 19 151 2 15 125 20 6 202 1 56 60 20 20 323 2 4 5 913 2,076 647 1,548 1,869 2,846 2,375 12,169 735 672 1,187 2,270

= No interaction

= When i = j Source: Derived from West Yorkshire Police

4 76 3 12 84 50 105 86 26 2 4 69 455 8 21 280 7 13 10 9 2 4 2 37 2 15 28 178 73 2 6 27 1,700

30 3 63 33 2 1 7 81 29 13 2 34 18 23 34 186 99 7 8 5 13 52 10 18 228 1 3 1,089 166 3 6 139 2 1 1 16 3 9 2 6 9 15 1 4 1 10 47 21 5 3 15 28 16 25 37 12 1 5 2 5 9 15 846 1,886

7 3 1 1 28 14 1 1 158 17 18 46 38 10 10 111 68 33 18 27 3 1 4 5 4 5 2 2 2 7 9 6 18 39 10 119 62 4 1 10 5 10 11 12 148 9 18 33 9 26 12 193 75 16 5 10 26 12 31 30 15 20 21 15 179 19 102 42 7 206 20 7 56 15 7 6 11 12 88 98 46 45 148 59 38 9 69 39 40 34 29 53 4 15 5 2 6 16 13 2 5 1 11 1 3 10 3 1 2 12 39 2 12 3 3 1 2 1 25 6 3 64 8 74 36 14 20 11 27 9 51 23 252 82 6 39 2 2 15 3 14 4 6 1 8 27 1 1 14 10 7 1 13 245 24 98 47 9 6 7 11 34 87 29 815 4 27 5 5 8 16 40 3 8 8 11 17 779 4 70 61 3 11 6 10 46 4 11 1 298 2 1 47 2 1 1 2 41 6 3 5 337 135 8 4 4 13 9 1 3 9 21 6 172 914 1 3 3 4 5 12 5 117 2 4 406 2 6 1 1 1 35 3 10 2 350 10 1 13 1 2 11 1 285 69 1 1 21 4 1 2 1 99 454 2 1 31 11 20 28 17 33 14 15 2 1,089 42 51 4 26 3 16 6 2 2 5 9 355 2 11 5 50 1 2 46 5 1 2 2 183 26 7 49 28 9 33 6 31 10 47 18 77 106 16 36 32 11 29 15 21 8 25 9 44 44 6 44 11 10 20 16 11 1 1 6 14 2 1 1 1 4 3 13 13 3 6 14 8 9 2 10 17 22 48 6 6 34 6 5 1 72 27 17 8 11 1,944 1,289 1,181 1,313 1,451 868 594 1,117 834 1,665 787 1,086

1 23 12 13 19 183 17 36 29 6 43 115 13 22 7 5 4 2 5 3 2 3 110 9 4 1,322 31 7 4 140 2 2,192

1 100 17 56 48 214 231 149 25 16 8 290 116 44 50 96 26 21 16 8 17 1 2 4 187 13 27 78 971 55 5 19 29 2,940

13 32 4 10 45 11 151 39 73 1 9 31 62 23 24 135 5 12 8 2 25 7 4 3 16 1 4 12 73 331 1 9 11 1,187

4 3 5 3 3 28 21 8 2 1 3 19 2 8 3 6 7 5 2 9 2 19 3 7 31 5 4 376 12 1 602

1 1 762 7 180 3,082 5 670 5 20 1,916 5 81 2,723 63 12 3,751 7 19 3,111 21 142 5,285 5 6 1,021 5 647 32 1 613 22 32 4,001 3 2 1,267 1 538 14 33 2,625 5 11 2,210 2 5 1,637 687 1 15 763 5 1,411 3 2 903 1 477 8 518 29 794 23 10 3,529 6 5 652 3 3 656 140 3 3,188 5 20 3,102 2 3 1,028 4 2 450 554 2 1,335 6 732 1,772 949 1,386 57,124


169

Table 8.8: Offender flows of burglary dwelling in Leeds Burglary Dwelling

Destination ( j ) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Out eboroArmley c k & Keestoramlermantpel All& Holookrid& SwHaltonarehiladingorsforHunsle Kirkstaiddletooortowrley Nley SNorth& Whas ey Nsey Smondothwoundheacro e niv erseetw oetherhinmo Wortley flow

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Origin ( i ) 3 2 1 38 Aireborough 2 58 5 1 31 5 1 7 2 8 1 22 27 2 10 9 2 4 13 4 4 4 3 14 6 4 4 Armley 22 1 4 1 Barwick & Kippax 1 1 49 1 3 18 1 5 9 2 4 2 3 1 2 1 1 2 Beeston 2 22 4 13 80 11 1 19 4 10 2 8 6 3 14 3 13 8 5 4 7 5 4 Bramley 2 4 3 97 11 5 1 1 27 9 2 1 2 3 1 6 3 32 30 13 5 Burmantofts 2 2 2 1 3 61 2 2 2 6 37 1 2 7 33 12 21 22 Chapel Allerton 12 2 29 8 6 2 116 2 7 5 11 11 20 8 5 27 4 10 20 3 7 3 7 3 2 8 City and Holbeck 3 3 1 2 1 2 2 47 10 19 22 1 1 2 3 2 3 3 Cookridge 2 1 11 1 20 1 Garforth&Swillington 1 10 1 17 1 1 11 1 1 2 7 3 3 2 7 Halton 1 5 3 3 24 15 9 4 78 33 9 27 1 9 2 3 5 2 48 5 32 Harehills 3 4 7 2 1 72 2 8 2 13 3 1 2 1 4 16 Headingley 4 1 2 1 3 Horsforth 2 6 1 2 16 9 2 5 1 2 58 1 31 4 13 2 5 3 3 7 2 19 19 1 1 Hunslet 15 4 23 5 24 12 1 130 26 114 2 1 3 3 16 1 1 3 1 10 Kirkstall 1 3 1 2 1 1 1 1 11 70 2 1 3 1 1 Middleton 1 5 1 3 1 1 19 6 9 Moortown 1 3 1 1 9 2 1 Morley North 2 3 1 1 1 2 23 75 1 Morley South 1 1 1 15 62 1 4 North 1 8 Otley & Wharfedale 3 1 1 1 4 1 Pudsey North 1 2 1 1 11 Pudsey South 2 4 5 4 18 14 4 8 11 11 8 3 6 1 3 2 4 4 1 87 9 7 1 15 Richmond Hill 2 2 1 1 3 1 2 28 1 Rothwell 1 2 1 1 9 2 1 2 3 1 8 3 Roundhay 2 2 2 5 2 3 4 3 3 7 10 3 4 2 3 3 1 1 6 1 9 99 14 Seacroft 3 2 1 3 15 11 12 6 49 9 5 7 8 11 5 3 3 1 1 17 1 113 University 2 2 2 1 25 31 2 4 5 2 4 13 1 2 1 17 Weetwood Wetherby 1 2 13 1 1 6 18 1 1 1 1 2 4 1 1 2 11 Whinmoor 11 29 3 1 1 51 4 1 1 1 4 1 2 Wortley 75 127 68 187 149 170 151 310 121 53 62 162 443 124 136 236 138 162 74 118 169 48 51 39 150 114 205 157 272 Inflow Note:

= No interaction

= when i = j Source: Derived from West Yorkshire Police

6

14 2 56 11 11 4 3 14 6 30

1

9

1

1 1 18 13 1 1 1 6 10 2 3

4

10 2

2

4

3 1

1 10 7

2 4

1

1 1

2

1 1 11 17

1

2 1

20 1 12 2

1 1

55 2 49 195 30 115 117

44 269 28 107 281 276 276 365 144 36 82 330 156 11 232 436 100 46 20 109 90 12 8 19 237 41 35 210 299 132 12 124 161 4,728


170

Figure 8.3 shows the catchment area of crime for Burmantofts, one of the most deprived wards in Leeds and one that contains a large number of offenders. The catchment area is quite tight with most crime being local (average distance is 2.151 km). However, Figure 8.4 shows the catchment area for the ward of Headingley. This ward contains a mixture of income groups, but is especially dominated by the large Leeds student population. Thus, it provides a wealth of opportunities for the criminal community. This is reflected in a wider catchment area (average distance is 2.646 km) for criminal activity in Headingley.

Offender from other wards to Burmantofts

Offender from other wards to Burmantofts 0-1 % 1-5 % 5-10 % 10-20% More than 20 %

N

Burmantofts

W

E S

Figure 8.3: Catchment area for crime in Burmantofts Source: Derived from West Yorkshire Police

Offender from other wards to Headingley

Headingley

Offender from other wards to Headingley 0-1 % 1-5% 5-10 % 10-20 % More than 20 %

N W

E S

Figure 8.4: Catchment area for crime in Headingley Source: Derived from West Yorkshire Police


8.2.4

171

Inflow/Outflow Ratio

As mentioned in the previous section the inflow and outflow totals can be derived for each origin and destination from the offender flow matrix. Inflow and outflow can be compared to get an inflow/outflow ratio which is a measure of the relative flow of offenders residing in one area but committing crimes in other areas. A ratio greater than one means there is a greater number of offenders coming into an area than going out, while a ratio less than one means there are less number of offenders coming in than going out. Figure 8.3 plots inflow/outflow ratios and shows certain inner city wards to have more offenders going out to commit burglary dwelling than coming in (more burglars living in an area than burglary dwelling committed in an area), notably Harehills and Armley (ratio of less than 0.5). On the other hand, outer city wards have more offenders coming in than going out, especially in Roundhay, Horsforth, and Pudsey North with a ratio greater than 5. Inflow/outflow ratios reflect push and pull factors. If the ratio is very low it means that an area may be too poor and does not have valuable things to steal or that there is more policing/less opportunities and therefore offenders are encouraged to travel for some distance to commit burglary dwelling. If the ratio is very high it means that an area is very attractive to burglars.

Inflow/ Outflow (Burglary Dwelling)

Horsforth

Roundhay

Harehills

Pudsey North

Armley

N W

E S

Figure 8.3: Inflow/Outflow of burglary dwelling in Leeds Source: Derived from West Yorkshire Police

Inflow/Outflow 0 - 0.5 0.5-1 1 - 2.5 2.5 - 5 More than 5


8.2.5

172

Self-Containment

It should be noted that most offenders commit crime in their local areas (not far from where they live). To measure this it is also useful to define‘self-containment’ as

SelfContai nment =

Oij Cj

Where Oij = Number of offenders travelling from origin i to destination j Cj = Number of crimes occuring in area j And we only consider cases where i = j The ‘degree of self-containment’ shows the proportion of crimes committed by people who live in that area. It can be measured at any spatial scale. Table 8.9 shows ‘degree of selfcontainment’ for crime in Leeds by ward. The table shows degree of self-containment for every crime type on the left and for burglary dwelling on the right. As can be seen, most areas have high degrees of self-containment. For burglary dwelling, generally about half of burglary incidents in each ward are perpetrated by their residents, and the rest come from elsewhere. However, it is interesting to note that Horsforth (0.02), Roundhay (0.04), Pudsey North (0.08), and Weetwood (0.09) have very low degrees of self-containment which means burglary dwelling incidents in these areas are mainly committed by offenders from other areas. This implies they have high attractiveness to offenders living in other areas. As mentioned in § 8.2.2 Roundhay is dominated by output areas that are classified as ‘prospering suburb’. Horsforth is the same (Table 8.5). Being prospering suburbs makes them more attractive than other areas. In Weetwood most of the burglary dwelling crimes were committed by offenders living in Chapel Allerton and Kirkstall (Table 8.8) which are connected to Weetwood from the south east and from the south west respectively.

It should be noted that there is high negative correlation between ‘degree of selfcontainment’ and ‘inflow/outflow ratio’. As can be seen from Figure 8.4, areas with a high ‘degree of self-containment’ generally have a low inflow/outflow ratio, as might be expected.


173

Table 8.9: Self-containment of crime in Leeds by ward Total Crime Ward Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth & Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor

No of offenders in area i

Total Inflow of area j

762 3,082 670 1,916 2,723 3,751 3,111 5,285 1,021 647 613 4,001 1,267 538 2,625 2,210 1,637 687 763 1,411 903 477 518 794 3,529 652 656 3,188 3,102 1,028 450 1,335

Note: i = origin ward; j = destination ward Source: Derived from West Yorkshire Police

913 2,076 647 1,548 1,869 2,846 2,375 12,169 735 672 1,187 2,270 1,700 846 1,886 1,944 1,289 1,181 1,313 1,451 868 594 1,117 834 1,665 787 1,086 2,192 2,940 1,187 602 949

Burglary Dwelling

Number of crime When i = j 586 1,047 391 599 1,181 1,405 1,173 3,042 382 336 224 1,141 455 228 1,089 815 779 298 337 914 406 350 285 454 1,089 355 183 1,322 971 331 376 554

Self-containment 0.64 0.50 0.60 0.39 0.63 0.49 0.49 0.25 0.52 0.50 0.19 0.50 0.27 0.27 0.58 0.42 0.60 0.25 0.26 0.63 0.47 0.59 0.26 0.54 0.65 0.45 0.17 0.60 0.33 0.28 0.62 0.58

No of Burglars in area i 44 269 28 107 281 276 276 365 144 36 82 330 156 11 232 436 100 46 20 109 90 12 8 19 237 41 35 210 299 132 12 124

Total Inflow of area j 75 127 68 187 149 170 151 310 121 53 62 162 443 124 136 236 138 162 74 118 169 48 51 39 150 114 205 157 272 195 30 115

Number of Burglary Dwelling When i = j 38 58 22 49 80 97 61 116 47 11 11 78 72 2 58 114 70 19 9 75 62 8 4 11 87 28 8 99 113 17 12 55

Self-containment 0.51 0.46 0.32 0.26 0.54 0.57 0.40 0.37 0.39 0.21 0.18 0.48 0.16 0.02 0.43 0.48 0.51 0.12 0.12 0.64 0.37 0.17 0.08 0.28 0.58 0.25 0.04 0.63 0.42 0.09 0.40 0.48


174

0.70

0.60

Self-Containment

0.50

0.40

0.30

0.20 2

R = 0.7767 0.10

0.00 0.00

2.00

4.00

6.00

8.00

10.00

12.00

Inflow/Outflow

Figure 8.4: Relationship between ‘inflow/outflow ratio’ and ‘degree of self-containment’

8.3 Spatial Interaction Model for Burglary Dwelling: Model Specification Some questions emerge from the offender flow matrix:

Why are there large flows between some areas but small or no flows between others?

What are the characteristics of destinations that make them attractive for burglary?

What are the characteristics of origins that have a large proportion of outflows that are unattractive to local offenders?

From the offender dataset we know the outflow from each origin and the inflow into each destination, so what is the pattern of flows likely to be between these areas? And what determines the magnitude of spatial interaction between where the offenders live and where the crimes occur?

To help answer these questions, a spatial interaction model is formulated. This can be used to explain and predict flows between areas. Spatial interaction models normally take the form of an equation which is made up of the independent variables that influence the level of


175

spatial interaction between areas (the dependent variable to be estimated). Once a spatial interaction model has been validated for a region, it can then be used for simulation and prediction purposes: such as how many additional flows would be generated if the population increased. The spatial interaction model in this study is designed for burglary dwelling only. The following sections will describe model formulation, attractiveness factors, model calibration, goodness-of-fit statistics and then a model summary.

8.3.1

Model Formulation

In general, flow-based spatial interactions are a function of the attributes of the places of origin, the attributes of the places of destination, and the friction of distance between the origins and the destinations. The basic formulation of the spatial interaction model is as follows:

Sij = Oi Ai W jα e

− β d ij

(8.1)

Sij = Interaction between area i (origin) and area j (destination), which is the flow of offenders from area i (offender living area) to area j (offence area). Oi = The attributes of the location of area i (origin), which is the number of offenders. Wj = The attributes of the location of destination j, which is a measure of the attractiveness of destination j dij = Distance between area i and area j Ai = A balancing factor, which formally is written as:

Ai =

1 − βd ∑Wk e ij

(8.2)

= Oi

(8.3)

k

To ensure that

∑S

ij

j

α (alpha) = The attractiveness power β (beta) = The distance decay parameter


8.3.2

176

Attractiveness Factors

The attractiveness factor is the ‘pulling power’ that a destination has. The more pulling power then the more likely the destination is to pull offenders across greater distances. However, it should be noted that the factors that prompt offenders to commit crimes are not the same as those inducing people to travel to work or to spend money in shops. For example in the case of burglary dwelling, attractiveness could be the number of affluent households that have more valuable things to be stolen. To be more accurate, the attractiveness factor may be a function of several factors. The factors used here are census variables, or derived from census variables (Table 8.10). In order to determine the best variables, they were regressed against crime inflow for burglary dwelling. The variable with the highest correlation coefficient is accepted as the most important determinant, though the match with the hypothetical ‘attractiveness’ will not be exact as the regression does not include the push factors (yet the flows are influenced by these). In addition, we should recall that we only have data for detected crimes, and the attractiveness associated with undetected crimes could be different. Nonetheless, as can be seen in Table 8.10 the ‘percentage of rented tenure multiplied by percentage of students’ has the highest R value. Therefore this factor has been used as the first (though see next section) attractiveness variable for the spatial interaction component of burglary dwelling.

Table 8.10: Correlations between potential attractiveness factors and inflow of burglary dwelling. Potential Attractive Factors All people Age 16-74 All HHs Number of Student** Percentage of Student** Number of HighClass Percentage of HighClass* All Car in area Number of Rented House** Percentage of Rented** Number of Unemployed** Multiple Deprivation Index* IndexOfWealth Percentage of Rented x Percentage of Student** FTStudent x HighClass** HHs x Student** HHs x Student*HighClass** Percentage of Rented x Percentage of HighClass** ** Correlation is significant at the 0.01 level (2-tailed) * Correlation is significant at the 0.05 level (2-tailed)

R 0.058 -0.120 0.762 0.793 -0.330 -0.349 -0.534 0.765 0.735 0.503 0.377 0.235 0.791 0.689 0.768 0.676 0.694


8.3.3

177

Scaled Attractiveness Factors

It was found that the first attempt using the attractiveness variable ‘percentage of rented * percentage of student’ did not produce a very accurate prediction (the results will be shown later in Table 8.14, § 8.3.6). Therefore a new attractiveness factor is incorporated by adapting the method used by Eyre (1999). The current attractiveness factor σ j is multiplied by a new weight factor γ j (which could be comprised of many variables).

W j =σ jγ j

(8.4)

The weight factor is made up of two components, a weight associated with the variable value and a weight associated with the correlation coefficient. The total weight factor is calculated thus: 1) The value-associated weight is set so that the weight associated with the most attractive ward for a given variable (i.e. the ward with the highest value for that variable) is 10. All the value-associated weights for the other areas are scaled relative to this using their variable levels. The scaled value of each weight for each area is given by η j , Where f is the variable being considered. f

η is formulated by η = f j

f j

ν jf ν jf= highest

× 10

(8.5)

Where ν j is equal to the value of the variable f for area j. f

2) Each of the correlation coefficients for the variables are not equal. Therefore, the importance of the variables need adjusting on this basis using a correlationassociated weight. This is given for each variable by

χf =

rf ∑r f f

where r is the correlation coefficient.

(8.6)


178

We thus have a value-associated weight for each area, and a correlation-associated weight for each variable. The final weight factor γ j used to adjust the prior attractiveness for each area j is χ f for the given area multiplied by the η j for each variable, summed together for f

the area.

γ j = ∑ χ f × η jf

(8.7)

f

In practice, to determine the appropriate variables to use a number of experiments were run using different variables and Standardised Root Mean Square Error (SRMSE) was used to compare which model was the best, including:

Deprivation and wealth (SRMSE = 0.312)

Deprivation, wealth, and working hours (SRMSE = 0.294)

Deprivation, and number of cars per household (SRMSE = 0.284)

Deprivation, and number of cars in an area (SRMSE = 0.282)

Deprivation, number of cars in an area, and wealth (SRMSE = 0.284)

Deprivation, wealth, and percentage of high-class people in an area (SRMSE = 0.298)

Number of cars per household, and full-time working (SRMSE = 0.31)

It was thus found that the ‘deprivation’ and ‘number of cars in an area’ variables gave the best results with an SRMSE of 0.282 (more information about SRMSE will be given in § 8.3.5). Therefore ‘deprivation’ and ‘number of cars in area’ were incorporated, as above, into the value of the current attractiveness variable, σ j (percentage of rented tenure * percentage of student).

γ j derived by experimentation: f1 = Deprivation variable f2 = Number of cars in area variable Step 1) The value-associated weight for ‘deprivation’ and ‘number of cars in area’ for each ward were derived using equation 8.5. The most attractive ward was set to 10. The real values of ‘deprivation’ and ‘number of cars in area’ can be found in Appendix C.


179

Step 2) The correlation-associated weights for these two variables were set using equation 8.6. The ‘deprivation’ variable and ‘number of cars in area’ variable originally had correlation coefficients of 0.377 and 0.534 respectively (Table 8.10). Therefore

∑r

f

= 0.911 (derived by 0.377+0.534) and this gives a weight of 0.414 (derived

f

by 0.377/0.911) for the ‘deprivation’ variable. In the same manner the ‘number of cars’ variable has a correlation coefficient of 0.534 and therefore it gives a weight of 0.586 in the attractiveness factor (derive by 0.534/0.911). Step 3) γ j is derived by equation 8.7 for each area. For example, γ j for Aireborough is derived by

γj

= (0.414*6.79) + (0.586*8.22) = 2.81 + 4.82 = 7.63 (see the right column of Table 8.11)

For Armley:

γj

= (0.414*2.08) + (0.586*4.62) = 0.86 + 2.71 = 3.57

Table 8.11 shows the γ j values calculated for the 33 wards in Leeds.


Table 8.11: The values calculated for

γj

for 33 wards Step 2: χ

Step 1:

η jf =

ν ν

f j

f j = highest

Aireborough Armley Barwick and Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth and Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley and Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

6.79 2.08 4.06 1.69 1.97 1.28 1.62 1.24 5.71 4.97 6.47 1.27 4.26 10.00 1.44 2.36 1.86 4.16 4.58 3.69 5.91 7.83 5.75 3.52 1.31 4.08 5.75 1.25 1.44 4.39 9.97 2.28 2.18

f

=

× 10

Ward

Scaled Deprivation

180

Scaled Number of Car in area

8.22 4.62 7.93 3.09 4.61 2.94 3.76 3.21 6.95 7.38 6.92 3.19 4.91 7.13 2.61 4.39 4.87 6.28 7.64 8.38 7.70 8.51 7.24 6.04 2.76 6.22 7.15 2.83 2.96 5.52 10.00 4.17 5.54

rf ∑r f

Step 3:

f

Original R= 0.377

Original R= 0.534

χf

χf

= 0.414

χ f × η jf ) 2.81 0.86 1.68 0.70 0.81 0.53 0.67 0.51 2.36 2.06 2.68 0.53 1.76 4.14 0.59 0.98 0.77 1.72 1.90 1.53 2.45 3.24 2.38 1.46 0.54 1.69 2.38 0.52 0.60 1.81 4.13 0.94 0.90

f

= 0.586

Weighted Number of Car in area

Weighted Deprivation (

γ j = ∑ χ f × η jf

(

χ f × η jf ) 4.82 2.71 4.65 1.81 2.70 1.72 2.21 1.88 4.07 4.33 4.06 1.87 2.88 4.18 1.53 2.57 2.85 3.68 4.48 4.91 4.52 4.99 4.25 3.54 1.62 3.65 4.19 1.66 1.73 3.24 5.86 2.44 3.25

7.63 3.57 6.33 2.51 3.52 2.25 2.88 2.39 6.43 6.38 6.74 2.40 4.64 8.32 2.13 3.55 3.62 5.40 6.38 6.44 6.96 8.23 6.63 5.00 2.16 5.34 6.57 2.18 2.33 5.05 9.99 3.39 4.15


8.3.4

181

Model Calibration

Calibration is a significant challenge associated with the use of spatial interaction models. It is the process by which model parameters, in particular the distance decay parameter (beta) and the attractiveness power (alpha), are set to ensure that the estimated flows are similar or close to the observed flows. Changing the value of alpha (α) and beta (β) will influence the estimated spatial interaction. For example, the relationship between distance and spatial interactions will change according to the beta exponent. The parameter beta is attached to the distance decay function. If the beta value is high (higher than 0.5), the friction of distance will be much more important or, in other words, people find it difficult (or are unwilling) to make long trips. A low value of beta (e.g. 0.25) indicates relative ease of travel. When beta is 0 distance has no effect and interactions will be the same even if distance is changed. Alpha (α) reflects scale economies - ‘bigger’ destinations are even more attractive when alpha is greater than 1. A value of 1 means there is a linear relationship. The spatial interaction model in this study is calibrated to reproduce existing interaction patterns between ‘where offenders live’ and ‘where crimes (burglary dwelling incidents) occur’. It is calibrated by trial and error using a small program written in Java called the ‘Spatial Interaction Calibrator’ which completes a brute-force search of the solution space to find the best represented parameters (alpha and beta values) that minimise the difference between the ‘predicted’ and ‘observed’ set of flows. To run the program the following data needs to be input. 1) Attractiveness (Wj) 2) Number of offenders at the origin (Oi) 3) Distance between the origin i and destination j (33 x 33 matrix) 4) Observed counts, which in this case is burglary dwelling flows (33 x 33 matrix) In addition the following control parameters were set:

Minimum value of beta = 0

Minimum value of alpha = 0

Maximum value of beta = 2

Maximum value of alpha = 2

Increasing value = 0.01 each iteration


182

Thus, every combination of alpha and beta between 0 and 2 is tried, with a resolution of 0.01. The program starts by calculating this following equation

Sij = Oi Ai W jα e

− βdij

(8.8)

At each alpha and beta values the predicted set of burglary dwelling flows of the 33 wards is compared against the observed counts. Highlight areas (in blue colour) in Figure 8.5a and 8.5b show those cells in the observed flows and the predicted flows to be compared. If Total Absolute Error (TAE) or sum of the residuals (sum of the ‘observed – predicted difference’ for each cell) (Figure 8.5c) improve then the alpha and beta value will be kept. The TAE is derived by

TAE =

∑S

ij

− Sˆ ij

(8.9)

ij

Where S ij is the observed count for the row i in column j

Sˆ ij is the predicted count for the row i in column j This process was continued until the maximum values of both parameters were reached. The final results are the value of alpha and beta that make the predicted flows as close to the observed flow as possible.


Destinations ( j )

Observed Flows ( S ij )

O R I G I N S (i)

1

2

3

…

…

33

… … … … … … …

… … … … … … …

… … … … … … …

1

38

0

0

2

2

58

5

3

0

0

22

:

… … … …

… … … …

… … … …

: : 33

183

8.5a: Observed burglary dwelling flows Destinations ( j )

Predicted Flows ( Sˆ ij )

O R I G I N S (i)

1

2

3

…

…

33

1

38

0

0

2

0

64

0

3

0

0

23

:

… … … …

… … … …

… … … …

… … … … … … …

… … … … … … …

… … … … … … …

: : 33

8.5b: Predicted burglary dwelling flows

Destinations ( j )

Absolute Error

S ij − Sˆ ij O R I G I N S (i)

1

2

3

…

…

33

1

0

0

0

2

2

6

5

3

0

0

1

:

… … … …

… … … …

… … … …

… … … … … … …

… … … … … … …

… … … … … … …

: : 33

TAE

TAE

∑S

ij

− Sˆ ij

ij

8.5c: Total Absolute Error (Sum of Residual) Figure 8.5: Burglary dwelling flows to be compared in the calibration process


8.3.5

184

Goodness-of-fit Statistic:

The assessment of a model’s ability to replicate a known dataset is an important part of model building. Model evaluation consists of measuring the accuracy with which a set of predicted data replicates a set of known data. Many goodness-of-fit statistics have been used for this purpose. However, for the spatial interaction model, Knudsen and Fortheringham (1986) examined the various statistics for comparing observed and predicted spatial interaction matrices. They found that for analysing the performance of two or more models in replicating the same dataset, the most appropriate statistic appears to be the Standardised Root Mean Square Error (SRMSE). SRMSE is a useful measure of model performance in predicting the observed values and it will be used as a method of comparing model results in this chapter to select the best model. It should be noted that the SRMSE is used in two different ways in this study: 1) ‘Matrix comparison’ where the whole grid of 33 x 33 ward relationships are tested 2) ‘Total inflow (by ward) comparison’ where only the inflows into 33 wards are tested. SRMSE is a general distance statistic which is characterised by functions of S ij - Sˆ ij . For matrix comparison the SRMSE is defined as

∑∑ (S SRMSE =

i

ij

− Sˆ ij ) 2 m × n

j

∑∑ S i

m×n

ij

(8.10)

j

Where S ij is an element of the matrix of observed flows

Sˆ ij is an element of the matrix of predicted flows m and n are matrix dimensions The differences are squared to avoid summing positive and negative differences. This statistic has a lower limit of ‘zero’ indicating a perfect fit. It has been noted that SRMSE should only be used when

∑∑ S i

interaction modelling.

j

ij

= ∑∑ Sˆij but this condition is usually met in spatial i

j


185

When comparing the result of ‘inflow burglary dwelling’ at ward level the SRMSE will be given by

SRMSE =

∑ (y

i

− yˆ i ) 2 n y

(8.11)

Where yˆ i is the predicted value of burglary at ward i

yi is the observed value of burglary in ward i y is the mean value of yi n

8.3.6

is the number of wards (which in this case is 33)

Model Summary

In practice there are many experiments which can be done including those mentioned in § 8.3.3. This section summarises 4 main models. 1) Model 1: The first estimation. Wj=

Percentage of rented tenure*percentage of student

Alpha= 0.25 Beta= 0.52 2) Model 2:

W j =σ jγ j : add deprivation and number of car factors to Wj of Model 1 Alpha= 0.25 Beta= 0.52 3) Model 3: Wj=

Predicted burglary dwelling from regression model

Alpha= 0.6 Beta= 0.56 4) Model 4: Wj =

Predicted burglary dwelling from inflow regression model

Alpha= 1.26 Beta= 0.52 5) Model 5: Wj =

Recorded burglary dwelling

Alpha= 0.77


186

Beta= 0.54 Note that the alpha and beta values of each model are different. They come from the ‘Spatial Interaction Calibrator’ program explained in § 8.3.4. Different Wjs give different alpha and beta values. The reason for using the regression equations to derive the attractiveness of the areas in models 3 and 4 is that crime will largely reflect the attractiveness of victims (though see § 8.3.2). The regressions can therefore be used to predict offender flows and crime. In model 5, however, the recorded burglary dwelling is used as attractiveness for comparison because it sheds light on current conditions. Table 8.12 shows the SRMSE values of the 5 models described above. There are two main comparisons in the table as mentioned earlier: ‘matrix comparison’ and ‘total inflow comparison (by ward)’. As can be seen the SRMSE values are very high for the matrix comparison, partly because of large numbers of cells without any interactions. Note that there are 1,089 cells in total for the burglary dwelling flows in Leeds but there are only 519 cells that are active (have a non-zero flow). This affects the mean value of observed counts used in equation (8.10). Therefore, another SRMSE value is included which calculates using only active cells. Model 2 shows the best result according to the SRMSE values of both the ‘matrix comparison’ and ‘total inflow comparison (by ward)’. Table 8.13 shows the predicted burglary dwelling flows from the best model. There are 492 (highlight in green) of 1,089 cells that have a perfect fit matching the exactly the number of observed flows. Table 8.14 compares the observed total inflow for burglary dwelling with the ‘first estimation’ (attractiveness = percentage of rented tenure*percentage of student) and the ‘best estimation’ (attractiveness = percentage of rented tenure*percentage of student multiplied by scaled deprivation and number of car). Overall the best prediction has greatly improved. As can be seen the average relative error is 24.87% compared with 34.92% for the ‘first estimation’. A large number of wards (25 wards) have improved, though there is poorer performance in 9 wards.


187

Table 8.12: Model summary Matrix Comparison

Model Description

SRMSE

TAE 1089 cells

Inflow Comparison (by ward) SRMSE

TAE by Ward

1089 cells

Active Cells

1.3192

0.6287

2886

0.3909

1524

1.3004

0.6197

2822

0.2817

1002

1.3839

0.6595

2904

0.3991

1528

1.3321

0.6349

2855

0.3437

1316

1.3821

0.6587

2956

0.4078

1521

st

1 Estimation: Wj = % of Rented * % of Student 1

Alpha=0.25 Beta=0.54

W j =σ jγ j : add deprivation and number of car factors to Wj of Model 1 2 Alpha=0.25 Beta=0.52 Wj = Predicted burglary dwelling from original regression 3

Alpha= 0.6 Beta=0.56 Wj = Predicted burglary dwelling from inflow regression

4

Alpha=1.26 Beta=0.52 Wj =Recorded Burglary dwelling

5

Alpha=0.77 Beta=0.54

Note:

Matrix comparison: n=1,089; Active cells= 519 Inflow comparison: n=33


188

Table 8.13: Predicted burglary dwelling flows

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Destination j

BurglaryDwelling Prediction Origin i Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton

44 269 28 107 281 276 276

City and Holbeck

365 0.021

Cookridge Garforth&Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley Inflow

144 36 82 330 156 11 232 436 100 46 20 109 90 12 8 19 237 41 35 210 299 132 12 124 161

Oi

Ai

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Out and Hoookridgh&SwHalton Harehileading orsforHunsleKirkstaMiddletooortowrley Norley SoNorth& Whadsey Ndsey So hmond Rothweoundh eacroniverseetwo Wetherb WhinmoWortley Flow Ø eborouArmleyck & KBeestoBramlermantopel Alle Wj Ö 25.3 14.4 19.6 10 16.1 9.9 12.6 12.2 23.7 19.5 18.2 11.2 38.1 31 8.5 19.9 13.2 20 19.5 21.5 24.8 26.5 20.5 17.5 9.0 17.7 23.3 9.2 17.7 28.4 31.1 13.2 14.7 0.034 38 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 2 1 0 0 0 0 0 0 0 0 0 0 44 0.016 0 64 0 4 18 1 4 9 5 0 0 1 43 8 1 35 1 4 5 1 1 1 5 8 1 0 1 0 10 19 0 0 18 269 0.041 0 0 23 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 28 0.028 0 4 0 30 1 1 1 12 0 0 1 1 7 0 6 3 6 1 6 6 0 0 0 1 2 1 1 0 5 2 0 0 6 107 0.018 1 19 0 1 83 0 2 3 9 0 0 1 21 32 0 28 0 3 3 0 1 2 22 15 0 0 1 0 4 20 0 0 9 281 0.023 0 2 2 2 1 63 12 6 1 2 20 35 14 0 5 3 2 9 1 1 2 0 0 0 18 3 21 19 17 4 0 7 1 276 0.016 0 5 0 2 2 6 54 6 5 0 2 14 48 2 2 10 1 32 1 0 4 1 1 1 3 0 18 4 31 18 0 2 2 276 0.019 0.039 0.030 0.020 0.011 0.018 0.032 0.013 0.036 0.017 0.031 0.034 0.029 0.029 0.025 0.028 0.028 0.039 0.021 0.027 0.015 0.013 0.031 0.037 0.024

0

18

1 2 0 0 0 0 0 3 0 6 0 0 0 3 1 33 0 1 0 1 0 0 0 1 0 0 0 0 0 0 0 1 0 2 0 0 0 0 0 1 0 9 0 4 0 0 0 0 0 15 43 193

0

24

5

6

11

93

2

0 0 3 0 2 0 66 1 0 0 0 0 0 0 3 0 0 4 1 1 0 1 2 1 32 24 7 2 0 1 3 1 7 4 3 0 0 1 0 0 0 1 0 16 1 7 4 16 1 0 2 25 1 9 7 15 0 6 0 1 0 3 0 0 0 0 1 4 0 2 0 1 0 0 0 0 0 0 3 0 0 0 1 0 0 0 0 0 1 0 3 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 0 0 0 0 1 6 1 21 7 11 1 0 0 0 0 0 0 0 0 0 0 1 2 0 0 3 1 0 18 7 2 1 0 4 3 7 23 16 3 0 0 4 0 3 1 10 0 0 0 0 0 0 0 6 0 0 4 2 0 0 0 6 7 0 1 6 1 43 114 163 178 182 206 134

1

3

7

48

2

12

15

7

7

0 0 0 9 12 0 6 0 5 27 3 0 0 0 0 0 0 0 5 45 2 1 0 1 0 1 1 1 10 74 24 1 4 5 1 20 0 0 2 68 2 1 14 0 7 0 0 0 0 6 0 0 0 0 3 7 5 10 0 63 3 22 2 0 0 2 86 18 1 114 1 11 1 1 1 2 0 7 1 48 0 0 0 1 6 0 0 2 0 16 0 0 0 0 0 0 0 0 0 0 0 0 1 0 2 0 8 0 0 0 1 2 1 0 1 0 5 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 1 0 1 0 0 4 17 14 11 0 16 3 6 4 3 2 0 0 0 1 0 2 0 0 1 2 2 0 0 0 0 3 2 17 19 6 0 2 1 1 7 0 2 11 59 2 4 13 1 14 0 0 1 21 5 0 13 0 7 0 0 0 0 0 0 0 0 0 1 9 4 1 0 0 0 0 2 0 0 1 11 3 1 8 1 1 52 142 201 501 102 128 279 110 164

7

5

1

0

2

0 0 0 0 0 0 1 1 1 0 0 0 3 8 2 1 3 13 0 0 12 1 7 79 0 0 0 0 0 0 1 0 1 2 0 1 0 0 0 0 1 1 0 0 0 0 0 0 14 3 69 123

4 0 0 4 1 0 0 2 0 2 0 0 64 0 0 0 1 0 1 3 2 2 0 2 0 96

7 0 0 0 0 0 0 2 0 0 0 0 1 9 0 0 0 0 0 0 0 1 0 0 0 28

2 0 0 0 1 1 0 7 0 0 0 0 0 0 4 2 0 0 0 0 1 1 0 0 4 55

3

9

1 0 0 0 0 3 0 11 1 1 0 0 1 19 6 1 0 2 0 0 1 0 1 0 0 0 0 0 1 0 9 0 0 61 0 1 0 0 0 6 1 5 1 0 0 0 0 1 10 1 62 147

2

5

39

12

0 1 0 2 20 3 0 0 0 0 3 2 4 1 0 2 38 18 26 8 0 2 0 11 16 0 0 0 0 1 12 3 2 12 2 0 3 1 14 61 6 0 0 2 0 0 4 0 2 4 0 0 0 0 0 1 0 0 1 0 0 5 0 1 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 7 7 16 3 28 0 0 0 0 0 17 2 1 1 1 31 52 6 2 1 9 3 81 15 0 2 0 4 49 0 0 0 0 0 1 12 11 1 1 0 0 0 4 4 74 185 127 291 268

No Note: Wj = (Percentage of rented tenure and percentage of student)* deprivation and number of car factors; Oi = Number of offender committed burglary dwelling is cell that exactly matches to the observed count (492 cells of 1,089 cells)

2

0

1

15

365

0 0 1 144 0 0 0 36 0 4 0 82 0 7 1 330 0 0 2 156 0 0 0 11 0 1 4 232 0 0 9 436 0 0 1 100 0 0 0 46 0 0 2 20 0 0 2 109 1 1 0 90 0 0 0 12 0 0 0 8 0 0 1 19 0 3 2 237 0 0 0 41 0 1 0 35 1 19 0 210 0 1 4 299 0 0 1 132 12 0 0 12 2 61 0 124 0 0 56 161 16 110 138 4,728


189

Table 8.14: Predicted inflows burglary dwelling by ward Ward


Total

Observed Inflow Burglary

75 127 68 187 149 170 151 310 121 53 62 162 443 124 136 236 138 162 74 118 169 48 51 39 150 114 205 157 272 195 30 115 117 4,728

First Estimated PredictedFlow

40 206 33 148 192 242 216 276 101 39 76 264 397 55 182 309 105 112 44 89 74 20 36 51 205 53 97 186 412 216 13 112 126

4,728

Residual

35 79 35 39 43 72 65 34 20 14 14 102 46 69 46 73 33 50 30 29 95 28 15 12 55 61 108 29 140 21 17 3 9 TAE=1,524

Best Estimated Error (%)

PredictedFlow

43 193 43 114 163 178 182 206 134 52 142 201 501 102 128 279 110 164 69 123 96 28 55 62 147 74 185 127 291 268 16 110 138

46.29 62.01 51.69 21.10 28.91 42.62 43.29 11.07 16.79 26.31 22.96 63.04 10.33 55.34 33.84 30.97 23.93 30.90 40.99 24.91 56.29 57.43 29.91 31.41 36.43 53.37 52.68 18.76 51.58 10.61 56.04 2.65 7.95 Average= 34.92

Source: West Yorkshire Police and spatial interaction model for burglary dwelling.

4,728

Residual

32 66 25 73 14 8 31 104 13 1 80 39 58 22 8 43 28 2 5 5 73 20 4 23 3 40 20 30 19 73 14 5 21 TAE=1,002

Error (%)

Improve ?

43.32 52.13 36.35 39 9.7 4.47 20.71 33.44 10.6 1.81 129.54 24.19 13.17 17.35 5.56 18.33 20.16 1.24 6.89 4.57 43.44 40.82 6.89 58.95 1.77 35.02 9.59 19.06 7.09 37.6 45.19 4.78 17.96 Average= 24.87

9 9 9 ° 9 9 9 ° 9 9 ° 9 ° 9 9 9 9 9 9 9 9 9 9 ° 9 9 9 ° 9 ° 9 ° ° 9


190

8.4 Concluding Comments This chapter started with an explanation about the movement of offenders. As found in previous studies the movements of offenders are relative short. It was found that offenders commit crimes close to where they live for most crime types except fraud and forgery. The distance decay principle matches the pattern that the number of burglary dwelling incidents that an offender commits drops steadily with an increase in distance from the offender’s residence. Generally, an area with high offender rates tends to have high crime rates while an area with low offender rates tends to have low crime rates. However, an area with low offender rates may have high burglary dwelling rates if it is very attractive for burglars. Movements of offenders can be seen as spatial interactions. Plotting the origin and destination of offenders creates a flow matrix. Inflow/outflow ratio and degree of selfcontainment give a picture of the attractiveness of an area. It can be argued that there is a high negative correlation between inflow/outflow ratio and degree of self-containment. Horsforth, Roundhay and Pudsey North wards show that they have high level of inflow/outflow ratio and have very low degrees of self-containment. This shows that these areas are very attractive for burglary. Most offenders committing burglary dwelling in these areas come from other wards. The spatial interaction model for burglary dwelling was formulated to replicate offender committed burglary dwelling flows in Leeds. There are large flows between some areas but small or no flows between others. The factors that determine the magnitude of spatial interaction between where offenders live and where the crimes occur are attractiveness of the destinations and distance between the two areas. It has been found that some areas are very attractive for offenders committing burglary dwelling but some are not.

The

characteristics of destinations that best replicate the known interaction are the 1) proportion of rented tenure type and students 2) scaled deprivation and number of cars in the area. The model can not only be used to explain spatial flows but also can be used to predict the consequences of changes in the conditions generating them. It will thus be used for simulation and prediction purposes in the next chapter.

Chapter 9- What-if Analyses

191

Chapter 9 What-if Analyses 9.1 Introduction 9.2 Policy and Scenario Issues 9.3 Modelling the Neighbourhood Renewal Strategy 9.3.1 Scenario 1: Economic Activity Changes in Gipton NRA 9.3.2 Scenario 2: Economic Activity Changes in Harehills NRA 9.3.3 Scenario 3: A Reduced Number of Offenders Committing Burglary Dwelling in the Gipton NRA 9.3.4 Scenario 4: A Reduced Number of Offenders Committing Burglary Dwelling in the Harehills NRA 9.4 Concluding Comments

9.1 Introduction One of the major advantages of the spatial microsimulation model is the ability to perform what-if analysis. It can be used to conduct policy simulations and for forecasting. In particular, spatial microsimulation models can be used to estimate the impacts of different policy scenario changes and their effects. This chapter aims to show how SimCrime, a spatial microsimulation for crime in Leeds, can be used for what-if analysis. In developing scenarios, this study focuses on those factors that are most likely to have a major impact on crime. Section 9.2 describes policy and scenario issues in Leeds, specifically focussing on the Neighbourhood Renewal Strategy. Section 9.3 gives examples of how SimCrime was used in order to perform what-if analysis. Changes in input parameters provide a new estimation of the likelihood of being the victim of crime. In particular, the impacts of socioeconomic changes on crime victimisation at output area level are illustrated in § 9.3.1 and 9.3.2. Plainly changes in socio-economic structure also mean changes in the number of offenders with a subsequent increase or decrease in criminal flows to different areas. These changes lead to further alterations in the number of crimes in different areas. Sections 9.3.3 and 9.3.4 examine the impact of a change in socio-economic structure on offender flows for burglary dwelling. Finally, § 9.4 offers some concluding comments.

9.2 Policy and Scenario Issues The key variables of SimCrime are important for the prediction of policy impacts. The variables that are focused upon are listed in Table 9.1. Spatial microsimulations, such as SimCrime, become policy relevant when the variables within them can be related to the targets of a specific policy scheme. It is not always the case that the variables that give a good current population distribution are those which are easily related to policy. However in SimCrime there is a good overlap. For example, SimCrime can be used to examine the impacts of the Leeds Neighbourhood Renewal Strategy on potential crime activity.


192

Table 9.1: SimCrime attributes Micro-unit attributes of SimCrime Location (place of residence) at output area level Age Sex Living arrangement Economic activity Tenure type Car or van availability Socio-economic classification

In January 2001 the Prime Minister, Tony Blair, launched the initiative ‘A New Commitment to Neighbourhood Renewal: A National Strategy Action Plan’. The Strategy sets out the government’s plan to narrow the gap between the most deprived neighbourhoods and the rest of the country. The aim is to bring economic prosperity, safe communities, high quality education, decent housing and better health to the poorest parts of the country. At the national level the Action Plan is carried out by the Neighbourhood Renewal Unit (NRU) of the Office of the Deputy Prime Minister (ODPM). At the local level, neighbourhood renewal is the responsibility of Local Strategic Partnerships (LSP). The Leeds Initiative, led by Leeds City Council, is the Leeds Local Strategic Partnership (LSP).

The Leeds Neighbourhood Renewal Strategy, produced by the Neighbourhood and Communities Partnership of the Leeds Initiative, was launched in late 2001. The Strategy aims to narrow the gap between the most disadvantaged neighbourhoods of Leeds and the rest of the city. It has been noted that almost 20% of the Leeds population live in areas officially rated as among the most deprived in the country. Many of these are in the innercity areas, but there are also pockets of deprived neighbourhoods in the more affluent outer areas. They suffer high crime rates, high levels of unemployment, low income, poor housing, poor health, family breakdowns, and low educational achievement (Leeds Initiative, 2005).

The Leeds Neighbourhood Renewal Strategy identifies key areas of the city in which regeneration activity should be focused. Two of these are the Harehills and Gipton areas (Figure 9.1). The aim of narrowing the gap between the communities in these areas and the rest of the population will only be achieved by simultaneously lowering unemployment,


193

reducing crime, improving health, increasing skills and providing better housing and physical environments. The crime aspect of the proposed programme of changes is examined here.

Figure 9.1: Neighbourhood Renewal Areas (NRAs) in Leeds Source: Leeds City Council

The micro-spatial impacts of the policy of Neighbourhood Renewal Areas (NRAs) can be estimated using spatial microsimulation models such as SimCrime. As mentioned in Chapter 7, the risk of burglary dwelling victimisation is related to socio-economic circumstances and, generally, economically disadvantaged households are at a higher risk. If the population changes (discussed by the plan) are carried out in the microsimulated population, it is then possible to use the new relationships between circumstances and crime to predict future crime levels. Broadly, one would expect, in areas where socio-economics are improved, that there would be a decrease in victimisation rates.

Changes in socio-economic structure cause changes in the chance of becoming a victim of crime but also in the number of offenders, which, in turn, can be related to changes in the number of victims and number of crimes (Figure 9.2). There is, then, the potential for two crime predictions: the first is to predict victims of crime, the second is to predict crime locations. The first proceeds by simulating the new population forecast under the plan, and attaching British Crime Survey (BCS) statistics to give general victim numbers. The second proceeds by then taking this population (without the crime data playing a part) and using it to predict both offenders and the attractiveness of the renewed areas, which can then be used in the spatial interaction model to predict the crime locations. It should be noted that both methodologies allow for the creation of new crimes and an analysis of their predicted locations.


194

Changes in Socio-economic

Increase/decrease in number of victims

Increase/decrease in number of offenders

Changes in Number of crimes/Crime rates

Figure 9.2: Effects of socio-economic changes on crime

Gipton and Harehills are the major Neighbourhood Renewal Areas in Leeds where unemployment and crime rates are greater than the Leeds average. The Neighbourhood Renewal Strategy aims to induce changes to the employment structure. A reasonable question would therefore seem to be ‘what will happen to crime rates if there is higher employment and less unemployment?’ It might, for example, be expected that a lowering in unemployment could mean a lower number of offenders, with subsequent decreases in the flow of criminals to other areas.

Figure 9.3 shows the estimated spatial distribution of those people living in the Gipton and Harehills NRAs which will be affected by the NRA initiative. The group dealt with, both in the figure and the analysis following, are people aged 16-74 who are economically active (employed, unemployed, or full-time students). Figure 9.4 shows the estimated spatial distribution of the subset of these people who are unemployed, and therefore likely to be affected the most.


195

People Age 16-74 Economically Active

Gipton N RA Harehills NRA People Aged 16-74 Active 24 - 94 95 - 129 N 130 - 161 162 - 197 W E 198 - 357 S

Source: SimCrime Figure 9.3: Spatial distribution of people aged 16-74 and economically active by output area

People Aged 16-74 Unemployed

Gipton NRA Harehills NRA People Aged 16-74 Unemployed 0-4 N 5-9 10 - 14 W E 15 - 20 S 21 - 35

Source: SimCrime Figure 9.4: Spatial distribution of unemployed people aged 16-74 by output area


196

9.3 Modelling the Neighbourhood Renewal Strategy The Gipton and Harehills NRAs will be used in what-if analyses that change employment in the following sections. Specifically, § 9.3.1 and 9.3.2 show the impact of socio-economic changes on the number of victims of burglary dwelling, while § 9.3.3 and 9.3.4 deal with the spatial interactions between offenders’ home addresses and where they commit crimes.

9.3.1

Scenario 1: Economic Activity Changes in Gipton NRA

The Gipton NRA is located east of the City Centre (Figure 9.5), and lies within the Burmantofts and Harehills electoral wards. This area has gained a poor reputation and has a predominance character of higher than average levels of deprivation. The 2001 Census statistics recorded 11,868 people (1.7% of the population of Leeds) living in the Gipton NRA in 4,638 households. Fifty-eight per cent of households in the Gipton NRA rent their homes, whether from the council or from private or social landlords, against only 38% for Leeds as a whole.

Some 56.72% of the population aged 16-74 is economically active, compared with 66.37% for Leeds as a whole. The Gipton NRA has 7,982 people in this age group of whom 4,527 are economically active. If the area met the city average, there would be 5,297 economically active people, i.e. about 770 more than at present. Economically active people fall into three groups:

Employed

Unemployed

Full-time students


Figure 9.5: Gipton Neighbourhood Renewal Area Source: Leeds City Council

197


198

Scenario 1 is to change the economic activity of the people aged 16-74 in the Gipton NRA to match the Leeds average, with the crimes being predicted through changes in victim numbers, not offenders. The figures shown in Table 9.2 illustrate how labour market conditions would have to change if Gipton NRA were to correspond to the Leeds average, while Table 9.3 summarises the number of people in each category. Note that the total number of economically active people in the Gipton NRA under Scenario 1 is derived from the number of people that already existed plus the increase/decrease required to correspond to the Leeds average. For example, the total number of employed people aged 16-74 under Scenario 1 is 4,754 (3,773+981).

Table 9.2: Economic activity in Gipton NRA and in Leeds

Status

Percentage of population aged 16-74 Gipton %

Employed 47.27 Unemployed 6.53 Full-time student 2.92 56.72 Total Source: Leeds City Council (2004a)

Leeds %

Number of population aged 16-74 Gipton Nos.

59.56 3.35 3.46 66.37

3,773 521 233 4,527

Increase required to correspond to Leeds average Nos. % 981 -254 43 770

26.01 -48.72 18.37 17.01

Table 9.3: Scenario 1 Scenario1: Total economically active people in the Gipton NRA under Scenario 1 Economic Activity of Gipton NRA Employed Unemployed Full-time student

Number 4,754 267 276

Figure 9.6 depicts the estimated spatial distribution of the number of victims of burglary dwelling in the Gipton NRA resulting from the changed economic activity under Scenario 1. As can be seen, most output areas would have a smaller number of victims. However, in 12 output areas the victims of burglary dwelling would increase and in two output areas they would remain the same (Figure 9.7). Overall, burglary dwelling decreases around 20% (from 308 incidents to 247 incidents) in the Gipton NRA under Scenario 1, assuming no change in offenders.


199

Scenario 1

Figure igure 9.6: How number of victims of burglary dwelling in the Gipton NRA would change under Scenario 1

Roundhay Seacroft

Harehills Gipton NRA

Burmantofts

Figure 9.7: Change in the number of victims of burglary dwelling in the Gipton NRA


Figure 9.8: Harehills Neighbourhood Renewal Area

200


9.3.2

201

Scenario 2: Economic Activity Changes in the Harehills NRA

The Harehills Neighbourhood Renewal Area (NRA) (Figure 9.8) is home to some 19,694 people, and lies within the Harehills, Burmantofts, and University wards. From the 2001 Census, it is clear that this area is one of the most deprived in Leeds. The Harehills NRA has 2.8% of the population of Leeds, but only 1.9% of the white population of Leeds, with 44% of the Bangladesh community, 18% of the Pakistani community and 15% of the ‘black’ community. Sixty-seven per cent of households in the Harehills NRA rent their homes, whether from the council or from private or social landlords, against only 38% for Leeds as a whole.

About 57% of the population aged 16-74 is economically active compared with 66.37% for Leeds as a whole. The Harehills NRA has 14,036 people in this age group, of whom 7,999 are economically active. If the area corresponded to the city average, there would be 9,315 economically active individuals in this group (i.e. 1,316 more than at present). Table 9.4 shows how the economic activity would have to change if the Harehills NRA were to correspond to the Leeds average. The most notable thing is the large decrease (56.30%) in unemployment that would be required. The required increase in employment is 28.19%. We use these changes in Scenario 2, and the resulting changes are shown in Table 9.5.

Table 9.4: Economic activity in Harehills NRA and in Leeds Percentage of population aged 16-74 Status

Harehills %

Employed 46.47 Unemployed 7.66 Full-time student 2.86 Total 56.99 Source: Leeds City Council (2004b)

Leeds % 59.56 3.35 3.46 66.37

Number of population aged 16-74 Harehills Nos. 6,522 1,075 402 7,999

Increase required to correspond to Leeds average Nos. % 1,838 -605 83 1,316

28.19 -56.30 20.64

16.45

Table 9.5: Scenario 2 Scenario 2: Total economically active people in the Harehills NRA under Scenario 2 Economic Activity of Harehills NRA Employed Unemployed Full-time student

Number 8,360 470 485


202

The Harehills NRA is bigger than the Gipton NRA in size, the number of output areas, and the number of individuals. Figure 9.9 illustrates the estimated spatial distribution of the victims of burglary dwelling in the Harehills NRA resulting from the changed economic activity under Scenario 2. It is interesting to note that there is no improvement in terms of burglary dwelling victimisation, indeed, things get worse. Under Scenario 2, the number of victims of burglary dwelling in the Harehills NRA would change from 531 to 595, a 12 per cent increase.

Scenario 2

Figure 9.9: How the number of victims of burglary dwelling in the Harehills NRA would change under Scenario 2.

It is always assumed that efforts to reduce crime in disadvantaged areas should aim at increasing local socio-economic status. Logically we might expect decreases in victimisation and in the number of crimes in areas where socio-economic conditions have improved. However, in some cases it may be that the areas in question become more attractive (improving the socio-economic structure attracts more affluent residents) and that causes a higher risk of victimisation. Moreover, the process of socio-economic improvement brings with it social instability. Improved economic conditions are not necessarily associated with lower levels of crime. If disparities within a region remain large, improvements in wealth and income may encourage the less advantaged groups to commit crimes against the more affluent groups (Glaeser and Sacerdote, 1999). It seems that this is the condition present in the Harehills NRA, and these results point towards future difficulties that may arise under the NRA scheme, if carried out in Harehills in isolation.


203

Chapel Allerton Harehills

University

Burmantofts

City and Holbeck Richmond Hill

Figure 9.10: The changing number of victims of burglary dwelling in the Harehills NRA under Scenario 2

As can be seen in Figure 9.10, there are a large number of output areas that would have more victims of burglary dwelling if Scenario 2 occurred. Output areas that would have an increase in the number of victims of burglary dwelling are mostly concentrated within pockets of the Harehills NRA itself and particularly those output areas on the edge of the NRA. This is likely to be happening in the model because the Harehills NRA improvement attracts offenders living in surrounding wards such as Chapel Allerton, Richmond Hill and City and Holbeck to come into the Harehills NRA. These wards have a very high number of offenders (Table 8.8 in previous chapter). This shows that the chance of becoming a victim of burglary dwelling is higher not only in disadvantaged areas, but also in the areas that are undergoing socio-economic improvement.


9.3.3

204

Scenario 3: A Reduced Number of Offenders Committing Burglary Dwelling in the Gipton NRA

Section 9.3.1 and 9.3.2 have already shown the result of changing the employment status of people in the Gipton and Harehills NRAs on the number of victims of burglary dwelling. It should be noted, however, that changing the employment structure impacts not only on victim numbers but also on offenders. Scenario 3 is therefore to add in these offenders to the previous microsimulation changes and to predict their movements. In this study, it has been found that the ‘number of offenders committing burglary dwelling’ (burglars) has a very high correlation with the ‘number of unemployed people’. The correlation coefficient is high at 0.819 (significant at the 0.01 level). Therefore, following the regression analysis, the ‘number of offenders’ in each area is adjusted using the following simple equation:

Number of burglars in area = -133.336 + 0.529* Number of Unemployed (9.1)

Table 9.6: Scenario 3 Scenario 3: Total offenders in the Gipton NRA under Scenario 3 Gipton NRA Burmantofts Harehills

Number of Burglars 77 108

As with § 4.2.4, it should be noted that ‘offenders’ in this sense are essentially offenceevents as offenders that commit multiple crimes are listed multiple times. This does not cause problems because the numbers are being calculated by regression. It is not, here, important which people are doing the offending, just total numbers. Table 9.6 shows the number of offenders committing burglary dwelling (burglars) calculated from the above equation. In line with the spatial interaction model described in Chapter 8, the attractiveness is derived from the ‘percentage of rented tenure * percentage of students’ multiplied by ‘scaled attractive factors’ (deprivation and number of car). Under Scenario 3 there is an 18.37 % increase in full-time students in the Gipton NRA (see Table 9.2). Therefore, the attractiveness factor in the spatial interaction model is also adjusted.

The impact of the changes in the number of offenders and attractiveness factors can be assessed in terms of the changes in offender flows. Given the lower number of burglars in the Burmantofts and Harehills areas of the Gipton NRA, and looking at the flow (travel to crime) patterns, it is reasonable to expect that Harehills and Burmantofts, as well as the surrounding areas, will have lower numbers of burglary dwelling. In particular, for


205

Roundhay the previous chapter shows that 23.42% of offenders who commit burglary dwelling in Roundhay live in Harehills, while 14.63% come from Burmantofts. Therefore, it can be reasonably expected that Roundhay will be strongly affected by the improvement in employment.

Figure 9.11 shows the decrease in burglary dwelling in Leeds under Scenario 3. The overall effect is that every ward would have less burglary dwelling. Harehills and Burmantofts themselves decrease more than 30%. As for surrounding areas, the largest effects are in Roundhay and Seacroft, with decreases in burglary dwelling of 22.75% and 20.93% respectively.

Decrease in 'burglary dwelling' under scenario 3

Roundhay

Gipton NR A Decrease in burglary dwelling Less than 1% 1-5 % N 5-10% 10-20% W E 20-30% S More than 30%

Figure 9.11: Decrease in burglary dwelling under scenario 3 Table 9.7 summarises offender flows under Scenario 3. Columns on the left show the number of offenders travelling from Burmantofts and Harehills to other wards, calculated both the base line model (in Chapter 8) and under Scenario 3. Columns on the right show ‘total inflow’ into each ward of burglary dwelling offenders. Again there is a comparison between the baseline spatial interaction model and Scenario 3. Each category shows ‘number’ and ‘percentage share’. Note that burglary dwelling flows in Leeds is initiated within a closed system. Therefore, flows within this system are generated by area conditions only, and without external inputs. Once the number of offenders is lowered, it is no surprise that each ward has a decrease in the total number of inflows. The reason for the expression of the figures as the ‘percentage share’ is that it allows comparison as a proportion of the total flows, useful when comparing results within this closed system.


206

Table 9.7: Estimated burglary dwelling flows under Scenario 3 Number of flows From Burmantofts

Number of flows From Harehills

Destination Ward Baseline SI Model Scenario3 Baseline SI Model Scenario3 Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth & Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

Total

0 2 2 2 1 63 12 6 1 2 20 35 14 0 5 3 2 9 1 1 2 0 0 0 18 3 21 19 17 4 0 7 1 276

0 1 1 1 0 18 3 2 0 1 5 10 4 0 1 1 1 3 0 0 1 0 0 0 5 1 6 5 5 1 0 2 0 77

0 3 1 2 1 32 24 7 2 1 10 74 24 1 4 5 1 20 1 1 4 0 0 0 11 2 38 18 26 8 0 7 1 330

0 1 0 1 0 11 8 2 1 0 3 25 8 0 1 2 0 6 0 0 1 0 0 0 3 1 12 6 8 3 0 2 0 108

Total inflow of burglary dwelling to each ward Number Scenario3 Change Baseline SI Model Number PercentageShare Number PercentageShare 43 193 43 114 163 178 182 206 134 52 142 201 501 102 128 279 110 164 69 123 96 28 55 62 147 74 185 127 291 268 16 110 138 4,728

0.90 4.09 0.92 2.41 3.46 3.76 3.86 4.36 2.83 1.10 3.01 4.26 10.60 2.17 2.72 5.91 2.33 3.47 1.46 2.61 2.02 0.60 1.15 1.31 3.12 1.57 3.92 2.69 6.16 5.67 0.35 2.32 2.92 100

42 190 41 111 162 116 156 197 131 49 120 132 474 101 122 274 108 143 68 122 91 28 54 62 126 71 143 100 260 259 16 99 136 4,307

0.99 4.40 0.95 2.57 3.76 2.69 3.63 4.57 3.05 1.15 2.79 3.06 11.01 2.36 2.83 6.35 2.50 3.33 1.58 2.84 2.12 0.65 1.26 1.43 2.93 1.64 3.32 2.33 6.05 6.02 0.37 2.30 3.16 100

-0.09% -1.82% -5.25% -2.97% -0.80% -34.81% -14.29% -4.66% -1.90% -5.16% -15.47% -34.58% -5.39% -1.02% -5.12% -2.03% -2.18% -12.67% -1.14% -0.75% -4.32% -1.04% -0.65% -0.74% -14.40% -4.78% -22.75% -20.93% -10.60% -3.37% -2.76% -9.47% -1.28% -8.91%

Percentage Share Change 0.09 0.32 0.04 0.16 0.31 -1.07 -0.23 0.20 0.22 0.05 -0.22 -1.20 0.41 0.19 0.11 0.45 0.17 -0.14 0.12 0.23 0.10 0.05 0.10 0.12 -0.19 0.07 -0.60 -0.35 -0.11 0.35 0.02 -0.01 0.24 0.00


207

It is interesting to note how offenders’ flows patterns would change under Scenario 3. The main trend is a decline in the total inflows (Table 9.7). The ‘percentage share’ of burglary dwelling inflow would be decreased in Harehills (-1.20), Burmantofts (-1.07), Roundhay (0.60), and Seacroft (-0.35). The impact on the rest of the system is also important. It has been found that Chapel Allerton, Halton, Richmond Hill, Moortown, University, and Whinmoor also decrease in ‘percentage share’.

The ‘percentage share’ change under scenario 3 is demonstrated in Figure 9.12. It shows that those areas in close proximity to the Gipton NRA would have less ‘percentage share’. Although the amount of burglary dwelling would decrease across the city, those wards further from the Gipton NRA would have a higher ‘percentage share’ under Scenario 3.

Change in 'Percentage Share' Under Scenario3

Gipton NRA 'Percentage Share' Change Decrease more than 1 Decrease 0-1 Increase 0.01 - 0.45 N W

E S

Figure 9.12: ‘Percentage share’ change under Scenario 3


9.3.4

208

Scenario 4: A Reduced Number of Offenders Committing Burglary Dwelling in the Harehills NRA

Here, under Scenario 4, the number of burglars are changed in the same manner as in § 9.3.3. The numbers of burglars living in the wards lying in the Harehills NRA are given in Table 9.8. Moreover, under Scenario 4 full-time students in the Harehills NRA are increased by 20.64% (see Table 9.4).

Table 9.8: Scenario 4 Scenario 4: Total offenders in the Harehills NRA under Scenario 3 Harehills NRA Burmantofts Harehills University

Number of Burglars 46 72 70

Figure 9.13 shows the decrease in burglary dwelling in Leeds under Scenario 4. As with Scenario 3, the overall effect is that every ward should see a decline in burglary dwelling. Harehills and Burmantofts decline by more than 40% while the University ward decline by 30%. Surrounding areas such as Roundhay, Chapel Allerton, and Seacroft would be the most affected, with a decrease in burglary dwelling of 30.05%, 26.48%, and 25.91% respectively.

Decrease in 'burglary dwelling' under scenario 4

Roundhay

Harehills NRA Decrease in burglary dwelling Less than 5 % 5-10 % 10-20 % N 20-30 % W E 30-40 % S More than 40 %

Figure 9.13: Decrease in burglary dwelling under Scenario 4


209

Table 9.8: Estimated burglary dwelling flows under Scenario 4 Number of flows From Burmantofts

Number of flows From Harehills

Number of flows From University

Destination Ward Baseline SI Model Scenario4 Baseline SI Model Scenario4 Baseline SI Model Scenario4 Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth & Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

Total

0 2 2 2 1 63 12 6 1 2 20 35 14 0 5 3 2 9 1 1 2 0 0 0 18 3 21 19 17 4 0 7 1 276

0 0 0 0 0 11 2 1 0 0 3 6 2 0 1 0 0 2 0 0 0 0 0 0 3 1 3 3 3 1 0 1 0 46

0 3 1 2 1 32 24 7 2 1 10 74 24 1 4 5 1 20 1 1 4 0 0 0 11 2 38 18 26 8 0 7 1 330

0 1 0 0 0 7 5 1 1 0 2 17 5 0 1 1 0 4 0 0 1 0 0 0 2 0 8 4 6 2 0 1 0 72

0 9 0 4 3 7 23 16 3 0 2 11 59 2 4 13 1 14 1 1 2 0 1 1 5 1 9 3 81 15 0 1 4 299

0 2 0 1 1 2 5 4 1 0 1 3 14 0 1 3 0 3 0 0 0 0 0 0 1 0 2 1 20 3 0 0 1 70

Total inflow of burglary dwelling to each ward Number Percentage Scenario4 Change Share Baseline SI Model Number Percentage Share Number Percentage Share Change 43 193 43 114 163 178 182 206 134 52 142 201 501 102 128 279 110 164 69 123 96 28 55 62 147 74 185 127 291 268 16 110 138 4,728

0.90 4.09 0.92 2.41 3.46 3.76 3.86 4.36 2.83 1.10 3.01 4.26 10.60 2.17 2.72 5.91 2.33 3.47 1.46 2.61 2.02 0.60 1.15 1.31 3.12 1.57 3.92 2.69 6.16 5.67 0.35 2.32 2.92 100

42 182 40 107 159 99 134 182 128 49 115 110 424 100 118 262 106 129 67 122 89 28 53 61 118 69 130 94 202 246 16 97 133 4,011

1.06 4.54 1.01 2.66 3.97 2.48 3.34 4.54 3.20 1.21 2.87 2.75 10.56 2.48 2.94 6.54 2.65 3.21 1.67 3.03 2.23 0.69 1.33 1.51 2.95 1.73 3.23 2.35 5.04 6.13 0.40 2.41 3.31 100

-0.24% -5.76% -6.45% -6.60% -2.48% -44.00% -26.48% -11.73% -4.19% -6.51% -19.15% -45.26% -15.47% -2.81% -8.22% -6.05% -3.68% -21.57% -3.02% -1.49% -6.45% -2.23% -2.01% -2.37% -19.62% -6.50% -30.05% -25.91% -30.56% -8.37% -3.50% -11.71% -3.84% -15.16%

0.16 0.45 0.09 0.24 0.52 -1.28 -0.51 0.18 0.37 0.11 -0.14 -1.51 -0.04 0.32 0.22 0.63 0.32 -0.26 0.21 0.42 0.21 0.09 0.18 0.20 -0.16 0.16 -0.69 -0.34 -1.12 0.45 0.05 0.09 0.39 0.00


210

Table 9.8 shows the estimated burglary dwelling flows under Scenario 4 in the same format as Table 9.7. The ‘percentage share’ of total inflow of burglary dwelling declines in 10 of 33 wards including Burmantofts (-1.28), Harehills (-1.51), University (-1.12), Roundhay (0.69), Chapel Allerton (-0.51), Seacroft (-0.34), Moortown (-0.26), Richmond Hill (-0.16), Halton (-0.14), and Headingley (-0.04). The same pattern can be found here under Scenario 4. There would be a decline in the ‘percentage share’ of total inflow of burglary dwelling in those areas with a close proximity to the Harehills NRA, while to balance the calculation the rest of the city would see an increase.

Change in 'Percentage Share' Under Scenario 4

Harehills NRA 'Percentage Share' Change Decrease more than 1 Decrease 0-1 Increase 0.01 - N0.63 W

E S

Figure 9.14: ‘Percentage share’ change under Scenario 4


211

9.4 Concluding Comments This chapter has demonstrated the usefulness of the spatial microsimulation model in the evaluation of policies such as the Neighbourhood Renewal Strategy. In particular, SimCrime was used to estimate the impact of changing the socio-economic structure of the Gipton and Harehills NRAs in Leeds, specifically how this changed the number and location of victims of burglary dwelling. Further, examples were given of analyses in which the spatial impacts of a change in the offender flows were investigated using the synthetic populations and the spatial interaction model. When interpreting results from scenarios presented in this chapter, it is important to understand that they do not predict the future crime trend. Instead, they estimate what changes in crime are likely to occur as a result of demographic and socioeconomic changes.

The findings resultant upon the improved socio-economic status of areas under Scenarios 1 and 2 shed new light on the way in which an area’s socio-economic structure affects the risk of victimisation. The risks are not intuitive, indicating the worth in making such a model. The results from Scenario 1 and 2 support the social disorganisation study of Van Wilsem et al. (2006) that victimisation is more likely in disadvantaged neighbourhoods, but may increase in neighbourhoods where socio-economic improvements are taking place.

When an area’s socio-economic conditions, such as employment structure, are changed, this affects the number of offenders and attractiveness of destination areas, resulting in a different pattern of offender flows. The results from Scenarios 3 and 4 show that burglary dwelling in an area is not only dependent upon individual and area characteristics, but also upon the city context. The total inflow to one area is dependent on what is happening in neighbouring areas. Improving the socio-economic structure in the Gipton NRA and Harehills NRA would have a significant impact on the existing offenders’ flows patterns, especially in surrounding areas.

One of the drawbacks of the what-if analysis presented in this chapter is that the validity of the estimated spatial impacts of the Neighbourhood Renewal Area cannot be assessed. However, the results from Scenario 3 and Scenario 4 can be compared with the police recorded crime data for the latest year available to see if the trends they reveal correspond to real data about the areas, which are currently under the Strategy. Note that the spatial interaction model can provide a predictive accuracy of around + 25% on average (Table 8.14 previous chapter); with matches of + 4% for Burmantofts, + 7% for University, and + 24% for Harehills.


212

Table 9.9: Recorded burglary dwelling (2001/02 and 2003/04) Recorded Burglary Dwelling 2001/2002

Ward Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth & Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley Total

Number Percentage Share 295 1.88 380 2.42 203 1.29 287 1.83 469 2.99 612 3.90 729 4.65 632 4.03 426 2.71 216 1.38 287 1.83 702 4.47 1,391 8.86 453 2.89 295 1.88 832 5.30 338 2.15 567 3.61 210 1.34 261 1.66 551 3.51 240 1.53 364 2.32 259 1.65 711 4.53 272 1.73 694 4.42 547 3.49 997 6.35 645 4.11 168 1.07 334 2.13 326 2.08 15,693 100

2003/2004 Number Percentage Share 311 2.25 792 5.73 181 1.31 296 2.14 600 4.34 508 3.67 429 3.10 597 4.32 381 2.75 214 1.55 290 2.10 450 3.25 1,002 7.24 446 3.22 275 1.99 790 5.71 246 1.78 367 2.65 193 1.40 238 1.72 348 2.52 305 2.20 335 2.42 357 2.58 557 4.03 194 1.40 468 3.38 395 2.86 735 5.31 533 3.85 233 1.68 261 1.89 506 3.66 13,833 100

Number Change 5.42% 108.42% -10.84% 3.14% 27.93% -16.99% -41.15% -5.54% -10.56% -0.93% 1.05% -35.90% -27.97% -1.55% -6.78% -5.05% -27.22% -35.27% -8.10% -8.81% -36.84% 27.08% -7.97% 37.84% -21.66% -28.68% -32.56% -27.79% -26.28% -17.36% 38.69% -21.86% 55.21% -11.85%

Percentage Share Change 0.37 3.30 0.01 0.31 1.35 -0.23 -1.54 0.29 0.04 0.17 0.27 -1.22 -1.62 0.34 0.11 0.41 -0.38 -0.96 0.06 0.06 -1.00 0.68 0.10 0.93 -0.50 -0.33 -1.04 -0.63 -1.04 -0.26 0.61 -0.24 1.58 0.00

Table 9.9 summarises the recorded burglary dwelling between 2001/02 and 2003/04. It is interesting to note that recorded burglary dwelling in Leeds has decreased 11.85% within three years. Twenty-four of the thirty-three wards have seen a decrease in the number of burglary dwelling. Changes in recorded burglary dwelling are shown in Figure 9.15 and the change in ‘percentage share’ of recorded burglary dwelling is illustrated in Figure 9.16. The red colour depicts an increase whilst the green colour depicts a decrease. As can be seen, areas close to the Gipton and Harehills NRAs have decreased in the amount of burglary dwelling. This decrease is likely to be a result of the Neighbourhood Renewal Area Strategy, which started in late 2001.


213

Change in 'Recorded Burglary Dwelling' (2001/02 and 2003/04)

N W

E

Harehills NRA Gipton NRA Burglary Changes Decrease more than 30% Decrease 15-30% Decrease up to 15% Increase up to 30% Increase 30-50% Increase more than 50%

S

Figure 9.15: Change in recorded burglary dwelling (2001/02 and 2003/04) Source: Derived from West Yorkshire Police

Change in 'Percentage Share' of Recorded 'Burglary Dwelling' (2001/02 and 2003/04)

Harehills NRA Gipton NRA 'Percentage Share' Decrease Increase N W

E S

Figure 9.16: Change in ‘percentage share’ of recorded burglary dwelling (2001/02 and 2003/04) Source: Derived from West Yorkshire Police


214

It is interesting to compare the ‘percentage shares’ of burglary dwelling under Scenario 3 and Scenario 4 with the real statistics (Table 9.10). As can be seen, they are a good match. Twenty-seven of thirty-three wards show the same trend as the real data. Examining the effects of socio-economic change on crime using such examples as presented in this chapter could, therefore, be an important step in the process of developing crime prevention initiatives.

Table 9.10: Trends of ‘percentage share’ of recorded burglary dwelling by ward (between 2001/02 and 2003/04) Change in ‘Percentage Share’ of burglary dwelling Ward Aireborough Armley Barwick & Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth & Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley & Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley

Recorded Burglary Dwelling

Under Scenario 3

Under Scenario 4

+ + + + + + + + + + + + + + + + + + +

+ + + + + + + + + + + + + + + + + + + + + + +

+ + + + + + + + + + + + + + + + + + + + + + +

Note: + = increase

- = decrease. Highlights are wards with the same trends as the recorded burglary dwelling

Chapter 10- Conclusions

215

Chapter 10 Conclusions 10.1 10.2 10.3 10.4 10.5

Introduction Summary of the Research Findings Evaluation and Limitations of the Research Possibilities for Future Research Concluding Statements

10.1 Introduction This final chapter concludes the thesis by providing reflection on the results of previous chapters and outlining the potential for future work. Section 10.2 summarises how the aims and objectives which were stated in Chapter 1 were achieved, and highlights the main findings made with reference to the research objectives. An evaluation and the limitations of the research are given in § 10.3. The possibilities for future research are detailed in §10.4. The chapter ends with concluding statements in § 10.5.

10.2 Summary of the Research Findings As stated in Chapter 1, the principal aim of this thesis was to investigate the potential of spatial microsimulation for modelling crime. Chapter 1 established that in order to achieve this research aim a number of research objectives needed to be accomplished.

The first objective was to review the geography of crime and crime modelling; this was achieved in Chapter 2. Crime mapping is a key component for studying the geography of crime and the discussion focused on the location of crimes in respect to demographic and socio-economic characteristics. The chapter summarised crime attributes relating to offenders, victims, and offence areas derived from the literature reviewed in Table 2.2. These variables were explored (in Chapter 5) to find out whether or not they are good predictors of crime in Leeds and to assess how these could be linked in a spatial microsimulation. Chapter 2 also reviewed the work on crime modelling. Crime models are generally based on statistical regression approaches. Most of them are based on long-run aggregate relationships between recorded crime and macroeconomic and demographic factors. In the UK, the smallest area of crime modelling to date is at the police force area level.


216

Chapter 3 fulfilled the requirements of objective two: to review microsimulation models and the procedures involved in creating a synthetic population microdata dataset. The chapter introduced microsimulation models and described the difference between static and dynamic, and spatial and aspatial models. Selected microsimulation models were reviewed and the advantages and disadvantages of microsimulation models were summarised. There have been a large number of microsimulation models to date. However, they have not been applied to the study of crime. The chapter then reviewed and compared the two main methods used to create a synthetic population: synthetic reconstruction and combinatorial optimisation. There are some key advantages of combinatorial optimisation using simulated annealing over other methods. Therefore, this method was adopted to create a microdata dataset and was described in detail.

Objective three, to investigate and review the available data for modelling crime in Leeds, was attained in Chapter 4. Four main sources of data were used in this study: the 2001 Census, the 2001/2002 British Crime Survey (BCS), police recorded crime datasets (2000/01, 2001/02, 2002/03, 2003/04), and a known offender dataset (2001-2004). General information on, and limitations of, the datasets were described and discussed. Although the census provides a comprehensive spatial coverage for small geographical areas, it has limitations. The main problem for this study is the discrepancies in census counts between tables because of actions taken to preserve the confidentiality of individuals. The BCS provides a rich source of individual and household information together with crime victimisation data, but is not available for small geographical areas. Police recorded crime data and data on known offenders were also reviewed and their advantages and drawbacks were discussed. Geographical referencing of the latter two datasets provided very useful information on where crimes occur and where offenders live.

Chapter 5 focused on objectives four and five: to explore the geography of crime in Leeds and to study the relationship between crime and its related determinants in the Leeds context. The chapter presented Leeds crime figures and trends. The geographical variations at the police division and ward levels were presented. Different types of crime tend to occur in different types of areas. However, it is clear that most of the crime types clustered in certain areas (close to the city centre). The City and Holbeck ward has the highest number of crimes and crime rates for every type except burglary dwelling. Known offenders and victims in Leeds were also described.


217

Pearson’s correlation coefficients provided a better understanding of the relationship between crime and its determinants. Focus was placed on a range of factors correlated with crime. It was argued that ‘student’ factor is one of the most important for modelling crime. The multiple regression model and correlation coefficients developed in the chapter suggested that ‘age’, ‘sex’, ‘economic activity’, ‘socio-economic classification’, ‘car ownership’, and ‘tenure type’ play an important role for burglary dwelling and need to be included in the spatial microsimulation for modelling crime.

Chapter 6 addressed objective six: to use the knowledge that has been gained in objective 15 to build a static spatial microsimulation model for crime in Leeds, by presenting SimCrime, a spatial microsimulation model for crime in Leeds. The knowledge gained from Chapters 2, 3, 4, and 5 was used to create SimCrime. The model in effect added ‘geography’ to the existing BCS data. The chapter demonstrated how to spatial disaggregate the BCS to output area level, so that the information pertaining to the characteristics of victims of crime can be used to estimate similar information for zones as small as output areas. Specifically, SimCrime combines individual microdata from the BCS with census statistics for smaller areas to create synthetic microdata estimates for small output areas in Leeds using a combinatorial optimisation simulated annealing method.

The output of SimCrime is a list of 514,523 individuals aged 16-74 living in households in Leeds whose characteristics match the characteristics of the 514,523 individuals living in Leeds, as shown in the 2001 Census. Attached to these individuals are attributes including victimisation-related variables from the 2001/2002 British Crime Survey. Moreover, it is possible to estimate the geographical and socio-economic distribution of the people which are unavailable from published sources. Validation is difficult, however, the individual synthetic population was aggregated to output area level and then evaluated in terms of their match to the constraint tables from the census to give a model calibration error. SimCrime constraint variables are age, sex, living arrangement, economic activity, tenure type, car or van availability, and socio-economic classification. It was argued that the constraint tables should be adjusted to minimise discrepancies added by the confidentiality process between the total populations in small areas. The adjustment method proposed in the chapter ensured the constraint tables were more consistent. The quality of the synthetic population is likely to be affected by the size of the sample used as a microdata database, the number of constraint variables, the consistency of constraint tables, and the value of the control parameters of the simulated annealing method.


218

Spatial microsimulation models such as SimCrime provide an ideal framework in which crime victimisation for small areas can be estimated. To accomplish objective seven, to use this model to estimate crime victimisation at ward level, Chapter 7 demonstrated the potential of spatial microsimulation for modelling crime victimisation at the ward level in Leeds. For ease of comparison, the chapter estimated victims of burglary dwelling for which the place of residence and victimisation are the same. The assumption is that if the synthetic population from SimCrime have the same (demographic and socio-economic) characteristics as the population from the British Crime Survey, they will have the same propensity to be a victim of crime. SimCrime estimated that the likelihood of being a victim of burglary dwelling is much higher in inner-city areas, particularly in the Headingley and University wards. However, the model showed that using the information from the BCS alone underplayed the importance of geographical factors. SimCrime also showed the risk for different types of household. It was argued that the proximity to offenders’ living areas raises the risk of victimisation. Moreover, it was noted that if affluent areas are surrounded by poor areas, they tend to have higher burglary dwelling rates that the model might otherwise estimate. The results confirm that the likelihood of victimisation varies dramatically with demographic, socio-economic and area characteristics.

As SimCrime initially predicted all crimes, but the crime statistics are for recorded crimes, recorded burglary dwelling was also estimated using SimCrime. SimCrime gave a much better match with real data for Leeds as a whole compared with results at ward level. The flip side of this analysis is that SimCrime can also be used to estimate the reporting rate. It has been found that reporting rates in Leeds vary considerably by crime type, matching the conclusions of the British Crime Survey. Insurance claims are the main reason for reporting crime in Leeds.

Objectives eight and nine: to investigate movement of offenders and to explore the interaction between the location of offence and the location of offender by linking the microsimulation model with a spatial interaction model, were examined in Chapter 8. It has been found that movements of offenders are relative short. Offenders commit crimes close to where they live for most crime types except fraud and forgery. The journey to crime helps to explain why areas with more offenders are not necessarily areas with more offences and areas with a smaller number of offenders are not necessarily areas with a smaller number of offences.

The spatial interaction model for burglary dwelling was formulated to model offender flows for burglary dwelling. The factors that determine the magnitude of spatial interaction


219

between ‘where offenders live’ and ‘where the crimes occur’ are attractiveness of the home area and distance between the two areas. The characteristics of destinations that affect the interaction are the proportion of rented tenure type and students, scaled deprivation and number of cars in the area.

To fulfil objective ten: to enable the what-if analysis of a range of policy scenarios, Chapter 9 demonstrated how SimCrime and spatial interaction models can be used for what-if analysis. Scenarios presented in the chapter attempt to describe how current conditions may change in the future. They are not forecasts per se; they are descriptions and portrayals of events and trends as they could change dependent on policy decisions – not as they will. SimCrime was used to estimate the impact of changing the socio-economic structure of the Gipton and Harehills Neighbourhood Renewal Areas (NRAs) in Leeds. The results of presented scenarios shed new light on the way in which an area’s socio-economic structure affects the risk of victimisation, specifically for burglary dwelling. It has been found that burglary dwelling victimisation is more likely in disadvantaged neighbourhoods, but, surprisingly, may also increase in neighbourhoods where socio-economic improvements are taking place. SimCrime indicated the likely effects of demographic and socio-economic changes on crime victimisation trends assuming no other factors come into play.

When the socio-economic conditions of areas are changed, it also results in a different pattern of offender flows. This was assessed by using the spatial interaction model. It has been found that burglary dwelling in an area is not only dependent upon individual and area characteristics, but also upon the city context. The offender flows to one area is dependent on what is happening in neighbouring areas. Changing the socio-economic structure in one area would have a significant impact on the existing offenders’ flows patterns, especially in surrounding areas.

10.3 Evaluation and Limitations of the Research As stated in Chapter 1, the principal aim of this thesis was to investigate the potential of spatial microsimulation for modelling crime. This aim has been achieved to a large extent. The work in this thesis has shown that the spatial microsimulation model has enabled the modelling of crime victimisation at small area levels (e.g. ward level). Before this the smallest area of modelling crime in the UK was at the police force area level. In practice, SimCrime is able to identify those most at risk of becoming a victim of crime down to the output area level or community level. This could help in the planning of crime prevention programmes in the future.


220

Adding geo-references into the British Crime Survey makes it more valuable and the spatial aspect is capable of providing geographical detail for different scales. This is one of the advantages of spatial microsimulation. Moreover, it is possible to estimate new geographical distributions of geo-demographic which are unavailable from published sources. For example, it becomes possible to identify individuals with the characteristics associate with a higher propensity to commit crime, i.e. male aged 16-24, unemployed, in rented tenure. This might be useful for monitoring change over time and can provide the basis for policy initiatives centred on ‘crime prevention’ because crime can be prevented by having an impact on those characteristics which lead people to be more likely than others to commit crimes – for example, by increasing their income.

Spatial microsimulation has a great advantage over other approaches for modelling crime in that it can be applied for policy analysis. It can identify which types of households and which geographical areas would be affected under different policy scenarios. The examples of what-if analyses that were presented in Chapter 9 may be useful for policy makers to evaluate the socio-economic impacts, as well as spatial impacts, of proposed policy changes. Examining the effects of socio-economic change on crime, and understanding the effects of possible trends on crime, is an important step in the process of developing crime prevention initiatives. Very little work has been undertaken to date in this field.

Although crime hotspot analysis has proved to be useful for developing crime reduction strategies and policing plans, it neglects the dynamics of the system – the interaction between where the offenders/victim live and where the crimes occur, as well as the negative side effects of crime on an area. Crime is the product of an interaction between the person and the setting. What attracts offenders to commit crime in some areas, but not others? This thesis has shown the strength of a spatial microsimulation framework that has been successfully linked with a spatial interaction model. The spatial interaction model developed in this study can be used not only to explain spatial flows but also the consequences of changes in the conditions of areas.

However, it is important to recognise the limitations of the research. As Ballas (2001) pointed out, caution is essential when using spatial microsimulation methodologies to carry out what-if policy analysis because the output of all microsimulation models is always simulated and not actual data. Moreover, it should be noted that the model presented in this thesis cannot be used to analyse the longer-term effects of policy changes. It is only a snapshot and does not deal with many of the complexities involved in the dynamics of urban renewal - for example the problems of developing strong community economies.


221

Models can always be improved, and this model is no exception. The quality of the synthetic population in this thesis is likely to be affected by the size of the sample used as a microdata database, the number of constraint variables, the consistency of constraint tables, and the value of the control parameters of the simulated annealing method.

SimCrime was constructed from the 2001 Census and the 2001/02 British Crime Survey. The constraint variables from the census were from the only data available at the time of the research (see Table 4.2 in § 4.2.1). One limitation of the UK Census is the discrepancies in census counts between tables. This produced some inconsistencies in the constraint tables used in this study (see § 4.2.1 and § 6.2.2). This means that in some cases there is no possible combination of individuals that would match every constraining table perfectly, as pointed out by Huang and Williamson (2001). This thesis has demonstrated a method to minimise discrepancies between the totals of the constraint tables. However, some areas still have different numbers of total population (up to + 5 people).

The validation of the synthetic population has limitations. The problem is that there is no available microdata dataset for comparison. Moreover, the validation of being a ‘victim of crime’ is also problematic. It has long been noted that many crimes are not reported to the police and not all those reported are recorded. Therefore, there is no available data that captures all the victims that can be compared with the results of SimCrime.

10.4 Possibilities for Future Research Having summarised the research achievements, evaluated the research and discussed its limitations, this section proposes how the work presented in this thesis can be extended. Some potential future investigations are also provided.

Spatial microsimulation models such as SimCrime can be further developed by including more constraint variables to improve the quality of the microdata distribution. It should be noted that the method used to generate the microdata distribution in this study is flexible and it is possible to add more constraint variables or change some of them. One of the strengths of using spatial microsimulation is the ability to apply it to different scales. The framework developed within this thesis has the potential to be applied to different areas and different spatial scales. For example, it would be interesting to apply this method to district or police force area levels across the country to see the difference in levels of victimisation. To do this the constraint tables need to be selected to correspond to the area of the study.


222

The British Crime Survey is accepted as the most authoritative and reliable indicator of crime trends. The surveys have been carried out on a continuous basis since 2001. As well as using different geographical scales, the experimentation could also be used for a different period of time (for example, annually along with the British Crime Survey) to see the trend of crime victimisation at the small area level. However, the static nature of the census data may cause a problem because it represents a snapshot only every ten years.

Static spatial microsimulations such as SimCrime can be extended to be dynamic. At the moment SimCrime is based on snapshots of the current circumstances of a sample of the population in 2001. To be dynamic we need to build up a synthetic longitudinal microdata dataset that describes the individuals’ lifetimes in terms of their criminality or victimhood into the future. However this is beyond the remit of the current study.

10.5 Concluding Statements In this thesis, an innovative spatial microsimulation framework has been presented which models crime at the small area level. Within this framework, victims, offenders and locations were examined. The model provides a predictive capacity which can be used to inform policy making. Crimes at the small area scale often do not match expectations based on national averages, and this study found that patterns of crime vary considerably across Leeds both geographically and by type of area. Many of these differences result from the variation in the demographic and socio-economic make-up of these areas, for both victims and offenders. The risk of becoming a victim or an offender can be very different depending on both demographic attributes and the socio-economic characteristics of the local neighbourhood. This model takes such conditions into account in order to provide a more accurate reflection of crime across the city and its outlying regions.

References

223

References Andreassen, L., Fredriksen, D. and Ljones, O. (1994), The Future Burden of Public Pension Benefits: A Microsimulation Study, Discussion Papers 115, Research Department of Statistics Norway, [online] http://ideas.repec.org/p/ssb/dispap/115.html, accessed 17/03/2006. Baldwin, J. and Bottoms, A. E. (1976), The Urban Criminal: A study in Sheffield, London, Tavistock. Ballas, D. (2001), A Spatial Microsimulation Approach to Local Labour Market Policy analysis, unpublished PhD Thesis, School of Geography, University of Leeds, Leeds. Ballas, D. and Clarke, G. P. (2000), GIS and microsimulation for local labour market policy analysis, Computers, Environment and Urban Systems, 24, 305-330. Ballas, D. and Clarke G. P. (2001a), Modelling the Local Impacts of National Social Policies: A Spatial Microsimulation Approach, Environment and Planning C: Government and Policy, 19, 587-606. Ballas, D. and Clarke, G. P. (2001b), Towards Local Implications of Major Job Transformations in the City: A Spatial Microsimulation Approach, Geographical Analysis, 31, 291-311. Ballas, D., Clarke, G. P., Dorling, D., Eyre, H., Rossiter, D. and Thomas, B. (2005a), SimBritain: A Spatial Microsimulation Approach to Population Dynamics, Population, Place and Space, 11, 13-34. Ballas, D., Clarke, G. P., Wiemers, E. (2005b), Building a Dynamic Spatial Microsimulation Model for Ireland, Population, Space and Place, 11 (3), 157-172. Ballas, D., Kingston, R., Stillwell, J. and Jin, J. (2004), Building a Spatial Microsimulation Decision Support System, Paper presented at 7th AGILE Conference on Geographic Information Science, 29 April-1 May 2004, Heraklion, Greece, [online] http://www.agile-secretariat.org/Conference/greece2004/papers/8-2-2_Ballas.pdf, accessed 24/11/2005. Ballas, D., Rossiter, D., Thomas, B., Clarke, G. P., and Dorling, D. (2005c), Geography Matters: Simulating the local impacts of national social policies, York, York Publishing Services. Barberet, R., Fisher, B. S., Farrell, G. and Taylor, H. (2003), University Student Safety, Finding 194, Home Office, London, [online] http://www.homeoffice.gov.uk/rds/pdfs2/r194.pdf, accessed 6/05/2003. Becker, G. (1968), Crime and Punishment: An Economic Approach, Journal of Political Economy, 76, 169-217 Birkin, M., Clarke, G. P. and Clarke, M. (1996), Urban and Regional Modelling at the Microscale, in Clarke, G. P. Ed., Microsimulation for Urban and Regional Policy Analysis, London, Pion, pp 10-27.

References

224

Birkin, M., Clarke, G. P. and Clarke, M. (2002), Retail Geography and Intelligent Network Planning, Chichester, Wiley. Birkin, M. and Clarke, M. (1988), Synthesis- A Synthetic Spatial Information System for Urban and Regional Analysis: Method and Examples, Environment and Planning A, 20, 1645-1671. Bottoms, A. E., Mawby, R. I. and Xanthos, P. (1989), A Tale of Two Estates, in Downes, D. Ed., Crime and the City, London, Macmilan. Bottoms, A, E. and Wiles, P. (1986), Housing Tenure and Residential Community Crime Careers in Britain, Crime and Justice, 8, 101-162. Bottoms, A. E. and Wiles, P. (1988), Crime and Housing Policy: A Framework for Crime Prevention Analysis, in Hope, T. and Shaw, M. Eds., Communities and Crime Reduction, London, Home Office Research and Planning Unit. Bottom, A. E. and Wiles, P. (2002), Environmental Criminology, in Maguire, M., Morgan, R. and Reiner, R. Eds., The Oxford Handbook of Criminology, Oxford, Oxford University Press, pp 620-656. Bowers, K. and Hirschfield, A. (1999), Exploring Links between Crime and Disadvantage in North-West England: An Analysis Using Geographical Information Systems, International Journal of Geographical Information Science, 13, 159-184. Bowers, K. and Hirschfield, A. (2001), Introduction, in Hirschfield, A. and Bowers, K. Eds., Mapping and Analysing Crime Data: Lessons from Research and Practice, London, Taylor Francis, pp 1-8. Brantingham, P. J. and Brantingham, P. L. (1981), Mobility, Notoriety, and Crime: A Study in the Crime Patterns of Urban Nodal Points, Journal of Environmental Systems, 11(1), 89-99. Brown, L. and Harding, A. (2002), Social Modelling and Public Policy: Application of Microsimulation Modelling in Australia, Journal of Artificial Societies and Social Simulation, 5(4), [online] http://jasss.soc.surrey.ac.uk/5/4/6.html, accessed 24/11/2005. Budd, T. (1999), Burglary of Domestic Dwellings: Findings from the British Crime Survey, Home Office Statistical Bulletin Issue 4/99, [online] http://www.homeoffice.gov.uk/rds/pdfs/hosb499.pdf, accessed 16/06/2004. Budd, T. (2003), Alcohol-related Assault: Finding from the British Crime Survey, Home Office Online Report 35/03, [online], http://www.homeoffice.gov.uk/rds/pdfs2/rdsolr3503.pdf, accessed, 11/08/2004. Caldwell, S. B., Clarke, G. P. and Keister, L. A. (1998), Modelling Regional Changes in US Household Income and Wealth: A Research Agenda, Environment and Planning C: Government and Policy, 16, 707-722. Caldwell, S. B. and Keister, L. A. (1996), Wealth in America: Family Stock Ownership and Accumulation, 1960-1995, in Clarke, G. P. Ed., Microsimulation for Urban and Regional Policy Analysis, London, Pion, pp 88-116. Carstairs, V. and Morris, R. (1991), Deprivation and Health in Scotland, Aberdeen, Aberdeen University Press.

References

225

Cater, J. and Jones, T. (1989), Social Geography: An Introduction to Contemporary Issues, New York, Routledge. Ceccato, V., Haining, R. P. and Signoretta, P. E. (2002), Exploring Offence Statistics in Stockholm City using Spatial Analysis Tools, Annals of the Association of American Geographer, 92, 29-51. Chainey, S. and Ratcliffe, J. (2005), GIS and Crime Mapping, Chichester, Wiley. Chamlin, M. and Cochran, J.K. (2000), Unemployment, Economic Theory, and Property Crime: A Note on Measurement, Journal of quantitative Criminology, 16, 443-455. Clarke, G. P. (1996), Microsimulation: An introduction, in Clarke, G. P. Ed., Microsimulation for Urban and Regional Policy Analysis, London, Pion, pp 1-9. Clarke, S. (2003), Estimating Personal Income for Small Areas, The Yorkshire and Humber Regional Review, 13 (3), 20-22. Cohen, L. E. and Felson, M. (1979), Social Change and Crime Rate Trends: A Routine Activity Approach, American Sociological Review, 44, 588-608. Costello, A. and Wiles, P. (2001), GIS and the Journey to Crime: An Analysis of Patterns in South Yorkshire, in Hirschfield, A. and Bowers, K. Eds., Mapping and Analysing Crime Data: Lessons from Research and Practice, London, Taylor Francis, pp 27-60. Cornish, D. and Clarke, R. (1986), The Reasoning Criminal: Rational Choice Perspectives on Offending, New York, Springer-Verlag. Craglia, M., Haining, R. and Signoretta, P. (2001), Modelling High-Intensity Crime Areas in English Cities. Urban Studies, 38, 1921-1941. Craglia, M., Haining, R. and Wiles, P. (2000), A Comparative Evaluation of Approaches to Urban Crime Pattern Analysis, Urban Studies, 37, 711-729. Croall, H. (1998), Crime and Society in Britain, London, Longman. CUSU (2005), Reporting Crime, [online] http://www.cusu.cam.ac.uk/welfare/safety/crime.html, accessed 11/04/2006. Dahlbäck, O. (1998), Modelling the Influence of Societal Factors on Municipal Theft Rates in Sweden: Methodological Concerns and Substantive Findings, Acta Sociologica, 41, 37-57. Dale, A. and Teague, A. (2002), Microdata from the Census: Samples of Anonymised Records, in Rees, P., Martin, D. and Williamson, P. Eds., The Census Data System, Chichester, Wiley, pp 203-212. Darden, J. T. (2001), Race Relation in the City, in Paddison, R. Ed., Handbook of Urban Studies, London, SAGE. Denham, C. and Rees, P. (2002), An Output Strategy for the 2001 Census, in Rees, P., Martin, D. and Williamson, P. Eds., The Census Data System, Chichester, Wiley, pp 305-326.

References

226

Dhiri, S., Brand, S., Harries, R. and Price, R. (1999), Modelling and Predicting Property Crime Trends in England and Wales, Home Office Research Study 198, [online] http://www.homeoffice.gov.uk/rds/pdfs/hors198.pdf, accessed 14/01/2004. Dodd, T., Nicholas, S., Povey, D. and Walker, A. (2004), Crime in England and Wales 2003/2004, Home Office Statistical Bulletin 10/04, [online] http://www.homeoffice.gov.uk/rds/pdfs04/hosb1004.pdf, accessed 17/07/2006. Edmark, K. (2005), Unemployment and Crime: Is There a Connection? Scandinavian Journal of Economics, 107 (2), 353-373. Ellingworth, D., Hope, T., Osborn, D. R., Trickett, A., and Pease, K. (1997), Prior Victimisation and Crime Risk, International Journal of Risk, Security and Crime Prevention, 2, 201-214. Ellis, L. (1988), The Victimful- Victimless Crime Distinction, and Seven Universal Demographic Correlates of Victimful Criminal Behavior, Personality and Individual Differences, 9 (3), 525-548. Entorf, H. and Spengler, H. (2000), Socioeconomic and Demographic Factors of Crime in Germany: Evidence from Panel Data of the German States, International Review of Law and Economics, 20, 75-106 Evandrou, M., Falkingham, J., Johnson, P. and Rake, K. (2001), SAGE: Simulating Social Policy for an Ageing Society: A Research Agenda, SAGE Discussion Paper No. 1, London School of Economics, [online] http://www.lse.ac.uk/collections/SAGE/pdf/SAGE_DP1.pdf, accessed 23/09/2004. Evans, D. J (1989), Geographical Analyses of Residential Burglary, in Evans, D. J. and Herbert D. T. Ed., The Geography of Crime, London, Routledge. Eyre, H. A. (1999), Measuring the Performance of Spatial Interaction Models in Practice, Unpublished PhD Thesis, School of Geography, University of Leeds, Leeds. Falkingham, J. and Hills, J. (1995a), The Effects of the Welfare State over the Life Cycle. In Falkingham, J. and Hills, J Eds., The Dynamic of Welfare: The Welfare State and the Life Cycle, New York, Prentice Hall/Harvester Wheatsheaf, pp 83-107. Falkingham, J. and Hills, J. (1995b), Redistribution between People or Across the Life Cycle? in Falkingham, J. and Hills, J Eds., The Dynamic of Welfare: The Welfare State and the Life Cycle, New York, Prentice Hall/Harvester Wheatsheaf, pp 137-149. Felson, M. (1994), Crime and Everyday Life: Insight and Implications for Society, London, Pine Forge. Field, S. (1998), Trends in Crime Revisited, Home Office Research Study 195, [online] http://www.homeoffice.gov.uk/rds/pdfs/hors195.pdf, accessed 8/01/2004. Fredriksen, D. (1998), Projections of Population, Education, Labour Supply and Public Pension Benefits: Analyses with the Dynamic Microsimulation Model MOSART, Social and economic studies 101, Statistics Norway [online] http://www.ssb.no/emner/02/03/sos101/sos101.pdf, accessed 24/06/2005.

References

227

Gaviria, A. and Pagés, C. (2002), Pattern of Crime Victimisation in Latin American Cities, Journal of Development Economics, 67, 181-203. Glaeser, E. L. and Sacerdote, B. (1999), Why is There More Crime in Cities?, Journal of Political Economy, 107 (6), 225-258. Gottfredson, M. (1984), Victim of Crime: The Dimensions of Risk, Home Office Research Study No. 81, [online] http://www.homeoffice.gov.uk/rds/pdfs05/hors81.pdf, accessed 11/10/2004 Groff, E. R. and La Vigne, N. G. (2001), Mapping an Opportunity Surface of Residential Burglary, Journal of Research in Crime and Delinquency, 38 (3), 257-278. Guerry, A. M. (1833), Essai sur la Statistique Morale de la France, Paris, Grochard. A Translation of Andre-Michel Guerry's Essay on the Moral Statistics of France (1833): A Sociological Report to the French Royal Academy of Science, edited and translated by Whitt, H. P. and Reinking, V. W. (2002), Studies in French Civilization vol. 26, New York, Edwin Mellea Press. Haining, R. (2003), Spatial Data Analysis: Theory and Practice, Cambridge, Cambridge University Press. Hale, C. (1997), The Labour Market and Post-War Crime Trends in England and Wales, in Carlen, P. and Morgan, R. Eds., Crime Unlimited: Questions for the 21st Century, Basingstoke, Macmillan, pp 30-56. Hancock, R., Mallender, J. and Pudney, S. (1992), Constructing a Computer model for Simulating the Future Distribution of Pensioner's Incomes for Great Britain, in Hancock, R. and Sutherland, H. Eds., Microsimulation Models for Public Policy Analysis: New Frontiers, London, STICERD, pp 33-66. Hansen, K., and Machin, S. (2003), Modelling Crime at Police Force Area Level, in Modelling Crime and Offending: Recent Developments in England and Wales, Occasional Paper No 80, Section E, Home Office, [online] http://www.homeoffice.gov.uk/rds/pdfs2/occ80modelling.pdf, access 05/12/2003. Harries, K. D. (1974), The Geography of Crime and Justice, New York, McGraw-Hill. Heidensohn, F. (2002), Women and Policing in the Twentieth and Twenty –First Century: Gender, Identity and Choices, Abstract presented at the Australian Institute of Criminology Conference: Third Australasian Women and Policing Conference: Women and Policing Globally, 20-23 October 2002, Canberra, Australia, [online] http://www.aic.gov.au/conferences/policewomen3/heidensohn2.html, access 21/04/2005. Herbert, D. T. (1976), The Study of Delinquency Areas: A social Geographical Approach, Transactions of the Institute of British Geographers, 1, 472-92. Herbert, D. T. (1979), Urban Crime: A Geographical Perspective' in Herbert D. T. and Smith D. M. Eds., Social Problems and the City: Geographical Perspectives, Oxford, Oxford University Press, pp 117–35. Herbert, D. T. (1982), The Geography of Urban Crime, Harlow, Longman.

References

228

Herbert, D. T. (1983), Crime and Delinquency, in Pacione, M. Ed., Progress in Urban Geography, London, Croom Helm, pp 75-102. Herbert, D. T. (1989), The Geography of Urban Crime, New York, Longman. Hirschfield, A., Bowers, K and Brown, P. J. B. (1995), Exploring Relations between Crime and Disadvantage on Merseyside, European Journal on Criminal Policy and Research, 3 (3), 93-112. Hirschfield, A. and Bowers, K. J. (1997), The Effect of Social Cohesion on Levels of Recorded Crime in Disadvantaged Areas. Urban Studies, 34, 1275-1295. Holm, E., Holme, K., Mäkilä, K., Mattsson-Kauppi, M. and Mörtvik, G. (2002), The SVERIGE Spatial Microsimulation Model: Content, Validation, and Example Applications, Spatial Modelling Centre, Kiruna, Sweden, [online] http://www3.umu.se/soc_econ_geography/smc/PDF/SVERIGE_validation.pdf, accessed 8/01/2004. Home Office (2006a), Changes in How Police Record Crime, [online] http://www.crimestatistics.org.uk/output/Page107.asp#General%20changes, accessed 24/03/2006 Home Office (2006b), Counting Rules, [online] http://www.homeoffice.gov.uk/rds/countrules.html, accessed 24/03/2006. Home Office (2006c), Student Safety, [online] http://www.homeoffice.gov.uk/crimevictims/how-you-can-prevent-crime/student-safety/?version=1, accessed 15/07/2006. Hope, T., Bryan, J., Trickett, A. and Osborn, R. D. (2001), The Phenomena of Multiple Victimization: The Relationship between Personal and Property Crime Risk, British Journal of Criminology, 41(4), 595-617. Huang, Z. and Williamson, P. (2001), A Comparison of Synthetic Reconstruction and Combinatorial Optimisation Approaches to the Creation of Small-Area Microdata. Working paper 2001/02, Department of Geography, University of Liverpool, [online] http://pcwww.liv.ac.uk/%7Ewilliam/microdata/Methodology/workingpapers/hw_wp_ 2001_2.pdf, accessed 27/01/2004. Kelly, M. (2000), Inequality and Crime, Review of Economics and Statistics, 82, 530-539. Kershaw, C., Budd, T., Kinshott, G., Mattinson, J., Mayhew, P. and Myhill, A. (2000), The 2000 British Crime Survey, Home Office Statistical Bulletin 18/00, [online] http://www.homeoffice.gov.uk/rds/pdfs/hosb1800.pdf, accessed 18/03/2004. Kershaw, C., Matthews, N. C., Thomas, C., Aust, R. (2001), The 2001 British Crime Survey: First Results, England and Wales, Home Office Statistical Bulletin 18/01, [online] http://www.homeoffice.gov.uk/rds/pdfs/hosb1801.pdf, accessed 7/08/2004. King, A., Bækgaard, H. and Robinson, M. (1999), Dynamod 2: An Overview, Technical Paper No 19, National Centre for Economic Modelling, University of Canberra, [online] http://www.natsem.canberra.edu.au/publications/papers/tps/tp19/tp19.pdf, accessed 8/08/2005. Kirkpatrick, S., Gelatt, C. D. Jr. and Vecchi, M. P. (1983), Optimization by Simulated Annealing, Science, 220, 671-680.

References

229

Knudsen D. C, and Fotheringham A. S (1986), Matrix Comparison, Goodness-of-Fit and Spatial Interaction Modelling, International Regional Science Review, 10 (2), 127147. Kongmuang, C. (1995), The spatial Analysis of Crime in Amphoe Muang Changwat Chiangmai, Unpublished Master Thesis, Department of Geography, Chiangmai University, Thailand. Lambert, S., Percival, R., Schofield, D. and Paul, S. (1994), An Introduction to STINMOD: A Static Microsimulation Model, Technical Report 1, National Centre for Social and Economic Modelling (NATSEM), University of Canberra, [online] http://www.natsem.canberra.edu.au/publications/papers/tps/stp1/stp1.pdf, accessed 14/02/2005. Laub, H. J. (1997), ‘Who are the victims?’, in Davis, R. C., Lurigio, A.J. and Skogan, W. G. Eds., Victims of Crime 2nd ed., London, SAGE. Laycock, G. (2003), Introduction, in Crime Uncovered: A Nation under the Cosh? The Truth about Crime in Britain in 2003, Supplement the Observer 27 April 2003. Leeds City Council (2004a), Skills Audit Gipton Neighbourhood Renewal Area, [online] http://www.leedsinitiative.org/initiativeDocuments/200534_82575626.pdf, access 15/09/2005. Leeds City Council (2004b), Skills Audit Harehills Neighbourhood Renewal Area, [online] http://www.leedsinitiative.org/initiativeDocuments/200534_21905154.pdf, access 15/09/2005. Leeds City Council (undated), Gipton Neighbourhood Renewal Area Map, [online] http://www.leeds.gov.uk/files/2005/week24/inter__35f47844-9e59-409d-a7e1c52bf4284aae_a0c011c5-264e-4279-b990-729b774a70da.pdf, accessed, 15/09/2005. Leeds City Council (undated), Harehills Neighbourhood Renewal Area Map, [online] http://www.leeds.gov.uk/files/2005/week24/inter__35f47844-9e59-409d-a7e1c52bf4284aae_f6097b0b-a638-4038-a4c8-592fc91f69e0.pdf, accessed, 15/09/2005. Leeds Community Safety (2004), Leeds Crime, Disorder and Drugs Audit 2004, [online] http://www.leeds-csp.org.uk/files/Leeds%20Crime,%20Disorder%20and%20Drug %20Audit%202004(1).pdf, accessed 24/11/2005. Leeds Initiatives (2005), Leeds Regeneration Plan 2005 to 2008, [online] http://www.leedsinitiative.org/initiativeDocuments/20051129_98887271.pdf, accessed 16/03/2006. Leeds Statistics (2004), Area Statistics: Police Division, [online] http://www.leeds-statistics.org/PDF_Downloads/Police.htm, accessed 21/04/2006. Levine, N. (2006), Crime Mapping and the CrimeStat Program, Geographical Analysis, 38 (1), 41-56. Levy H., Mercader-Prats, M. and Planas, M. (2001), An Introduction to ESPASIM: A Microsimulation Model to Assess Tax-Benefit Reforms in Spain, Brazilian Electronic Journal of Economics, 4 (1), 1- 23, [online] http://www.beje.decon.ufpe.br/, accessed 5/11/2004.

References

230

Lloyd, R. and Harding, A. (2004), Getting Down to Small Areas: Estimating Regional Income and Wealth in Australia using Spatial Microsimulation, Paper presented at Regional Science Association International- British and Irish Section Conference, 1820 August 2004, Cork, Ireland, [online] http://www.natsem.canberra.edu.au/publications/papers/cps/cp04/2004_006/cp2004_0 06.pdf, accessed 5/12/2004. Longley, P. A., Goodchild, M. F., Maguire, D. J. and Rhind, D. W. (2001), Geographic Information Systems and Science, Chichester, Wiley. Maguire, M. (2002), Crime Statistics: The ‘Data Explosion’ and its Implications, in Maguire, M., Morgan, R. and Reiner, R. Eds., The Oxford Handbook of Criminology 3rd Edn., Oxford, Oxford University Press. Maguire, M. and Bennett, T. (1982), Burglary in a Dwelling: The Offence, the Offender and the Victim, London, Heinemann. Mannheim, H. (1965), Comparative Criminology (Volume 1), London, Routledge and Kegan Paul. Martin, D. (2002), Spatial Patterns in Residential Burglary: Assessing the Effect of Neighborhood Social Capital, Journal of Contemporary Criminal Justice, 18, 132146. Martini, A. and Trivellato, U. (1997), The Role of Survey Data in Microsimulation Models for Social Policy Analysis, Labour, 11, 83-112. Master, T. (1995), Advanced Algorithms for Neural Networks: A C++ Sourcebook, New York, John Wiley & Sons. Mayhew, P., Maung, N. A. and Mirlees-Black, C. (1993), The 1992 British Crime Survey, Home Office Research Studies 132, London, HMSO. McClintock, F. H. and Avison, H. (1968), Crime in England and Wales, London, Heinemann. Melhuish, T., Blake, M. and Day, S. (2002), An Evaluation of Synthetic Household Populations for Census Collection Districts Created Using Spatial Microsimulation Techniques, Paper presented at The 26th Australian & New Zealand Regional Science Association International (ANZRSAI) Annual Conference, 29 Sept – 2 Oct 2002, Gold Coast, Queensland, Australia, [online] http://www.natsem.canberra.edu.au/publications/papers/cps/cp02/2002_013/cp2002_0 13.pdf, accessed 25/08/2004. Mertz, J. (1991), Microsimulation - A Survey of Principles Developments and Applications, International Journal of Forecasting, 7, 77-104. Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H. and Teller, E. (1953), Equation of State Calculations by Fast Computing Machines, Journal of Chemical Physics, 21, 1087-1092. Miethe, T. D. and Meier, R. F. (1990), Opportunity, Choice and Criminal Victimisation: A Test of a Theoretical Model, Journal of Research in Crime and Delinquency, 27, 243266.

References

231

Mitton, L., Sutherland, H. and Weeks, M. (2000), Microsimulation Modelling for Policy Analysis, Cambridge, Cambridge University Press. Murie, A. (1997), Linking Housing Changes to Crime, Social Policy and Administration, 31, 22-36. Nelson, A. L., Bromley, R. D. F. and Thomas, C. J. (1996), The Geography of Shoplifting in a British City: Evidence from Cardiff, Geoforum, 27 (3), 409-423. O’Donoghue, C. (2001a), Dynamic Microsimulation: A Methodological Survey, Brazilian Electronic Journal of Economics, 4(2), [online] http://www.beje.decon.ufpe.br/v4n2/cathal.htm, accessed 11/06/2005. O’Donoghue, C. (2001b), Introduction to the Special Issue on Dynamic Microsimulation Modelling, Brazilian Electronic Journal of Economics, 4(2), [online] http://www.beje.decon.ufpe.br/v4n2/intro.htm, accessed 11/06/2005. Office for National Statistics (2003), Indices of Deprivation for Wards 2000, [online], http://www.neighbourhood.statistics.gov.uk/dissemination/viewFullDataset.do?instanc eSelection=06473&productId=802&$ph=60_61&datasetInstanceId=6473&startColum n=1&numberOfColumns=8&containerAreaId=543394, accessed 24/01/2005. Openshaw, S., Charlton, M., Wymer, C. and Craft, A. (1987), A Mark I Geographical Analysis Machine for the Automated Analysis of Point Data Sets, International Journal of Geographical Information Systems, 1, 335-358. Orcutt, G. H. (1957), A new type of socio-economic system, The Review of Economics and Statistics, 39, 116-123 Orcutt, G. H., Greenberger, M., Korbel, J. and Rivlin, A. (1961), Microanalysis of Socioeconomic Systems: A Simulation Study, New York, Harper and Row. Orcutt, G. H., Merz, J. and Quinke, H. (1986), Microanalytic Simulation Models to Support Social and Financial Policy, Amsterdam: North-Holland, Elsevier. Osborn, D. R., Tseloni, A. (1998), The Distribution of Household Property Crimes, Journal of Quantitative Criminology, 14, 307-330. Pham, D. T. and Karaboga, D. (2000), Intelligent Optimisation Techniques: Genetic Algorithms, Tabu Search, Simulated Annealing and Neural Networks, London, Springer. Portnov, B. A. and Rattner, A. (2003), Spatial Patterns of Crime in Israel: Investigating the Effects of Inter-Urban Inequality and Proximity, Paper presented at 43rd European Congress of the Regional Science Association, 27-30 August 2003, Jyvaskyla, Finland, Paper provided by European Regional Science Association in its series ERSA Conference Papers, [online] http://www.ersa.org/ersaconfs/ersa03/cdrom/papers/512.pdf, accessed 24/11/2004. Raikes, S. (2002), Never Too Early, Criminal Justice Matters, 50, 24-25, [online] http://www.kcl.ac.uk/depsta/rel/ccjs/cjm/contents/41-50/cjm50.pdf, accessed 23/05/2004.

References

232

Ratcliffe, J. and McCullagh, M. (2001), Crime Repeat Victimisation and GIS, in Hirschfield, A. and Bowers, K. Eds., Mapping and Analysing Crime Data: Lessons from Research and Practice, London,Taylor and Francis. Read, T. and Oldfield, D. (1995), Local Crime Analysis, Police Research Group: Crime Detection and Prevention Series, Paper No 65, [online] http://www.homeoffice.gov.uk/rds/prgpdfs/fcdps65.pdf, accessed 8/06/2003. Rees, P., Martin, D. and Williamson, P. (2002), Census Data Resources in the United Kingdom, in Rees, P., Martin, D. and Williamson, P. Eds., The Census Data System, Chichester, Wiley, pp 1-24. Reilly, B. and Witt, R. (1992), Crime and Unemployment in Scotland: An Econometric Analysis Using Regional Data, Scottish Journal of Political Economy, 39(2) 213-228. Rephann, T. J., MAKILA, K. and HOLM, E. (2005), Microsimulation for Local Impact Analysis: An Application to Plant Shutdown. Journal of Regional Science, 45, 183222. Rephann, T. J., Öhmann, M. (1999), Building a Microsimulation Model for Crime in Sweden: Issues and Applications, Paper presented at the Seminarium om Ekobrottsforskning, 22 February 1999, Stockholm, Sweden, [online] http://www.equotient.net/papers/crimemic.pdf, accessed 30/09/2002. Schmid, C. F. (1960), Urban Crime Areas, American Sociological Review, part I: 25(4), 527542, part II: 25 (5), 655-78. Schuerman, L. and Kobrin, S. (1986), Community Careers in Crime, Crime and Justice, 8, 67-100. Scott, P. (1972), The Spatial Analysis of Crime and Delinquency, Australian Geographical Studies, 10, 1-18. Senior, M. (2002), Deprivation Indicators, in Rees, P., Martin, D. and Williamson, P. Eds., The Census Data System, Chichester, Wiley, pp 123-137. Shaw, C. R. and McKay, H. D. (1942), Juvenile Delinquency and Urban Areas, Chicago, University of Chicago Press. Shaw, C. R. and McKay, H. D. (1969), Juvenile Delinquency and Urban Areas, Chicago, University of Chicago Press, revised Edn. Shepherd, P., See, L., Kongmuang, C., Clarke, G. P. (2004), An Analysis of Crime and Disorder in Leeds 2000/01 to 2003/04, Report to Leeds Community Safety Partnership, August 2004. Simmons, J. (2002), Crime in England and Wales 2001/02, Home Office Statistical Bulletin 07/02, [online] http://www.homeoffice.gov.uk/rds/pdfs2/hosb702.pdf, accessed 18/11/2004. Simmons, J. and Dodd, T. (2003), Crime in England and Wales 2002/2003, Home Office Statistical Bulletin 07/03, [online] http://www.homeoffice.gov.uk/rds/pdfs2/hosb703.pdf, accessed 24/11/2005.

References

233

Simmons, J., Legg, C. and Hosking, R. (2003a), National Crime Recording Standard (NCRS): An analysis of the impact on recorded crime. Part One: The National Picture, [online] http://www.homeoffice.gov.uk/rds/pdfs2/rdsolr3103.pdf, accessed 10/07/2006. Simmons, J., Legg, C. and Hosking, R. (2003b), National Crime Recording Standard (NCRS): An analysis of the impact on recorded crime. Part Two: Impact on Individual Police Forces, [online] http://www.homeoffice.gov.uk/rds/pdfs2/rdsolr3203.pdf, accessed 10/07/2006. Smith, D. (2002), Crime and the Life course, in Maguire, M., Morgan, R. and Reiner, R. Eds., The Oxford Handbook of Criminology, Oxford, Oxford University Press, pp 702745. Smith, S. J. (1989), The Challenge of Urban Crime, in Herbert, D. T., and Smith, D. M. Eds., Social Problems and the City, Oxford, Oxford University Press, pp 271-288. South Yorkshire Police (undated), South Yorkshire Police, [online] http://southyorks.police.uk, accessed 11/04/2006 Spielauer, M. (2002), Dynamic Microsimulation of Health Care Demand, Health Care Finance and the Economic Impact of Health Behavior: Part II: Survey and Review, Interim Report IR-02-036/ May, International Institute for Applied Systems Analysis Schlossplatz 1, Laxenburg, Austria, [online] http://www.iiasa.ac.at/Publications/Documents/IR-02-036.pdf, accessed 24/11/2004. Stillwell, J., Birkin, M., Ballas, D., Kingston, R. and Gibson, P. (2004), Simulating the City and Alternative Futures, in Unsworth, R. and Stillwell, J. Eds., Twenty-first Century Leeds: Geographies of a Regional City, Leeds, Leeds University Press, pp 435-364. Tanton, R., Jones, R. and Lubulwa, G. (2001), Modelling Crime Victimisation and the Propensity to Report Crime to the Police, Paper prepared for the Methodology Advisory Committee Meeting, 22 June 2001, Australian Bureau Statistics, Australia, [online] http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/E1399B98149C1E89CA2571 300081B6FE/$File/1352055040_Jun2001.pdf, accessed 11/11/2004. Thomas, N. and Feist, A. (2004), Chapter 7: Detection Crime, in Dodd, T., Nicholas, S., Povey, D. and Walker, A. Eds., Crime in England and Wales 2003/2004, Home Office Statistical Bulletin 10/04, pp 103-125, [online] http://www.homeoffice.gov.uk/rds/pdfs04/hosb1004.pdf, accessed 17/07/2006. Townsend, P. (1979), Poverty in the United Kingdom: A Survey of Household Resources and Standards of Living, London, Allen Lane. Trickett, A., Osborn, D. R. and Ellingworth, D. (1995), Property Crime Victimisation: The Roles of Individual and Area Influences’, International Review of Victimology, 3, 27395. Tseloni, A., Osborn, D. R., Trickett, A. and Pease, K. (2002), Modelling Property Crime using the British Crime Survey: What Have We Learnt?, British Journal of Criminology, 42, 109-128.

References

234

Turton, I. and Openshaw, S, (2001), Automated Crime Pattern Analysis using the Geographical Analysis Machine, in Hirchsfield, A. and Bowers, K. Eds., Mapping and Analysing Crime Data, Taylor and Francis, pp 11-26. Van Wilsem, J., Wittebrood, K. and De Graaf, N. D. (2006), Socioeconomic Dynamics of Neighbourhoods and the Risk of Crime Victimization: A Multilevel Study of Improving, Declining, and Stable Areas in the Netherlands, Social Problems, 53 (2), 226-247. Vickers, D. W. (2006a), Multi-Level Integrated Classifications Based on the 2001 Census, Unpublished PhD Thesis, School of Geography, University of Leeds, Leeds. Vickers, D. W. (2006b), The National Classification of Census Output Areas, [online] http://www.geog.leeds.ac.uk/people/d.vickers/OAclassinfo.html, accessed 1/06/2006. Victim of Crime in Scotland (undated), Anyone Can be a Victim of Crime, [online] http://www.scottishvictimsofcrime.co.uk, accessed 11/04/2006. Waller, I. and Okihiro, N. (1978), Burglary: The Victim and the Public, Toronto, University of Toronto Press. WikstrÖm, P. H. (1991), Urban Crime, Criminals, and Victims: The Swedish Experience in an Anglo-American Comparative Perspective, New York, Springer-Verlag. WikstrÖm, P. H., Loeber, R., (2000), Do Disadvantaged Neighbourhoods Cause Welladjusted Children to Become Adolescent Delinquents?: A Study of Male Serious Juvenile Offending, Individual Risk and Protective Factors, and Neighbourhood Context, Criminology, 38, 1109-42. Wiles, P. and Costello, A. (2000), The ‘Road to Nowhere’: The Evidence for Travelling Criminals, Home Office Research Study No. 207, Home Office, [online] http://www.homeoffice.gov.uk/rds/pdfs/hors207.pdf, accessed 12/08/2004. Williamson, P. (1992), Community Care Policies for the Elderly: A Microsimulation Approach, unpublished PhD Thesis, School of Geography, University of Leeds, Leeds. Williamson P. (2001), Modelling Alternative Domestic Water Demand Scenarios, in Clarke, G. P. and Madden, M. Eds. Regional Science in Business, Berlin, Springer-Verlag, pp 243-268. Williamson, P., Birkin, M. and Rees, P. (1998), The Estimation of Population Microdata by Using Data from Small Area Statistics and Samples of Anonymised Records, Environment and Planning A, 30, 785-816. Williamson P. (2002), Synthetic Microdata, in Rees, P., Martin, D. and Williamson, P. Eds., The Census Data System, Chichester, Wiley, pp 231-241. Williamson, P., Clarke, G. P., McDonald, A. T. (1996), Estimating Small Area Demands for Water with the Use of Microsimulation, in Clarke, G. P. Ed., Microsimulation for Urban and Regional Policy Analysis, London, Pion, pp 117-148. Witt, R., Clarke, A. and Fielding, N. (1999), Crime and Economic Activity: A Panel Data Approach, British Journal of Criminology, 39, 391-400.

References

235

Yarwood, R. (2001), Crime and Policing in the British Countryside: Some Agendas for Contemporary Geographical Research, Sociologia Rurail, 41, 201-219. Zaidi, A. and Rake, K. (2001), Dynamic Microsimulation Models: A Review and Some Lessons for SAGE, SAGE Discussion Paper No. 2, ESRC-SAGE Research Group, London School of Economics, [online] http://www.lse.ac.uk/collections/SAGE/pdf/SAGE_DP2.pdf, accessed 1/10/2004. Zaidi, A. and Scott, A. (2001), Base dataset for the SAGE model, Technical Note 1, ESRC SAGE Research Group, London School of Economics, [online] http://www.lse.edu/collections/SAGE/pdf/SAGE_TN1.pdf, accessed 1/10/2004. Zedner, L. (2002), Victims, in Maguire, M., Morgan, R. and Reiner, R. Eds., The Oxford Handbook of Criminology, Oxford, Oxford University Press, pp 419-456.

21 58 6 37 29 72 182 448 21 18 53 161 141 14 28 68 16 62 14 14 43 8 11 17 59 8 136 49 374 62 6 25 24 2,285


Sexual Offences

16 169 21 427 14 146 20 241 19 260 31 359 41 336 66 3,139 6 302 6 190 5 375 19 308 8 805 5 560 19 240 19 590 12 404 5 296 9 371 8 424 4 448 6 397 6 271 5 294 14 437 10 317 9 611 23 164 34 1,394 8 406 6 201 7 297 7 298 488 15,477

Violent Crime

8 371 32 484 11 181 14 303 41 400 40 680 37 503 202 7,334 8 253 12 217 15 705 38 522 22 834 13 384 21 517 18 739 20 319 13 360 15 684 15 639 8 319 18 361 18 397 17 340 24 400 15 293 13 646 13 329 45 1,176 18 491 121 297 16 228 19 293 940 21,999

Robbery

Other Theft

Other Crime 0 2 0 0 1 0 0 1 0 0 0 4 1 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 2 0 0 1 0 14


6 14 1 7 12 23 11 57 3 4 2 10 5 2 12 7 10 4 5 6 3 3 1 2 10 3 2 15 16 7 2 8 2 275

Homicide

Fraud & Forgery 126 133 17 115 73 258 92 947 27 69 113 131 167 115 143 173 125 134 180 97 56 48 124 22 51 101 100 59 219 119 32 93 77 4,336

Handling

Drug Offences

Criminal Damage

28 62 11 38 31 50 153 401 28 9 41 101 72 20 39 90 19 18 24 28 18 18 31 7 27 9 18 68 127 38 25 14 31 1,694

150 287 106 200 214 269 387 1,025 135 123 284 319 476 233 237 484 170 213 187 168 215 163 143 140 379 197 392 185 834 323 107 143 205 9,093

74 216 46 119 228 203 251 1,426 96 63 97 235 155 69 153 188 139 98 94 126 88 66 115 132 191 79 102 197 361 109 79 79 139 5,813

1,851 3,262 1,199 2,068 2,734 3,817 3,765 18,419 1,823 1,305 2,847 3,619 4,628 2,455 2,713 4,249 2,477 2,545 2,517 2,757 2,453 1,896 1,959 1,930 3,351 1,939 3,464 2,543 6,569 3,001 1,480 1,812 2,356 105,803

236

252 323 307 484 387 655 146 231 283 228 271 475 514 282 630 622 332 878 732 251 789 457 1,220 1,696 285 258 401 149 202 243 256 380 521 653 351 767 1,163 176 603 373 315 352 317 441 546 801 254 818 361 352 530 618 330 394 189 412 333 229 564 439 524 335 392 199 254 354 246 304 292 169 293 492 519 331 909 195 446 265 566 438 431 418 182 841 564 394 1,029 525 355 540 164 220 220 219 259 423 470 335 456 13,607 11,478 18,304

Total 2000/01

Appendix A

Aireborough Armley Barwick and Kippax Beeston Bramley Burmantofts Chapel Allerton City and Holbeck Cookridge Garforth and Swillington Halton Harehills Headingley Horsforth Hunslet Kirkstall Middleton Moortown Morley North Morley South North Otley and Wharfedale Pudsey North Pudsey South Richmond Hill Rothwell Roundhay Seacroft University Weetwood Wetherby Whinmoor Wortley Total

Burglary Elsewhere

WARDNAME

Burglary Dwelling

Appendix A: Recorded crime by crime type at ward level

27 58 6 69 46 124 231 720 17 15 55 207 242 64 46 150 20 70 24 27 39 13 33 18 91 16 155 39 578 57 11 18 21 3,307


Sexual Offences

Robbery

Other Theft

28 457 39 480 6 195 21 388 35 477 34 877 42 503 251 8,506 8 269 7 272 17 799 31 621 24 930 15 424 31 661 34 751 24 421 17 359 19 883 23 744 12 307 15 420 26 513 16 353 32 518 10 374 10 546 39 535 42 1,492 22 506 80 290 14 279 16 305 1,040 25,455

6 232 23 478 2 144 8 408 13 211 18 441 38 361 89 3,998 5 390 3 189 5 357 25 363 24 837 14 478 22 278 26 506 16 298 6 301 5 490 6 557 6 266 6 346 7 260 11 311 18 438 8 344 7 662 19 215 44 1,519 25 505 9 136 6 172 18 373 538 16,864

Violent Crime

0 1 0 0 0 0 1 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 2 1 0 0 0 0 10


3 10 1 3 13 18 13 61 1 4 2 12 6 2 16 8 8 7 5 2 2 0 4 0 10 5 4 10 16 5 1 7 8 267

Other Crime

213 187 27 207 82 261 97 705 55 111 143 142 114 135 274 279 79 172 262 142 75 52 211 27 39 202 163 108 287 128 53 148 186 5,366

Homicide

Fraud & Forgery

20 71 6 39 44 35 151 323 10 9 24 87 40 17 24 108 22 16 19 19 19 15 44 9 33 20 15 80 118 42 40 15 25 1,559

Handling

Drug Offences

Criminal Damage

295 226 299 380 410 718 203 226 302 287 284 632 469 329 742 612 376 1,209 729 291 953 632 1,585 2,179 426 244 419 216 270 350 287 416 404 702 392 925 1,391 178 755 453 290 328 295 394 902 832 201 834 338 332 652 567 329 394 210 438 463 261 669 715 551 226 331 240 224 321 364 475 343 259 398 528 711 373 1,093 272 445 449 694 441 460 547 224 1,064 997 502 1,266 645 282 507 168 191 182 334 373 417 326 395 510 15,693 12,429 21,646

141 290 126 230 221 338 276 1,094 175 169 275 295 416 274 368 343 273 161 332 311 132 161 216 149 406 235 246 308 905 262 90 184 222 9,624

121 255 60 152 245 286 347 1,591 90 79 85 305 180 79 231 225 152 97 90 197 102 95 113 123 220 91 94 230 393 134 74 121 168 6,825

Total 2001/02

2,068 3,400 1,304 2,728 2,927 4,629 4,033 21,738 2,109 1,694 2,869 4,107 5,137 2,573 3,542 4,297 2,635 2,496 3,240 3,673 2,068 1,908 2,609 2,203 3,982 2,471 3,497 3,420 8,160 3,120 1,325 2,088 2,573 120,623

237


Burglary Elsewhere

WARDNAME

Burglary Dwelling


9 10 4 6 4 19 15 76 1 4 4 11 11 2 20 8 12 2 7 9 5 7 3 2 15 6 2 7 14 4 3 5 5 312

1 0 0 2 1 1 0 1 0 0 0 0 0 0 1 1 0 0 0 1 1 1 0 1 0 0 1 0 0 0 0 0 1 14

11 584 49 618 13 267 25 513 39 525 27 810 57 535 214 9,185 9 273 9 334 22 925 38 606 24 1,015 13 461 27 638 33 842 33 460 19 358 16 882 31 766 15 386 16 434 27 614 27 381 31 597 14 343 16 578 24 563 84 1,474 12 480 22 310 12 268 22 396 1,031 27,421

27 76 7 66 53 121 179 475 20 10 59 178 206 28 51 117 26 66 22 20 33 14 18 28 76 11 89 49 433 48 6 19 46 2,677

9 232 34 651 6 171 33 264 28 285 32 353 130 426 124 2,956 5 215 8 195 4 406 26 336 13 1,339 4 316 14 283 25 420 19 330 8 324 10 557 18 770 9 255 15 247 12 231 17 342 27 423 11 425 8 510 21 270 38 1,390 13 479 17 176 8 168 8 319 754 16,064

Violent Crime



Sexual Offences

Robbery

Other Theft

Other Crime

238 357 14 237 97 292 145 1,056 110 208 232 203 176 247 315 286 103 231 340 280 56 66 292 54 47 267 255 145 272 236 75 235 293 7,460

Homicide

Fraud & Forgery

27 62 38 31 36 64 290 429 15 11 83 124 49 12 29 122 24 34 17 27 10 8 62 15 24 15 40 83 178 61 10 15 42 2,087

Handling

Drug Offences

Criminal Damage

372 276 450 563 475 1,013 235 308 297 378 311 761 481 329 1,010 629 358 1,338 588 172 922 704 1,336 2,040 539 170 404 273 344 348 420 556 472 552 272 810 1,104 209 734 559 247 391 438 527 862 727 194 738 379 337 798 683 348 444 327 635 557 378 601 768 533 343 278 316 207 315 394 401 365 356 337 542 643 417 1,114 293 325 490 543 332 391 514 251 1,029 679 424 1,326 718 236 435 217 185 203 313 296 481 526 420 700 16,374 12,179 22,826

113 169 392 448 145 113 232 304 250 407 366 438 252 408 907 2,315 138 141 167 143 288 167 339 441 378 278 165 119 359 338 356 336 329 295 130 193 314 188 345 292 104 142 119 113 155 172 177 235 385 369 254 170 238 124 384 326 777 710 185 204 86 112 240 182 252 342 9,321 10,734

Total 2002/03

2,518 4,748 1,618 3,163 3,545 4,848 4,119 21,818 2,040 2,054 3,638 3,936 5,536 2,564 3,902 4,205 3,145 2,840 3,872 4,306 2,170 1,878 2,746 2,514 4,168 2,624 3,127 3,666 7,799 3,111 1,422 2,242 3,372 129,254

238


Burglary Elsewhere

WARDNAME

Burglary Dwelling


6 13 5 0 14 18 14 59 4 8 9 15 4 7 12 11 8 7 8 8 6 2 5 2 14 3 1 9 19 4 1 4 10 310

1 0 0 0 1 1 0 3 1 0 0 1 1 0 0 0 0 1 0 0 1 0 1 0 0 0 0 0 2 0 0 0 1 15

22 583 56 614 8 405 13 495 42 592 56 856 57 612 237 9,013 16 352 5 331 19 604 36 568 59 895 19 438 22 678 35 763 27 532 25 344 19 700 38 760 13 336 19 538 21 613 25 454 37 603 14 455 28 543 34 555 92 1,351 24 537 27 473 13 244 24 453 1,182 27,290

16 77 10 38 63 96 169 267 32 23 34 110 146 31 32 92 21 42 16 21 24 12 20 24 80 11 88 27 244 46 3 17 49 1,981

6 242 41 742 9 171 22 330 34 256 35 340 88 304 104 2,404 4 105 10 200 13 334 31 280 27 809 16 177 21 283 26 346 16 386 14 247 9 496 9 541 8 230 15 263 13 275 19 428 26 363 7 450 14 432 40 170 63 1,284 22 415 13 221 9 152 16 469 800 14,145

Violent Crime



Sexual Offences

Robbery

Other Theft

Other Crime

259 461 61 170 108 249 103 957 154 226 209 90 150 345 226 278 79 267 301 312 50 57 307 53 24 244 259 87 238 243 153 115 253 7,088

Homicide

Fraud & Forgery

14 85 15 31 27 90 308 482 11 10 30 104 53 8 56 124 26 30 12 38 20 15 81 10 36 12 26 101 173 63 43 44 34 2,212

Handling

Drug Offences

Criminal Damage

311 216 437 792 479 1,330 181 245 354 296 261 700 600 245 1,140 508 347 1,490 429 192 1,004 597 1,241 2,372 381 312 447 214 296 455 290 414 547 450 124 862 1,002 188 681 446 212 399 275 352 960 790 201 726 246 293 821 367 242 507 193 488 538 238 504 747 348 204 291 305 266 385 335 369 397 357 328 722 557 395 1,375 194 377 436 468 258 420 395 230 1,218 735 395 1,099 533 254 496 233 198 254 261 281 520 506 442 822 13,833 10,849 24,952

153 215 251 663 135 182 205 415 239 618 342 840 222 665 615 2,687 76 198 121 207 186 258 255 748 344 368 118 200 244 448 273 479 269 345 107 236 195 258 216 431 91 205 96 193 163 225 165 367 427 654 174 216 151 201 287 561 575 932 192 319 80 188 151 289 221 497 7,339 15,308

Total 2003/04

2,481 5,604 1,781 2,976 3,979 5,268 4,167 21,038 2,093 2,106 2,947 3,674 4,727 2,416 3,609 4,144 3,069 2,436 3,233 3,863 1,827 2,166 2,825 2,954 4,591 2,593 2,889 3,714 7,202 3,148 1,887 2,100 3,797 127,304

239


Burglary Elsewhere

WARDNAME

Burglary Dwelling


Note: * per 1,000 households

0.23 0.59 0.21 0.00 0.63 0.97 0.77 2.85 0.19 0.33 0.40 0.71 0.15 0.32 0.74 0.55 0.38 0.33 0.32 0.27 0.27 0.08 0.22 0.09 0.79 0.14 0.05 0.51 0.89 0.19 0.04 0.23 0.43

0.04 0.00 0.00 0.00 0.04 0.05 0.00 0.15 0.05 0.00 0.00 0.05 0.04 0.00 0.00 0.00 0.00 0.05 0.00 0.00 0.05 0.00 0.04 0.00 0.00 0.00 0.00 0.00 0.09 0.00 0.00 0.00 0.04

0.86 2.54 0.34 0.79 1.88 3.03 3.13 11.46 0.75 0.21 0.84 1.71 2.27 0.88 1.36 1.75 1.29 1.16 0.77 1.30 0.59 0.78 0.92 1.13 2.09 0.67 1.28 1.92 4.31 1.16 1.02 0.75 1.04

22.76 27.87 17.23 30.08 26.55 46.25 33.62 435.92 16.44 13.85 26.56 26.98 34.50 20.31 41.97 38.22 25.33 16.02 28.21 25.99 15.15 21.96 26.84 20.47 34.06 21.66 24.88 31.31 63.24 25.92 17.87 14.15 19.60

0.62 3.49 0.43 2.31 2.83 5.19 9.28 12.91 1.49 0.96 1.50 5.23 5.63 1.44 1.98 4.61 1.00 1.96 0.64 0.72 1.08 0.49 0.88 1.08 4.52 0.52 4.03 1.52 11.42 2.22 0.11 0.99 2.12

0.23 1.86 0.38 1.34 1.52 1.89 4.83 5.03 0.19 0.42 0.57 1.47 1.04 0.74 1.30 1.30 0.76 0.65 0.36 0.31 0.36 0.61 0.57 0.86 1.47 0.33 0.64 2.26 2.95 1.06 0.49 0.52 0.69

9.45 33.68 7.28 20.06 11.48 18.37 16.70 116.27 4.91 8.37 14.69 13.30 31.18 8.21 17.52 17.33 18.38 11.50 19.99 18.50 10.37 10.74 12.04 19.30 20.50 21.42 19.80 9.59 60.10 20.03 8.35 8.81 20.29

5.97 11.39 5.74 12.46 10.72 18.48 12.19 29.74 3.55 5.06 8.18 12.11 13.26 5.47 15.10 13.67 12.81 4.98 7.86 7.39 4.10 3.92 7.14 7.44 24.12 8.28 6.92 16.19 26.91 9.27 3.02 8.75 9.56

Violent Crime

Theft of MV

Theft from MV

Sexual Offences

Robbery

Other Theft

Other Crime

10.11 20.92 2.60 10.33 4.84 13.45 5.66 46.29 7.19 9.46 9.19 4.28 5.78 16.00 13.99 13.93 3.76 12.43 12.13 10.67 2.26 2.33 13.44 2.39 1.36 11.61 11.87 4.91 11.14 11.73 5.78 6.67 10.95

Homicide

0.55 3.86 0.64 1.88 1.21 4.86 16.92 23.31 0.51 0.42 1.32 4.94 2.04 0.37 3.47 6.21 1.24 1.40 0.48 1.30 0.90 0.61 3.55 0.45 2.03 0.57 1.19 5.70 8.10 3.04 1.62 2.55 1.47

Handling

Fraud & Forgery

Criminal Damage 17.06 60.36 15.06 42.54 51.13 80.51 55.15 114.72 20.88 19.04 24.05 40.95 26.25 18.50 59.42 36.37 39.09 23.61 21.68 25.55 13.13 15.72 17.39 32.55 77.67 20.75 19.25 68.72 51.44 23.94 9.59 30.15 35.56

Drugs Offences

8.43 21.74 10.42 15.86 10.99 18.75 10.55 60.02 14.58 12.39 18.20 5.89 7.25 9.83 21.79 10.07 13.95 11.27 19.67 17.24 9.20 10.86 16.16 14.79 22.31 17.94 11.82 12.98 18.49 12.26 7.48 16.29 19.12

8.39 30.09 7.74 25.22 27.72 45.39 36.53 129.96 9.25 8.66 11.34 35.53 14.18 9.28 27.73 23.99 16.43 10.99 10.40 14.74 9.25 7.88 9.85 16.55 36.94 10.28 9.21 31.65 43.62 15.40 7.10 16.75 21.50

240

29.17 85.21 18.76 41.97 66.55 61.43 53.97 59.37 41.34 22.30 30.53 56.50 104.54 50.36 39.63 90.49 28.51 41.43 18.53 19.32 36.65 29.48 34.79 38.59 72.08 22.22 54.21 53.11 67.72 57.34 21.75 36.74 51.43

Appendix B


Burglary Elsewhere

WARDNAME

Burglary Dwelling *

Appendix B: Crime rate per 1000 population by crime type at ward level (2003/04)

Note: * per 1,000 households

0.35 0.45 0.17 0.36 0.18 1.03 0.82 3.68 0.05 0.17 0.18 0.52 0.42 0.09 1.24 0.40 0.57 0.09 0.28 0.31 0.23 0.29 0.13 0.09 0.85 0.29 0.09 0.39 0.66 0.19 0.11 0.29 0.22

0.04 0.00 0.00 0.12 0.04 0.05 0.00 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.06 0.05 0.00 0.00 0.00 0.03 0.05 0.04 0.00 0.05 0.00 0.00 0.05 0.00 0.00 0.00 0.00 0.00 0.04

0.43 2.22 0.55 1.52 1.75 1.46 3.13 10.35 0.42 0.38 0.97 1.81 0.93 0.60 1.67 1.65 1.57 0.88 0.64 1.06 0.68 0.65 1.18 1.22 1.75 0.67 0.73 1.35 3.93 0.58 0.83 0.70 0.95

22.80 28.05 11.36 31.18 23.54 43.77 29.39 444.23 12.75 13.98 40.67 28.79 39.12 21.38 39.49 42.18 21.90 16.67 35.55 26.20 17.41 17.72 26.89 17.18 33.72 16.33 26.49 31.76 68.99 23.16 11.71 15.54 17.13

1.05 3.45 0.30 4.01 2.38 6.54 9.83 22.97 0.93 0.42 2.59 8.46 7.94 1.30 3.16 5.86 1.24 3.07 0.89 0.68 1.49 0.57 0.79 1.26 4.29 0.52 4.08 2.76 20.27 2.32 0.23 1.10 1.99

0.35 1.54 0.26 2.01 1.26 1.73 7.14 6.00 0.23 0.33 0.18 1.24 0.50 0.19 0.87 1.25 0.90 0.37 0.40 0.62 0.41 0.61 0.53 0.77 1.53 0.52 0.37 1.18 1.78 0.63 0.64 0.46 0.35

9.06 29.55 7.28 16.04 12.78 19.07 23.40 142.97 10.04 8.16 17.85 15.96 51.61 14.66 17.52 21.04 15.71 15.09 22.45 26.33 11.50 10.08 10.12 15.42 23.89 20.23 23.37 15.23 65.06 23.12 6.65 9.74 13.80

4.41 17.79 6.17 14.10 11.21 19.78 13.84 43.87 6.45 6.99 12.66 16.10 14.57 7.65 22.22 17.83 15.67 6.05 12.66 11.80 4.69 4.86 6.79 7.98 21.75 12.09 10.91 21.66 36.37 8.93 3.25 13.91 10.90

Violent Crime

Theft of MV

Theft from MV

Sexual Offences

Robbery

9.29 16.20 0.60 14.40 4.35 15.78 7.96 51.07 5.14 8.71 10.20 9.64 6.78 11.46 19.50 14.33 4.90 10.76 13.70 9.58 2.53 2.69 12.79 2.43 2.65 12.71 11.68 8.18 12.73 11.39 2.83 13.62 12.68

Other Theft

1.05 2.81 1.62 1.88 1.61 3.46 15.93 20.75 0.70 0.46 3.65 5.89 1.89 0.56 1.80 6.11 1.14 1.58 0.69 0.92 0.45 0.33 2.72 0.68 1.36 0.71 1.83 4.68 8.33 2.94 0.38 0.87 1.82

Other Crime

Fraud & Forgery

17.57 45.97 12.64 46.25 45.30 72.30 50.64 98.67 18.87 14.57 20.75 38.48 28.29 18.13 53.36 36.97 38.00 20.68 22.45 26.27 12.54 12.86 15.98 24.44 62.92 23.32 17.92 58.05 62.07 20.99 7.67 27.89 30.28

Homicide

Drugs Offences

10.77 21.56 13.10 18.90 14.75 19.34 9.45 64.62 7.94 14.40 24.45 12.92 8.06 11.46 32.62 9.72 16.05 16.21 25.59 20.55 15.47 8.45 17.56 15.20 23.55 15.47 15.21 14.16 19.85 11.39 6.99 17.16 18.17

Handling

Criminal Damage

34.89 60.57 24.36 53.59 53.35 76.06 73.97 70.01 58.49 28.44 44.21 69.31 115.18 63.11 63.11 83.28 43.93 77.11 31.40 30.68 56.13 30.55 40.91 38.48 83.21 33.57 62.90 69.11 62.56 77.25 20.25 44.06 53.46

6.60 20.33 4.81 18.48 18.25 23.67 22.41 111.97 6.59 5.99 7.34 20.95 10.71 5.52 20.92 16.83 14.05 8.99 7.58 9.99 6.40 4.61 7.53 10.60 20.84 8.09 5.68 18.39 33.23 9.85 4.23 10.55 14.80

241


Burglary Elsewhere

WARDNAME

Burglary Dwelling *

Appendix B: Crime rate per 1000 population by crime type at ward level (2002/03)

Appendix C: Demographic and socio-economic variables from the 2001 Census

Ward

AllPeople16to74

AllHHs

HH density

PopDensity

Number Student

% of Student

NumberOfHighClass

% of HighClass

NumberOfCarPerHHs

AllCarInArea

Rented

Aireborough

18581

10663

507

1218.12

985

5.3

5840

31.43

1.18

12579

2428

Armley

15920

9295

1718

4072.83

996

6.26

3171

19.92

0.76

7068

3960

17225

9647

118

287.97

721

4.19

4727

27.44

1.26

12143

2111

11588

7053

1610

3756.62

729

6.29

2028

17.5

0.67

4730

2834

Bramley

15988

9016

1385

3425.19

1658

10.37

2910

18.2

0.78

7059

3840

Burmantofts

12790

8270

1716

3839.63

727

5.68

1818

14.21

0.54

4494

5347

Chapel Allerton

13083

7949

1736

3975.11

1058

8.09

3912

29.9

0.72

5759

3597

City and Holbeck

14869

10056

1018

2092.71

1537

10.34

2282

15.35

0.49

4907

6523

Cookridge

15334

9216

827

1921.45

1074

7

5413

35.3

1.15

10633

2413

Garforth and Swillington

17354

9598

331

823.01

778

4.48

4558

26.26

1.18

11301

1865

Halton

16831

9500

496

1188.19

821

4.88

4935

29.32

1.12

10601

1028

Harehills

14062

7964

3040

8035.11

1321

9.39

2137

15.2

0.61

4886

4104

Headingley

23219

9585

3485

9434.55

14029

60.42

3839

16.53

0.78

7517

7181

Horsforth

16059

8857

766

1865.22

1462

9.1

5938

36.98

1.23

10916

1868

Hunslet

10975

6940

923

2148.27

473

4.31

1359

12.38

0.58

3998

4113

Kirkstall

15481

8730

1823

4167.85

2981

19.26

4048

26.15

0.77

6715

4482

Middleton

14781

8628

626

1522.99

614

4.15

2901

19.63

0.86

7455

3644

Moortown

14864

8858

1637

3969.13

1080

7.27

5382

36.21

1.08

9610

2303

Morley North

18375

10414

625

1490.21

796

4.33

5054

27.5

1.12

11701

2097

Morley South

21236

12320

599

1422.13

914

4.3

5653

26.62

1.04

12828

3586

North

15612

9496

223

519.59

1000

6.41

5972

38.25

1.24

11795

2389

Otley and Wharfedale

17643

10345

240

568.09

886

5.02

6100

34.57

1.26

13034

2214

Pudsey North

16941

9630

711

1685.24

875

5.17

5311

31.35

1.15

11087

1711

Pudsey South

15921

9252

987

2366.92

770

4.84

3848

24.17

1

9241

2898

Richmond Hill

12103

7727

819

1877.41

605

5

1632

13.48

0.55

4223

4623

Rothwell

15274

8729

420

1010.58

680

4.45

4008

26.24

1.09

9523

2371 1767

15515

8633

744

1879.67

1199

7.73

6663

42.95

1.27

10953

11572

7437

1734

4131.7

553

4.78

1387

11.99

0.58

4333

4973

University

17504

10854

2238

4404.95

6994

39.96

2572

14.69

0.42

4524

9078

Weetwood

15727

9295

1313

2926.69

3600

22.89

5275

33.54

0.91

8457

4049

Wetherby

19328

10714

139

344.3

1000

5.17

6974

36.08

1.43

15309

1939

Whinmoor

12191

7104

563

1366.8

666

5.46

2388

19.59

0.9

6383

2999

Wortley

16533

9839

803

1885.32

712

4.31

3219

19.47

0.86

8481

3628

242

Roundhay Seacroft

Appendix C

Barwick and Kippax Beeston

Appendix C: Demographic and socio-economic variables from the 2001 Census Ward Aireborough Armley Barwick and Kippax

% of Rented

MaleYoungAdult

MaleYoungAdultUnEmployed

YoungAdult

UnEmployed

UnemploymentRate

MaleUnEmployed

IndexOfMultipleDeprivation

22.77

1177

78

2346

410

2.21%

261

42.6

1213

102

2711

608

3.82%

415

10.15 33.2

21.88

1066

52

2093

421

2.44%

264

16.99

Beeston

40.18

956

86

1965

544

4.69%

381

40.73

Bramley

42.59

1061

128

3199

626

3.92%

397

35.06

Burmantofts

64.66

919

130

1990

775

6.06%

507

53.66

Chapel Allerton

45.25

1009

123

2122

671

5.13%

468

42.5 55.41

City and Holbeck

64.87

1490

171

3190

989

6.65%

692

Cookridge

26.18

1093

53

2095

345

2.25%

234

12.07

Garforth and Swillington

19.43

1118

63

2119

332

1.91%

211

13.87 10.65

Halton

10.82

1022

60

1993

339

2.01%

226

Harehills

51.53

1329

169

2868

889

6.32%

597

54.07

Headingley

74.92

7431

118

15732

513

2.21%

354

16.17

Horsforth

21.09

1013

60

2512

303

1.89%

187

6.89

Hunslet

59.27

813

132

1770

617

5.62%

413

47.97

Kirkstall

51.34

2299

87

4616

564

3.64%

396

29.22

Middleton

42.23

1021

119

2151

653

4.42%

424

37.04

Moortown

26

1073

76

2056

423

2.85%

273

16.55

Morley North

20.14

1170

50

2365

363

1.98%

202

15.04

Morley South

29.11

1374

105

2806

533

2.51%

343

18.65

North

25.16

952

74

1887

380

2.43%

247

11.66

21.4

1064

60

2060

348

1.97%

211

8.8

Pudsey North

17.77

1068

55

2180

351

2.07%

214

11.98

Pudsey South

31.32

1115

66

2195

420

2.64%

277

19.56

Richmond Hill

59.83

922

156

1925

722

5.97%

480

52.52

Rothwell

27.16

936

71

1872

417

2.73%

271

16.87

Roundhay

20.47

1143

61

2213

374

2.41%

252

11.99

Seacroft

66.87

930

141

2002

731

6.32%

465

55.07

Otley and Wharfedale

University

83.64

3389

148

7671

881

5.03%

595

47.76

Weetwood

43.56

1825

51

4539

353

2.24%

217

15.71

Wetherby

18.1

959

47

2221

329

1.70%

213

6.91

Whinmoor

42.22

914

85

1816

420

3.45%

275

30.28

Wortley

36.87

1152

131

2326

626

3.79%

417

31.61

243

Appendix D- Model File

244

Appendix D C:\Documents and Settings\geock\Desktop\MyWork\TOK\My Model\InputFile_Constraints C:\Documents and Settings\geock\Desktop\MyWork\TOK\My Model\BCS2001.csv C:\Documents and Settings\geock\Desktop\MyWork\TOK\My Model\NewGroupNumber_16-74 in HHs.csv

Selected Area Code Start Selected Area Code End Column&Name,Male_16-24_couple OR,SEX,=,1,INDIVI#DUAL AND,16.0,=

Modelling Crime: A Spatial Microsimulation Approach

Modelling Crime: A Spatial Microsimulation Approach

Suggest Documents

Modelling Crime: A Spatial Microsimulation Approach

The energy costs of commuting: a spatial microsimulation approach

A Spatial Microsimulation Approach to the Analysis of ... - Frontiers

OO dynamic spatial microsimulation

Comparing different spatial microsimulation frameworks

Microsimulation Modelling for Policy Analysis

MICROSIMULATION MODELLING OF INTELLIGENT SPEED ...

A spatial approach to combatting wildlife crime - Wiley Online Library

Building a Microsimulation Model for Crime in ... - Semantic Scholar

Using geographical information systems and spatial microsimulation ...

Measuring Small Area Inequality Using Spatial Microsimulation ...

Spatial microsimulation estimates of household ... - STICERD - LSE

Spatial microsimulation estimates of household ... - STICERD - LSE

towards a 3d spatial urban energy modelling approach - ISPRS Annals

Nets versus spraying: A spatial modelling approach reveals ... - Plos

A poisson regression approach for modelling spatial ... - ScienceOpen

A Pooling Approach to Modelling Spatial Relations for Image Retrieval ...

TOWARDS A 3D SPATIAL URBAN ENERGY MODELLING APPROACH

Building a spatial microsimulation-based planning support system for ...

simSALUD: A Web-based Spatial Microsimulation to Model the Health ...

A Fine Grained Hybrid Spatial Microsimulation Technique for ...

JAMSIM: a Microsimulation Modelling Policy Tool - Semantic Scholar

JAMSIM: a Microsimulation Modelling Policy Tool - Semantic Scholar

Building a Static Farm Level Spatial Microsimulation Model ...