Modeling vehicle accidents and highway geometric design relationships

The statistical properties of four regression models—two conventional linear regression models and two Poisson regression models—are investigated in terms of their ability to model vehicle accidents and highway geometric design relationships. Potential limitations of these models pertaining to their underlying distributional assumptions, estimation procedures, functional form of accident rate, and sensitivity to short road sections, are identified. Important issues, such as the treatment of vehicle exposure and traffic conditions, and data uncertainties due to sampling and nonsampling errors, are also discussed. Roadway and truck accident data from the Highway Safety Information System (HSIS), a highway safety data base administered by the Federal Highway Administration (FHWA), have been employed to illustrate the use and the limitations of these models. It is demonstrated that the conventional linear regression models lack the distributional property to describe adequately random, discrete, nonnegative, and typically sporadic vehicle accident events on the road. As a result, these models are not appropriate to make probabilistic statements about vehicle accidents, and the test statistics derived from these models are questionable. The Poisson regression models, on the other hand, possess most of the desirable statistical properties in developing the relationships. However, if the vehicle accident data are found to be significantly overdispersed relative to its mean, then using the Poisson regression models may overstate or understate the likelihood of vehicle accidents on the road. More general probability distributions may have to be considered.     

On statistical inference in time series analysis of the evolution of road safety

Data collected for building a road safety observatory usually include observations made sequentially through time. Examples of such data, called time series data, include annual (or monthly) number of road traffic accidents, traffic fatalities or vehicle kilometers driven in a country, as well as the corresponding values of safety performance indicators (e.g., data on speeding, seat belt use, alcohol use, etc.). Some commonly used statistical techniques imply assumptions that are often violated by the special properties of time series data, namely serial dependency among disturbances associated with the observations. The first objective of this paper is to demonstrate the impact of such violations to the applicability of standard methods of statistical inference, which leads to an under or overestimation of the standard error and consequently may produce erroneous inferences. Moreover, having established the adverse consequences of ignoring serial dependency issues, the paper aims to describe rigorous statistical techniques used to overcome them. In particular, appropriate time series analysis techniques of varying complexity are employed to describe the development over time, relating the accident-occurrences to explanatory factors such as exposure measures or safety performance indicators, and forecasting the development into the near future. Traditional regression models (whether they are linear, generalized linear or nonlinear) are shown not to naturally capture the inherent dependencies in time series data. Dedicated time series analysis techniques, such as the ARMA-type and DRAG approaches are discussed next, followed by structural time series models, which are a subclass of state space methods. The paper concludes with general recommendations and practice guidelines for the use of time series models in road safety research.     

Accident prediction models for urban roads

This paper describes some of the main findings from two separate studies on accident prediction models for urban junctions and urban road links described in [Uheldsmodel for bygader-Del1: Modeller for 3-og 4-benede kryds. Notat 22, The Danish Road Directorate, 1995; Uheldsmodel for bygader- Del2: Modeller for straekninger. Notat 59, The Danish Road Directorate, 1998] (Greibe and Hemdorff, 1995, 1988).The main objective for the studies was to establish simple, practicable accident models that can predict the expected number of accidents at urban junctions and road links as accurately as possible. The models can be used to identify factors affecting road safety and in relation to 'black spot' identification and network safety analysis undertaken by local road authorities.The accident prediction models are based on data from 1036 junctions and 142 km road links in urban areas. Generalised linear modelling techniques were used to relate accident frequencies to explanatory variables.The estimated accident prediction models for road links were capable of describing more than 60% of the systematic variation ('percentage-explained' value) while the models for junctions had lower values. This indicates that modelling accidents for road links is less complicated than for junctions, probably due to a more uniform accident pattern and a simpler traffic flow exposure or due to lack of adequate explanatory variables for junctions.Explanatory variables describing road design and road geometry proved to be significant for road link models but less important in junction models. The most powerful variable for all models was motor vehicle traffic flow.     

A trial linkage of the road accident and vehicle registration files

Vehicle characteristics are of importance in all phases of road accident. None, however, are recorded by the police on the road accident registration form in The Netherlands. By linking the Vehicle Registration file to the Road Accident file a number of important vehicle characteristics can be added to the accident record. Theoretically it had to be possible to link both files because there was a common key variable viz. the vehicle registration number, which in The Netherlands is unique and stays with the vehicle its whole life. The test linkage was aimed at determining whether it was also practically possible. It was furthermore aimed at determining the completeness and reliability of both files. The fatal accidents of 1981 were selected to serve for the test because only fatal accidents are completely recorded, and 1981 was the most recent available year at the time. These limitations made it possible for SWOV to carry out a manual control on vehicle make and model, as this is often recorded by the police. Based on the conformity found, the validity of the linking by vehicle registration number was found to be 94% and the reliability at least 95%. The conclusion is therefore that the linkage can be accomplished in practice and that it reliably adds useful vehicle characteristics to the police records on fatal road accidents. This complementary information is in the first place important for deeper analysis of the relationship between, on the one hand, accident and injury characteristics and, on the other hand, the vehicle characteristics.     

Disaggregate analysis of road accident severities

A comprehensive understanding of the relationship between road accident occurrence and severity of consequences permits the formulation of safety measures that are most cost-effective. A disaggregate model of road accident severity based on sequential logit models is presented. The sequential binary approach is able to account for the dependency between different levels of severity. Factors that affect the level of damage experienced by individuals involved in road accidents include: accident dynamics, seating position, vehicle condition, vehicle size, driver condition and driver action. Separate models are calibrated for three accident situations: single-vehicle accidents, two-vehicle accidents and multi-vehicle accidents. Ontario road accident police reports are used to calibrate and validate the models. The results of a simple application of the models to a safety protocol involving the effectiveness of passenger restraint devices is presented.     

The effects of congestion charging on road traffic casualties: A causal analysis using difference-in-difference estimation

This paper aims to identify the impacts of the London congestion charge on road casualties within the central London charging zone. It develops a full difference-in-difference (DID) model that is integrated with generalized linear models, such as Poisson and Negative Binomial regression models. Covariates are included in the model to adjust for factors that violate the parallel trend assumption, which is critical in the DID model. The lower Bayesian Information Criterion value suggests that the full difference-in-difference model performs well in evaluating the relationship between road accidents and the London congestion charge as well as other socio-economic factors. After adjusting for a time trend and regional effects, the results show that the introduction of the London congestion charge has a significant influence on the incidence of road casualties. The congestion charge reduces the total number of car accidents, but is associated with an increase in two wheeled vehicle accidents.     

Impact of traffic congestion on road accidents: A spatial analysis of the M25 motorway in England

Traffic congestion and road accidents are two external costs of transport and the reduction of their impacts is often one of the primary objectives for transport policy makers. The relationship between traffic congestion and road accidents however is not apparent and less studied. It is speculated that there may be an inverse relationship between traffic congestion and road accidents, and as such this poses a potential dilemma for transport policy makers. This study aims to explore the impact of traffic congestion on the frequency of road accidents using a spatial analysis approach, while controlling for other relevant factors that may affect road accidents. The M25 London orbital motorway, divided into 70 segments, was chosen to conduct this study and relevant data on road accidents, traffic and road characteristics were collected. A robust technique has been developed to map M25 accidents onto its segments. Since existing studies have often used a proxy to measure the level of congestion, this study has employed a precise congestion measurement. A series of Poisson based non-spatial (such as Poisson-lognormal and Poisson-gamma) and spatial (Poisson-lognormal with conditional autoregressive priors) models have been used to account for the effects of both heterogeneity and spatial correlation.The results suggest that traffic congestion has little or no impact on the frequency of road accidents on the M25 motorway. All other relevant factors have provided results consistent with existing studies.     

An approach to accidents modeling based on compounds road environments

The most common approach to study the influence of certain road features on accidents has been the consideration of uniform road segments characterized by a unique feature. However, when an accident is related to the road infrastructure, its cause is usually not a single characteristic but rather a complex combination of several characteristics. The main objective of this paper is to describe a methodology developed in order to consider the road as a complete environment by using compound road environments, overcoming the limitations inherented in considering only uniform road segments. The methodology consists of: dividing a sample of roads into segments; grouping them into quite homogeneous road environments using cluster analysis; and identifying the influence of skid resistance and texture depth on road accidents in each environment by using generalized linear models. The application of this methodology is demonstrated for eight roads. Based on real data from accidents and road characteristics, three compound road environments were established where the pavement surface properties significantly influence the occurrence of accidents. Results have showed clearly that road environments where braking maneuvers are more common or those with small radii of curvature and high speeds require higher skid resistance and texture depth as an important contribution to the accident prevention.     

Globalization in road safety: explaining the downward trend in road accident rates in a single country (Israel)

A theoretical model is proposed in which road safety in a single country depends upon parochial considerations, such as police enforcement, and upon global considerations, such as international road safety technology. We show that there is a non-spurious relationship between the downward trend in the rate of road accidents in Israel and the road accident rate abroad. We suggest that this reflects the international propagation of road safety technology as it is embodied in motor vehicles and road design, rather than parochial road safety policy. Recent developments in the econometric analysis of time series are used to estimate the model using data for Israel. We make no direct attempt to explain the downward trend in the rate of road accidents outside Israel.     

Prediction of road accidents: A Bayesian hierarchical approach

In this paper a novel methodology for the prediction of the occurrence of road accidents is presented. The methodology utilizes a combination of three statistical methods: (1) gamma-updating of the occurrence rates of injury accidents and injured road users, (2) hierarchical multivariate Poisson-lognormal regression analysis taking into account correlations amongst multiple dependent model response variables and effects of discrete accident count data e.g. over-dispersion, and (3) Bayesian inference algorithms, which are applied by means of data mining techniques supported by Bayesian Probabilistic Networks in order to represent non-linearity between risk indicating and model response variables, as well as different types of uncertainties which might be present in the development of the specific models.     

Predicting accident frequency at their severity levels and its application in site ranking using a two-stage mixed multivariate model

Accident prediction models (APMs) have been extensively used in site ranking with the objective of identifying accident hotspots. Previously this has been achieved by using a univariate count data or a multivariate count data model (e.g. multivariate Poisson-lognormal) for modelling the number of accidents at different severity levels simultaneously. This paper proposes an alternative method to estimate accident frequency at different severity levels, namely the two-stage mixed multivariate model which combines both accident frequency and severity models. The accident, traffic and road characteristics data from the M25 motorway and surrounding major roads in England have been collected to demonstrate the use of the two-stage model. A Bayesian spatial model and a mixed logit model have been employed at each stage for accident frequency and severity analysis respectively, and the results combined to produce estimation of the number of accidents at different severity levels. Based on the results from the two-stage model, the accident hotspots on the M25 and surround have been identified. The ranking result using the two-stage model has also been compared with other ranking methods, such as the naïve ranking method, multivariate Poisson-lognormal and fixed proportion method. Compared to the traditional frequency based analysis, the two-stage model has the advantage in that it utilises more detailed individual accident level data and is able to predict low frequency accidents (such as fatal accidents). Therefore, the two-stage mixed multivariate model is a promising tool in predicting accident frequency according to their severity levels and site ranking.     

Predicting accident frequency at their severity levels and its application in site ranking using a two-stage mixed multivariate model

Accident prediction models (APMs) have been extensively used in site ranking with the objective of identifying accident hotspots. Previously this has been achieved by using a univariate count data or a multivariate count data model (e.g. multivariate Poisson-lognormal) for modelling the number of accidents at different severity levels simultaneously. This paper proposes an alternative method to estimate accident frequency at different severity levels, namely the two-stage mixed multivariate model which combines both accident frequency and severity models. The accident, traffic and road characteristics data from the M25 motorway and surrounding major roads in England have been collected to demonstrate the use of the two-stage model. A Bayesian spatial model and a mixed logit model have been employed at each stage for accident frequency and severity analysis respectively, and the results combined to produce estimation of the number of accidents at different severity levels. Based on the results from the two-stage model, the accident hotspots on the M25 and surround have been identified. The ranking result using the two-stage model has also been compared with other ranking methods, such as the naïve ranking method, multivariate Poisson-lognormal and fixed proportion method. Compared to the traditional frequency based analysis, the two-stage model has the advantage in that it utilises more detailed individual accident level data and is able to predict low frequency accidents (such as fatal accidents). Therefore, the two-stage mixed multivariate model is a promising tool in predicting accident frequency according to their severity levels and site ranking.     

Time series count data models: an empirical application to traffic accidents

Count data are primarily categorised as cross-sectional, time series, and panel. Over the past decade, Poisson and Negative Binomial (NB) models have been used widely to analyse cross-sectional and time series count data, and random effect and fixed effect Poisson and NB models have been used to analyse panel count data. However, recent literature suggests that although the underlying distributional assumptions of these models are appropriate for cross-sectional count data, they are not capable of taking into account the effect of serial correlation often found in pure time series count data. Real-valued time series models, such as the autoregressive integrated moving average (ARIMA) model, introduced by Box and Jenkins have been used in many applications over the last few decades. However, when modelling non-negative integer-valued data such as traffic accidents at a junction over time, Box and Jenkins models may be inappropriate. This is mainly due to the normality assumption of errors in the ARIMA model. Over the last few years, a new class of time series models known as integer-valued autoregressive (INAR) Poisson models, has been studied by many authors. This class of models is particularly applicable to the analysis of time series count data as these models hold the properties of Poisson regression and able to deal with serial correlation, and therefore offers an alternative to the real-valued time series models. The primary objective of this paper is to introduce the class of INAR models for the time series analysis of traffic accidents in Great Britain. Different types of time series count data are considered: aggregated time series data where both the spatial and temporal units of observation are relatively large (e.g., Great Britain and years) and disaggregated time series data where both the spatial and temporal units are relatively small (e.g., congestion charging zone and months). The performance of the INAR models is compared with the class of Box and Jenkins real-valued models. The results suggest that the performance of these two classes of models is quite similar in terms of coefficient estimates and goodness of fit for the case of aggregated time series traffic accident data. This is because the mean of the counts is high in which case the normal approximations and the ARIMA model may be satisfactory. However, the performance of INAR Poisson models is found to be much better than that of the ARIMA model for the case of the disaggregated time series traffic accident data where the counts is relatively low. The paper ends with a discussion on the limitations of INAR models to deal with the seasonality and unobserved heterogeneity.     

Scenario analysis of freight vehicle accident risks in Taiwan

This study develops a quantitative risk model by utilizing Generalized Linear Interactive Model (GLIM) to analyze the major freight vehicle accidents in Taiwan. Eight scenarios are established by interacting three categorical variables of driver ages, vehicle types and road types, each of which contains two levels. The database that consists of 2043 major accidents occurring between 1994 and 1998 in Taiwan is utilized to fit and calibrate the model parameters. The empirical results indicate that accident rates of freight vehicles in Taiwan were high in the scenarios involving trucks and non-freeway systems, while; accident consequences were severe in the scenarios involving mature drivers or non-freeway systems. Empirical evidences also show that there is no significant relationship between accident rates and accident consequences. This is to stress that safety studies that describe risk merely as accident rates rather than the combination of accident rates and consequences by definition might lead to biased risk perceptions. Finally, the study recommends using number of vehicle as an alternative of traffic exposure in commercial vehicle risk analysis. The merits of this would be that it is simple and thus reliable; meanwhile, the resulted risk that is termed as fatalities per vehicle could provide clear and direct policy implications for insurance practices and safety regulations.     

An aggregate accident model based on pooled, regional time-series data.

The determinants of personal injury road accidents and their severity are studied by means of generalized Poisson regression models estimated on the basis of combined cross-section/time-series data. Monthly data have been assembled for 18 Norwegian counties (every county but one), covering the period from January 1974 until December 1986. A rather wide range of potential explanatory factors are taken into account, including road use (exposure), weather, daylight, traffic density, road investment and maintenance expenditure, accident reporting routines, vehicle inspection, law enforcement, seat belt usage, proportion of inexperienced drivers, and alcohol sales. Separate probability models are estimated for the number of personal injury accidents, fatal accidents, injury victims, death victims, car occupants injured, and bicyclists and pedestrians injured. The fraction of personal injury accidents that are fatal is interpreted as an average severity measure and studied by means of a binomial logit model.     

Causal models for road accident fatalities in Yemen.

An identification of the causes of road accident fatalities is becoming more important with the growth of technology, population, number of vehicles and the need for their use. Many authors have addressed the problem in the past but no universal findings have been obtained. The problem tends to be different under different environments and for different geographical regions. The aim of this paper is to develop a model for the analysis and forecasting of road accident fatalities in Yemen considering data restrictions. The proposed data has a particular structure of accident occurrence that has not been reported in any existing research using data in other countries. The available data for the period 1978-1995 is used to build models to understand the nature and extent of the causes of fatalities. Part of the data is used for model building and part of it for test purposes. The issues of correlation and causality have been addressed and multiple collinearity is investigated and dealt with. Two alternative models are proposed based on both statistical grounds and that of practicality in viable decision making. The influence of consuming a locally grown stimulant called Qat on road users has been addressed and it is found that it increases the risk of accidents. This is not the common understanding within the authorities in Yemen as growing and consuming Qat is unregulated.     

Profiling Casualty Severity Levels of Road Accident Using Weighted Majority Voting

To determine the individual circumstances that account for a road traffic accident, it is crucial to consider the unplanned connections amongst various factors related to a crash that results in high casualty levels. Analysis of the road accident data concentrated mainly on categorizing accidents into different types using individually built classification methods which limit the prediction accuracy and fitness of the model. In this article, we proposed a multi-model hybrid framework of the weighted majority voting (WMV) scheme with parallel structure, which is designed by integrating individually implemented multinomial logistic regression (MLR) and multilayer perceptron (MLP) classifiers using three different accident datasets i.e., IRTAD, NCDB, and FARS. The proposed WMV hybrid scheme overtook individual classifiers in terms of modern evaluation measures like ROC, RMSE, Kappa rate, classification accuracy, and performs better than state-of-the-art approaches for the prediction of casualty severity level. Moreover, the proposed WMV hybrid scheme adds up to accident severity analysis through knowledge representation by revealing the role of different accident-related factors which expand the risk of casualty in a road crash. Critical aspects related to casualty severity recognized by the proposed WMV hybrid approach can surely support the traffic enforcement agencies to develop better road safety plans and ultimately save lives.     

The effect of infrastructure and demographic change on traffic-related fatalities and crashes: a case study of Illinois county-level data

This paper presents analyses of data from the Highway Safety Information System (HSIS) for the State of Illinois. Our analyses focuses on whether various changes in road network infrastructure and geometric design can be associated with changes in road fatalities and reported accidents. We also evaluate models that control for demographic changes. County-level time-series data is used and fixed effect negative binomial models are estimated. Results cannot confirm the hypothesis that changes in road infrastructure and geometric design have been beneficial for safety. Increases in the number of lanes appears to be associated with both increased traffic-related accidents and fatalities. Increased lane widths appears to be associated with increased fatalities. Increases in outside shoulder width appear to be associated with a decrease in accidents. Inclusion of demographic results does not significantly change these results but does capture much of the residual time trend in the models. Potentially mis-leading results are found when the time trend is not included. In this case a negative association between vertical curvature and both accidents and fatalities. No statistical association with changes in safety is found for median widths, inside shoulder widths, and horizontal and vertical curvature.     

Relationships between road accidents and hourly traffic flow—I: Analyses and interpretation

This research extends the investigation of the relationships between measures of accidents and traffic flow, and considers the hourly flow instead of the average daily traffic (ADT), which has already been reported. The findings of this study serve as a basis for further clarification of the interactions between various levels of traffic flow and road accidents. Eight four-lane road sections were studied during an 8-year period, providing adequate data based on carefully predefined criteria. Power functions are fitted and classified according to: (1) time-sequence analysis for each roadway section; and (2) cross sectional analysis on a one year basis. The results are presented, separately for multi and single vehicle accidents, in a matrix-format. A linear dependency was observed between the power and the logarithm of the multiple constant. This was done in a similar fashion to the previously reported study of the relationships between road accidents and ADT. The results for each type of analysis and type of accident are discussed, and three examples of a practical application are given.