The weather and the city: road safety 1. Introduction perspective - PDF document

International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety Presentation Structure The weather and the city: road safety 1. Introduction perspective 2. Data description 3. Method 4. Results Sara Ferreira Isabel Iglesias António Lobo 5. Conclusions and further developments António Couto 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety 1. Introduction 2. Data description Crash data: (Police records) Porto city, Portugal Motivation: 23,155 crashes without victims 5 years (2001-2005) � 1 st study in Portugal – crashes versus weather. � Contribution to the literature o Urban areas (Theofilatos and Yannis, 2014): Objectives � Analyze the impact of the weather on crashes. � Distinguish crashes with and without victims � Investigate lagged effects � Add road classification 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety 2. Data description 2. Data description Crash data: (Police records) Porto city. Portugal Meteorological description: 4,695 crashes with victims 5 years (2001-2005) Portugal is located in the west coast of the Iberian Peninsula and it is characterized by a temperate climate with a dry season and hot summer. Porto is located at the Northwest region of Portugal. which is one of the wettest regions in continental Europe. 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic

International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety Statistical description Meteorological data sources: Variable Mean Standard Deviation Minimum Maximum Daily Precipitation (mm) - meteorological station located at Serra Number of crashes without 3.3 2.4 0 25 victims per day do Pilar, near the city center, and belonging to the Geophysics Institute of the University of Porto. Number of crashes with victims 0.7 0.9 0 6 per day Daily mean temperature(ºC) - meteorological station of the Daily precipitation (mm) 3.4 8.5 0.0 80.9 Portuguese Institute for Sea and Atmosphere (IPMA) located at Daily mean temperature (ºC) 14.9 4.4 3.6 30.4 Porto Airport, in the city surroundings. Lagged effects Daylight duration (min) 737.7 121.2 553.0 908.0 Daylight duration (min) - Lisbon Astronomical Observatory (OAL) Weekly precipitation (mm) 23.4 34.7 0.0 279.0 open database for the city of Porto Monthly precipitation (mm) 97.5 98.3 0.0 567.2 Weekly precipitation measures the impact of the accumulated precipitation during the previous 7 days. Monthly precipitation measures the impact of the accumulated precipitation during the previous 30 days. 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety Road hierarchy classification: Arterial Principal distributor Local distributor Local access urban road diversity in terms of geometric and functional characteristics proxy of the risk exposition 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety 4. Results Crashes with victims - Poisson regression model 3. Method Estimated Variable Standard Error P[Z>z] parameter Constant -0.9169 0.0978 0.0000 Count model – GLM Poisson vs. Negative Binomial Arterial (reference) 0 - - Principal distributor 0.3636 0.0557 0.0000 � � ~��������� � �, �ℎ��� � � = � �� � +ln ����� � Local distributor 0.7640 0.0519 0.0000 Local access 0.1011 0.0591 0.0871* Daily precipitation 0.0061 0.0021 0.0040 If reject the hypothesis of mean equal to variance, then: Weekly precipitation -0.0008 0.0007 0.2521** Monthly precipitation -0.0008 0.0003 0.0042 � � = � �� � +ln ����� � ~������1 � , 1 �� Temperature -0.0121 0.0044 0.0056 Time trend (ln γ ) 0.0039 0.0135 0.7742** Number of observations = 6,912; Overdispersion parameter tests: g= μ i: 2.204; g= μ i2: 2.408 Value inferior to χ 2 0.05GL=1 (3.84) thus, the hypothesis of mean equal to variance is not reject. Log likelihood = -5966,9; McFadden Pseudo R2 = 0,026; Akaike Information Criterion = 1.729 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic

International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety Crashes without victims - NB regression model 4. Results Estimated Variable Standard Error P[Z>z] parameter � Increasing daily precipitation – crashes with and without victims increase too Constant 1.0925 0.0415 0.0000 10% daily precipitation - 0.2% c. with victims 0.3% c. without victims Arterial 0 - - Principal distributor 0.7175 0.0224 0.0000 � Increasing monthly precipitation – crashes with and without victims decrease Local distributor 0.2078 0.0237 0.0000 10% monthly precipitation - 0.7% c. with victim 0.2% c. without victims Local access 0.7192 0.0216 0.0000 Daily precipitation 0.0097 0.0009 0.0000 � Increasing weekly precipitation – crashes without victims increase too Weekly precipitation 0.0008 0.0003 0.0042 10% weekly precipitation - 0.2% c. without victims Monthly precipitation -0.0002 0.0001 0.0284 Temperature -0.0153 0.0019 0.0000 � Increasing temperature – crashes without victims decrease Time trend (ln γ ) -0.0586 0.0058 0.0000 10% temperature - 2% c. without victims Number of observations = 6,912; Overdispersion parameter tests: g= μ i: 9.504; g= μ i2: 8.855 Value superior to χ 2 0.05GL=1 (3.84) thus, the hypothesis of mean equal to variance is reject. Log likelihood = -14.159.24; McFadden Pseudo R2 = 0.009; Akaike Information Criterion = 4.099 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety Crashes with victims Crashes without victims 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety 5. Conclusions: 5. Conclusions: � Monthly precipitation affects all crashes in the same line. � Daily precipitation affects all crashes in the same line. � Monthly precipitation may be associated to road users’ behavior rather to any poor road condition - during the rainfall season, drivers tend to be more � Daily precipitation may be associated to poor conditions (visibility, vision carefully. and friction conditions). � Weekly precipitation affects only crashes without victims – may induce in � Lagged effects are relevant to consider in the analysis of weather effects. drivers a feeling of adaptation to adverse weather conditions, leading to an underestimation of the risk. � Temperature affects only crashes without victims – positively associated to road conditions 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic

tend to be more carefulcautious International Co-operation on Theories and Concepts in Traffic Safety International Co-operation on Theories and Concepts in Traffic Safety 5. Further developments Thank you for your attention! � Studies by type of crash, particularly focusing on pedestrians. � National study, preferably by regions. � Comparison between regions and countries (transferability). � Add exposition variables Sara Ferreira: sara@fe.up.pt 30 th ICTCT Workshop 30 th ICTCT Workshop 26-27 October 2017 – Olomouc. Czech Republic 26-27 October 2017 – Olomouc. Czech Republic