The Effect of Police Enforcement on Road Traffic Accidents Andrew Scott Submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy Awarded by Edinburgh Napier University April 2010 Abstract The primary goal of this thesis is to investigate the effectiveness of police enforcement on Road Traffic Accidents; specifically, ‘Does police enforcement activity have any real effect on levels of Killed and Seriously Injured road traffic accidents?’ Data relating to forty one Police Force Areas in England and Wales was analysed by means of Zero Truncated Poisson regression, Cluster Analysis and Multilevel Modelling. Enforcement measures available to the police, for which data is available in this report, range from Prosecutions and Fixed Penalty Notices to Written Warnings and Vehicle Defect Rectification Notices. Results from the Zero Truncated Poisson regression models have significant effects (P < .05), in relation to both contemporary and lagged Annual data and contemporary Quarterly data, for all proxy variables except Prosecutions. Significant effects (P < .05) are also found for Fixed Penalty Notices lagged by two quarters, Vehicle Defect Rectification Notices and speeding related Fixed Penalty Notices lagged by one quarter. Results from Cluster Analysis verify the trend linking increased police enforcement with decreasing KSI rates. Clusters derived from population based KSI rates are more clearly defined than those using Vehicle kilometres travelled based KSI rates. Multilevel modelling found significant fixed effects (P < .05) for Fixed Penalty Notices and speeding related Fixed Penalty Notices in relation to both derived and regional clusters, linking an increase in enforcement to a decrease in the overall KSI rate. There would seem to be little doubt, based on the findings of this report, that higher levels of police enforcement, as measured here, lead to decreasing numbers of KSI accidents. ii Acknowledgements I am extremely grateful to all the people who have helped me with this thesis. In particular I would like to thank my supervisors Professor Robert Raeside and Phil Darby for their support, encouragement and boundless enthusiasm. Special thanks to my partner Hayley, always there, always supportive. Thank you, it meant so much. Also, to my daughter Amy, the sunshine in my life, just for being Amy. iii Author’s Declaration I declare that no material contained in this thesis has been used in any other submission for an academic award and that the work presented is my own. This thesis is copyright material and no quotation or excerpt from it may be published without proper acknowledgement Andrew Scott iv Contents List of Acronyms and Abbreviations ix List of Variables Descriptors ix Chapter 1: Introduction 1 1.2 Aims 3 1.3 Research Approach 3 1.4 Data 4 1.5 Analysis Plan 5 1.6 Structure of Thesis 7 Chapter 2: Literature Review 8 2.1 Introduction 8 2.2 Enforcement 9 2.3 Speed Cameras 13 2.4 Traffic Light Cameras 16 2.5 Seat Belt Use 17 2.6 Drink Driving 20 2.7 Attitudes and Behaviour 26 2.8 Road Safety Media Campaigns 30 Chapter 3: Research Methods 32 3.1 Introduction to Research Methods 32 3.2 Data Collection 34 3.3 Data Preparation 38 3.4 Exploratory Data Analysis 40 3.4.1 Introduction 40 3.4.2 Trends in Data 43 3.4.3 Annual Trends in Data 44 3.4.4 Quarterly Trends in Data 50 3.5 Proposed Methods of Analysis 51 v Chapter 4: Poisson Regression Analysis 54 4.1 Introduction to Poisson Regression Analysis 54 4.2 Data Analysis using Poisson Regression 57 4.3 Development of Poisson Regression Models 59 4.3.1 Modelling Annual Accident Data 59 4.3.2 Analysis of Annual Data 60 4.3.3 Model Fitting 61 4.4 Modelling Quarterly Accident Data 75 Chapter 5: Cluster Analysis 87 5.1 Introduction to Cluster Analysis 87 5.2 Ward Method Cluster Analyses on KSI Rate by Population and FPN_1000’s 88 5.2.1 Analysis of Variance 92 5.3 Ward Method Cluster Analysis on KSI Rate by Vkm and FPN_1000’s 93 5.3.1 Analysis of Variance for KSI RATE by Vkm and FPN_1000‟s 94 5.4 Fuzzy C-means Cluster Analysis 96 Chapter 6: Multilevel Modelling 99 6.1 Introduction to Multilevel Modelling 99 6.2 Multilevel Modelling of Accident Data 100 6.2.1 Multilevel Models using FPN‟s 100 6.2.2 Multilevel Models using ZFPN_G16_1000‟s 103 6.3 Multilevel Models based on Regional Clusters 105 6.3.1 Multilevel Models using ZFPN_1000‟s on Regional Clusters 106 6.3.2 Multilevel Models using ZFPN_G16_1000‟s 108 6.4 Discussion of Results relating to Annual Data 111 6.5 Multilevel Modelling of Quarterly Data 114 6.5.1 Quarter 3 Multilevel Models using ZFPN_1000‟s 115 vi 6.5.2. Quarter 3 Multilevel Models using ZLag1FPN_1000‟s and ZLag1_FPN_1000‟s 117 6.5.3 Quarter 3 Multilevel Models using ZFPN_G16_1000‟s 120 6.5.3.1 Quarter 3 Multilevel Models using ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16_1000‟s 120 6.5.4 Quarter 4 Multilevel Models using ZFPN_1000‟s 121 6.5.4.1 Quarter 4 Multilevel Models using ZLag1FPN_1000‟s and ZLag1_FPN_1000‟s 123 6.5.5 Quarter 4 Multilevel Models using ZFPN_G16_1000‟s 123 6.5.5.1 Quarter 4 Multilevel Models using ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16_1000‟s 124 6.6 Multilevel Models on Regional Clusters 124 6.6.1 Quarter 3 Multilevel Models using ZFPN_1000‟s on Regional Clusters 125 6.6.2. Quarter 3 Multilevel Models using ZLag1_FPN_1000‟s and ZLag1_FPN_1000‟s on Regional Clusters 127 6.6.3 Quarter 3 Multilevel Models using ZFPN_G16_1000‟s on Regional Clusters 128 6.6.3.1 Quarter 3 Multilevel Models using ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16 _1000‟s on Regional Clusters 129 6.6.4 Quarter 4 Multilevel Models using ZFPN_1000‟s on Regional Clusters 129 6.6.4.1 Quarter 4 Multilevel Models using ZLag1FPN_1000‟s and ZLag1_FPN_1000‟s on Regional Clusters 131 6.6.5 Quarter 4 Multilevel Models using ZFPN_G16_1000‟s on Regional Clusters 131 6.6.5.1 Quarter 4 Multilevel Models using ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16_1000‟s 132 vii 6.7 Discussion of Results relating to Quarterly Data 133 6.7.1 Results from Analysis of Derived Clusters 133 6.7.2 Results from Analysis of Regional Clusters 133 Chapter 7: Discussion and Conclusions 135 7.1 Introduction 135 7.2 Discussion of Results 136 7.2.1 Results from Zero Truncated Poisson Regression 136 7.2.2 Discussion of Results from Cluster Analysis 143 7.3 Multilevel Modelling 146 7.3.1 Discussion of Multilevel Modelling Results 146 7.4 Limitations 148 7.5 Contribution to Knowledge 149 7.6 Recommendations 150 7.6.1 Recommendations for Practice and Policy 150 7.6.2 Recommendations for Further Research 150 7.7 Conclusions 151 Appendices 152 Appendix 3 152 Appendix 4 159 Appendix 5 189 Appendix 6a 192 Appendix 6b 198 Appendix 6c 208 References 217 viii Acronyms and Abbreviations DV Dependent Variable FCM Fuzzy C-means Clustering FPN Fixed Penalty Notices FPN G16 Speeding Related Fixed Penalty Notices GA Geographical Area IMD Index of Mean Deprivation KSI Killed and Serious Injury/Killed and Seriously Injured NB Negative Binomial PFA Police Force Area Pop Population Pros Prosecutions QTR Quarter RTA Road Traffic Accident Std Standard VDRN Vehicle Defect Rectification Notices Vkm Vehicle Kilometres travelled WW Written Warnings ZTP Zero Truncated Poisson Variable Descriptors IMD Index of Multiple Deprivation Geographical Area sqkm Geographical area of each PFA Percent Motorway Percentage of total motorway in each PFA lnpop Log of Population lnvkm Log of Vehicle kilometres All Penalties Sum of all Penalties Lag1 All Penalties Annual one Year Lag FPN Fixed Penalty Notices Lag FPN Annual one Year Lag Lag 1 FPN One quarter, 3 months, Lag ix Lag 2 FPN Two quarter, 6 months, Lag FPN_G16 Speeding related FPN‟s Lag FPN 16 Annual one Year Lag Lag 1 FPN G16 One quarter, 3 months, Lag Lag 2 FPN G16 Two quarter, 6 months, Lag PROSECUTIONS Successful Prosecutions Lag Prosecutions Annual one Year Lag Lag 1 Prosecutions One quarter, 3 months, Lag Lag 2 Prosecutions Two quarter, 6 months, Lag VDRN Vehicle Defect Rectification Notices Lag VDRN Annual one Year Lag Lag 1 VDRN One quarter, 3 months, Lag Lag 2 VDRN Two quarter, 6 months, Lag WW Written Warnings Lag WW Annual one Year Lag Lag 1 WW One quarter, 3 months, Lag Lag 2 WW Two quarter, 6 months, Lag x