This report contains four main chapters, the first is a general introduction, the second an explanation of forecasting and time series, the third presents the available research data and the fourth results of the analyses. The first chapter focuses on traffic safety as a problem for our society (lives lost and the cost of traffic victims as a percentage of GNP). The policy objectives set in 2000 for 2010 and set in 2010 for 2020 are mentioned. The purpose of the research is indicated, this is predicting traffic fatalities for six countries by 2015 (Belgium, Czech Republic, Finland, Iceland, Poland and Sweden). This should be done aggregated as well as disaggregated by age, gender and transport mode. For Belgium an additional objective was formulated, this is combined disaggregated analysis of age and gender. It was not only the purpose to predict the fatalities, but also the fatalities per billion passenger kilometres, per 100.000 inhabitants or per 100.000 vehicles. However this could not be performed due to problems correcting the residual values. Further in chapter one, motivations for the choice of unsafety and exposure measures are given, the research questions are mentioned, a definition of risk is given, the chosen age, gender and vehicle classes are indicated, the main existing exposure classes are presented and a motivation for the choice of the six countries is given. Chapter two starts with “what is forecasting?” and the difference between univariate and multivariate forecasting. An introduction to time series is presented and the five patterns in time series, random, trend, seasonal, cyclical and autocorrelation are mentioned. Six different regression model techniques to analyse time series are indicated, those are, linear, non-linear, ARMA (auto- regressive moving average), ARIMA (Auto-Regressive Integrated Moving Average), DRAG (Demand for Road use Accidents and their Gravity) and state space models. In chapter two you can read a detailed presentation of the LRT (Latent Risk Time series) model that is used for the analyses of this research is given. Basically the model consists of six equations, two measurement and four state equations. There is one measurement equation for the fatality risk and one for the exposure component. Both measurement equations contain two state equations one for the level and one for the slope. Chapter two ends with some strengths of time series as a technique for future prediction. (Author/publisher)
Samenvatting