The main objective of the analysis work performed in the framework of the Work Package 4 of the DaCoTA project is to analyse the past evolution of the annual number of fatalities in the various member states, and to forecast this evolution up to 2020. The model applied for many countries was the Latent Risk Model, which defines the development of the annual numbers of fatalities as the result of the joint development of exposure and risk (see Bijleveld et al., 2008; Martensen & Dupont, 2010). Because it involves the simultaneous modelling of the risk and exposure trends, the LRT is a multivariate (bivariate) Time Series model. In this report, demonstration is made of the usefulness and appropriateness of multivariate time series models to further improve the analysis of the developments of annual fatality numbers by exploring models (1) integrating the fatality time series of a panel of countries, (2) handling the trends of fatality numbers for subgroups of road-users and for various accident types simultaneously, and (3) including economic variables such as GDP. The simultaneous modelling of multiple trends implies that attention is paid to possible correlations between the random variations of the trend components (level, slope among others). When these correlations are high, the components in question are said to be common for the different trends modelled. The identification of common components is important because it allows the improvement of models to make them more efficient, but also because common components are informative in themselves of the dynamics governing the evolution of the trends considered. In the next section, the notion of “correlation between the random variations of the components of different trends” is formally related to concepts that are fundamental in Time Series analysis, such as the concepts of stationarity, integration and co-integration. This is described on the basis of the results of the investigation of the correlations between fatality and exposure time series that has been conducted previously for the different member states. A first multivariate Time Series application is then presented and discussed, namely, the simultaneous analysis of the development of annual fatality series for the various member states — or subgroups of member states by means of macro panel data analysis. After having exposed the principles underlying the technique, an example application is presented for the development of fatality numbers in France and in the United Kingdom. The next section describes how the multivariate time series framework can be applied to fit a disaggregated model of the fatality trends for subgroups of road users (defined on the basis of age, gender, transport mode and others). This allows identifying subgroups of road users for which the evolution of fatality numbers is governed by common processes or components, and also subgroups for which this evolution appears problematic (or not as encouraging as that of others). An example is then provided: the model of the evolution of the number of fatalities for 6 different age groups in Spain, taking into account the evolution of the size of the population. A second application explores the relationship between the number of fatalities and GDP on a macropanel of 30 countries and shows how to articulate short-term and long-term variations between them in a coherent time series model. (Author/publisher)
Abstract