The present document describes prediction models for accident, fatalities, serious injuries and killed and seriously injured victims, adjusted to bidirectional data of Portuguese motorway sections. The study was developed at Laboratório Nacional de Engenharia Civil (LNEC) in the scope of Workpackage 2 - Safety Impact Assessment and Accident Prediction Model of the RIPCORD-ISEREST project, carried out under the 6th European Framework Program. It covers the issues related to modeling accidents using the Negative Binomial model regressions as well as detailed diagnostic checks of the models obtained.
The data used were collected over a five year period, ranging from 1999 to 2003. Several explanatory variables were measured concerning exposure, number of lanes, presence of an additional lane, lane widths, type and widths of the road’s shoulders and medians; response variables include the number of accidents, killed and serious injuries, fatalities and serious injuries that occurred on the five year period. The data set was further divided into four subsets corresponding to all road sections whose variables with missing values were removed, road sections without missing values and the equivalent for road sections with values of the annual average daily traffic greater than 5000 vehicles. Models were fitted to the four data sets with the response variables consisting on the number of accidents, fatalities, serious injuries and killed and seriously injured victims. Statistical techniques of model selection and model checking, including deviance and likelihood ratio tests and the AIC procedure, as well as graphical methods, were intensively used as measures of goodness-of-fit.
1 | INTRODUCTION
2 | GENERALISED LINEAR MODELS
2.1 MODEL SELECTION
2.2 MODEL CHECKING
2.3 OVERDISPERSION
2.4 THE NEGATIVE BINOMIAL FAMILY
3 | METHODOLOGICAL APPROACH
3.1 THE MOTORWAY DATA SET
4 | MODELING THE DATA EXCLUDING THE VARIABLES WITH MISSING VALUES
4.1 MODELING THE NUMBER OF ACCIDENTS
4.1.1 Diagnostic Checks
4.1.2 The predicted number of accidents
4.2 MODELING THE NUMBER OF KSI (KILLED AND SERIOUSLY INJURED VICTIMS)
4.2.1 Diagnostic Checks
4.2.2 The predicted number of killed and seriously injured victims
4.3 DISCUSSION
5 | MODELING THE DATA EXCLUDING THE MOTORWAY SECTIONS WITH MISSING VALUES
5.1 MODELING THE NUMBER OF ACCIDENTS
5.1.1 Diagnostic Checks
5.1.2 The predicted number of accidents
5.2 MODELING THE NUMBER OF KILLED AND SERIOUSLY INJURED VICTIMS
5.2.1 Diagnostic Checks
5.2.2 The predicted number of killed and seriously injured victims
5.3 DISCUSSION
6 | MODELING THE DATA WITH VALUES OF AADT GREATER THAN 5000 AND EXCLUDING THE VARIABLES WITH MISSING VALUES
6.1 MODELING THE NUMBER OF ACCIDENTS
6.1.1 Diagnostic Checks
6.1.2 The predicted number of accidents
6.2 MODELING THE NUMBER OF FATALITIES
6.2.1 Diagnostic Checks
6.2.2 The predicted number of fatalities
6.3 MODELING THE NUMBER OF SERIOUS INJURIES
6.3.1 Diagnostic Checks
6.3.2 The predicted number of serious injuries
6.4 MODELING THE NUMBER OF KILLED AND SERIOUSLY INJURED VICTIMS (KSI)
6.4.1 Diagnostic Checks
6.4.2 The predicted number of killed and seriously injured victims
6.5 DISCUSSION
7 | MODELING THE DATA WITH VALUES OF AADT GREATER THAN 5000 AND EXCLUDING THE MOTORWAY SECTIONS WITH MISSING VALUES
7.1 MODELING THE NUMBER OF ACCIDENTS
7.1.1 Diagnostic Checks
7.1.2 The predicted number of accidents
7.2 MODELING THE NUMBER OF FATALITIES
7.2.1 Diagnostic Checks
7.2.2 The predicted number of fatalities
7.3 MODELING THE NUMBER OF SERIOUS INJURIES
7.3.1 Diagnostic Checks
7.3.2 The predicted number of serious injuries
7.4 MODELING THE NUMBER OF KILLED AND SERIOUSLY INJURED VICTIMS (KSI)
7.4.1 Diagnostic Checks
7.4.2 The predicted number of killed and seriously injured victims
7.5 DISCUSSION
8 | CONCLUSIONS
9 | REFERENCES
S. Azeredo Lopes
João L. Cardoso