Structural road accident models : the international DRAG family 1. ed

Contents v Contributing authors xi Foreword: on a manuscript of 1984 xiii Sylvain Lassarre Research support, and more xvii Marc Gaudry, Sylvain Lassarre PART I. NATIONAL AND REGIONAL MODELS 1 MULTIPLE LEVELS, DAMAGES, FORMS, MOMENTS AND VARIABLES IN 1 ROAD ACCIDENT MODELS Marc Gaudry L L Introduction: the «Modening Quartet» in this book 1 L2. Problem Formulation 2 1.2.1. A multilevel structure; a multidamage application 2 1.2.2. Perspectives on problem structure 6 1.3. The quantification of effects 9 1.3.1. From fixed to flexible mathematical form 9 1.3.2. From monotonic to multitonic forms: the case of alcohol 14 1.4. Variables: multimoment, multivariate 19 1.4.1. The dependent variable: from observations to moments 19 1.4.2. The explanatory variables: not a triad, but a quatrain 24 1.4.3. Is pregnancy a risk factor? 27 1.5. Other modelling dimensions of interest 29 1.5.1. Multidata 29 1.5.2. Multiple documentation of reference results 30 1.6. Conclusion: matching tools to questions 31 1.7. Post Scriptum: model acronyms 31 1.8. References 32 2 THE DRAG-2 MODEL FOR QUEBEC 37 Frangois Fournier, Robert Simard 2.1. Introduction 37 2.2. The structure of the DRAG-2 model 3 8 2.2.1. A diagram of the model structure 3 8 2.2.2. Dependent variable graphs 40 2.2.3. The matrix of direct effects of independent variables over 45 dependent variables 2.3. Results on mathematical form and particular variables 49 2.3.1. Econometric results 5 0 2.3.2. Results on elasticity 51 2.3.3. Other results: forecasts for the period of 1997-2004 63 vi STRUCTURAL ROAD ACCIDENT MODELS 2.4. Other developments 66 3 THE SNUS-2.5 MODEL FOR GERMANY 67 Ulrich Blum, Marc Gaudry 3.L Context 67 3.2. Structure of model 68 3.2.1. The dependent variables 68 3.2.2. Visual analysis of the dependent variables 69 3.2.3. Matrix of direct effects 70 3.3. Results and their interpretation 73 3.3.1. Statistical results 73 3.3.2. Economic results: overall specific results 73 3.3.3. Decomposition of the impact by variable: results common to 75 other models 3.3.4. Results for other variables 79 3.4. Deriving other interesting results 82 3.4.1. The analysis of victims: direct, indirect and total elasticities 82 3.4.2. Multiple moments and their marginal rates of substitution 84 3.4.3. Marginal rates of substitution with comparable accident data 88 3.4.4. Marginal rates of substitution with disaggregated accident data 92 3.5. Policy implications 94 3.5.1. Higher prices save energy and lives 94 3.5.2. Risk substitution in terms of first moments 94 3.6. References 94 4 THE TRULS-1 MODEL FOR NORWAY 97 Lasse Fridstrom 4.1. Introduction 97 4.2. Structure of the Model TRULS-1 98 4.2.1. Dependent variables: definitions and relations 99 4.2.2. Visual analysis of dependent variables 100 4.2.3. Matrix of direct effects 107 4.2.4. The casualty subset test 108 4.3. Results on form and selected explanatory variables 111 4.4. References 126 5 THE DRAG-STOCKHOLM-2 MODEL 127 Goran Tegner, Ingvar Holmberg, Vesna Loncar-Lucassi, Christian Nilsson 5.1. Introduction 127 5.1.1. The Dennis agreement 127 5.1.2. The MAD-project 128 5.1.3. The concept of zero fatality 12 8 5.2. Structure of the DRAG model for Stockholm county 5.2.1. Introduction 128 Contents vii 5.2.2. Dependent variables: definitions and relations 129 5.2.3. Visual analysis of dependent variables 129 5.2.4. Matrix of direct effects 13 3 5.3. Model form and explanatory variables 13 4 5.3.1. Summary of econometric results 134 5.3.2. The demand for road use 134 5.3.3. Comparison between estimated and actual demand for road use 13 9 5.3.4. The contribution of road infrastructure to road traffic growth 139 5.4. The Road accident frequency and gravity models 140 5.4.1. Economic activities 141 5.4.2. Quality of vehicle fleet 141 5.4.3. Road network data 142 5.4.4. Weather data 142 5.4.5. Intervention measures 142 5.4.6. Gasoline price 142 5.5. The DRAG-Stockholm-2 model 143 5.5.1. The new model specification 143 5.5.2. Comparison of results between the "old" and "new" specification 143 5.6. Comparison of actual and estimated accident risks 146 5.7. Specific results on the DRAG-Stockholm model 148 5.8. Points of interest and conclusion 149 5.8.1. Alcohol consumption: the J-shaped relationship 150 5.8.2. Medicine consumption 151 5.8.3. Pregnancy—a new risk factor 151 5.8.4. Conclusions 154 5.9. References 154 6 THE TAG-1 MODEL FOR FRANCE 157 Laurence Jaeger, Sylvain Lassarre 6.1. Introduction 15 7 6.2. Structuring the TAG model 159 6.3. Econometric form of the TAG model 163 6.4. The estimates produced by the TAG model 167 6.4.1. Model of road transport demand 168 6.4.2. Constructing a model of average speed 170 6.4.3. Analysis of the results by risk indicator 173 6.4.4. Analysis of results by explanatory factor 175 6.5. Conclusion 181 6.6. References 182 7 THE TRACS-CA MODEL FOR CALIFORNIA 185 Patrick McCarthy lA. Introduction 185 viii STRUCTURAL ROAD ACCIDENT MODELS 7.2. TRACS-CA model structure 186 7.2.1. Exposure and crash losses 186 7.2.2. Historical trends 188 7.2.3. Determining variables included in the TRACS-CA structure 191 7.3. Estimation results 194 7.3.1. Statistical summary 194 7,32. Common variable results 195 7.3.3. Specific variable results 198 7.3.4. Further results 200 7.4. Discussion and future directions 200 7.5. References 203 8 COMPARING SIX DRAG-TYPE MODELS 205 Nicolas Chambron 8.1. Risk exposure 206 8.2. Driver behaviour 209 8.2.1. Speed 209 8.2.2. Seatbelt wearing 212 8.2.3. Consumption of alcohol 214 8.2.4. Consumption of medicines 216 8.3. Economic variables 217 8.3.1. Households' economic and financial situation 217 8.3.2. Fuel prices 220 8.3.3. Competing supply from public transport 222 8.4. Conclusion 222 8.5. References 224 PART II: OTHER MODELS AND ISSUES 9 THE ROAD, RISK, UNCERTAINTY AND SPEED 225 Marc Gaudry, Karine Vernier 9.1. Risk, uncertainty and observed road accident outcomes 225 9.2. Model structure: simultaneity and perceived risk 226 9.3. Selected results: accident frequency and severity 230 9.4. Selected results: speed 233 9.5. Conclusion 235 9.6. References 235 10 THE RES MODEL BY ROAD TYPE IN FRANCE 237 Ruth Bergel, Bernard Girard 10.1. Introduction 237 10.2. Structure of the model 23 8 10.2.1. General outline 238 10.3. 10.4. 10.5. 10.2.2. The data base 10.2.3. Economic formulation 10.2.4. Econometric specification 10.2.5. Algorithm The Results 10.3.1. Tests of functional form 10.3.2. Measuring elasticities 10.3.3. Short and medium term simulations Conclusion References Contents ix 239 242 243 244 245 245 246 248 248 249 11 POSTFACE AND PERSPECTIVES 251 Sylvain Lassarre 11.1. Relevance of models for understanding the influence of risk factors 252 11.2. Outlook for research in constructing risk models 25 8 11.2.1. Data extraction 258 11.2.2. Adding levels to the structure 259 11.2.3. Breakdown of indicators by user and road types 260 11.2.4. Disaggregation by location or vehicle x driver 260 11.3. Relevance of the models for managing road safety 260 11.4. References 262 PART III: ALGORITHMS AND DETAILED MODEL OUTPUTS 12 THE TRIO LEVEL-1.5 ALGORITHM FOR BC-GAUHESEQ REGRESSION Tran Liem, Marc Gaudry, Marcel Dagenais, Ulrich Blum 12.1. Introduction and statistical model 263 12.1.1. Introduction 263 12.1.2. Log-likelihood function 265 12.1.3. Computational aspects 270 12.1.4. Model types 275 12.1.5. Model estimation 276 12.2. Estimation results 278 12.2.1. Definitions of moments of the dependent variable 278 12.2.2. Derivatives and elasticities of the sample and expected values 283 of the dependent variable 12.2.3. Derivatives and elasticities of the standard error of the dependent 288 variable 12.2.4. Derivatives and elasticities of the skewness of the dependent 293 variable 12.2.5. Ratios of derivatives of the moments of the dependent variable 296 12.2.6. Evaluation of moments, their derivatives, rates of substitution 301 and elasticities 12.2.7. Student's t-statistics 305 X STRUCTURAL ROAD ACCIDENT MODELS 12.2.8. Goodness-of-fit measures 306 12.3. Special options 309 12.3.1. Correlation matrix and table of variance-decomposition proportions 3 09 12.3.2. Analysis of heteroskedasticity of the residuals 310 12.3.3. Analysis of autocorrelation of the residuals 311 12.3.4. Forecasting: maximum likelihood and simulation forecasts 314 12.4. References 321 13 THE IRPOSKML PROCEDURE OF ESTIMATION 325 Lasse Fridstrom 13.1. Accident frequency model specification 325 13.2. Severity model specification 331 13.3. References 334 14 TURNING BOX-COX INCLUDING QUADRATIC FORMS IN REGRESSION 335 Marc Gaudry, Ulrich Blum, Iran Liem 14.1. Model with two Box-Cox transformations on a same independent variable 335 14.1.1. Solution 335 14.1.2. First-order conditions 336 14.1.3. Second-order conditions 337 14.1.4. Special case: quadratic form 339 14.2. Model with powers A^ and A2 only on a same independent variable 339 14.2.1. First-order conditions 340 14.2.2. Second-order conditions 341 14.2.3. Special case: quadratic form 343 14.3. Two-step transformations on a same independent variable 343 14.4. References 346 15 APPENDIX 1. DETAILED MODEL OUTPUTS 347 Marc Gaudry, Sylvain Lassarre