Indexes and Models to Analyse Road Safety Within the Aggregate Approach
M N Postorino, University of Reggio Calabria, IT
Road safety is an important aspect of urban and extra urban transportation systems, particularly due to the high social costs it involves. While different actions have been started for resolving the problem of the atmospheric pollution caused by the vehicles moving on the transportation networks as well as different efforts have been made to limit the environmental pollution at the end-of-life of the vehicles, in the safety field the situation is still very serious. The resources devoted to the road safety each year are largely smaller than the real needs; the situation is various from Country to Country, both for the differences in the regulations in force and for the different sensibility of users to the safety problem.
Many studies carried out in this field linked the risk of accidents, the percentage or the number of accidents, the number of fatal accidents (dependent variables) to different factors or explanatory variables (independent variables) such as: age, and/or sex of the driver, expert or inexpert drivers, speed, length of the network, use of the safety belts, meteorological conditions and so on. Other kinds of studies concern the aggregate description of the accidents occurred in a region, by using indexes for identifying the trend of some relevant variables (total number of accidents, number of fatal accidents, and so on, in a given location and in a given time period).
Studies already carried out attribute 10-11% of accidents to the non-observance of the safety distance, 15% to driver inattention, 2-3% to meteorological conditions and 52% to undefined causes, among which the user guide behaviour. Then, undefined factors are one of the most important causes of accidents, but the knowledge of the main causes of accidents (not due to human factors) is a crucial aspect for the analysts of the transportation systems if some intervention have to be made for reducing the number of accidents (and mainly the fatal accidents).
Road accidents can be considered the consequence of the interactions among the users of the urban or extra urban transportation system and the environment in which they move, or in other words, the consequence of the interaction among human factors, technological factors and environmental factors. An accident is then the result of a sequence of actions and events due to this interaction, whose strong complexity makes difficult to establish which factor could be the main cause of the accident and how little variations on the initial conditions could transform a slight accident in a fatal one.
Models for the analysis of accidents can be specified at different aggregation/disaggregation levels, also due to the availability of the relative data base.
Generally, the use of disaggregate models presupposes the realization of specific data base collections, in order to register all accident data, at a microscopic level, that occur in a prefixed area and in a time interval as large as to obtain a representative sample of the event. On the contrary, the construction of aggregate models does not require this kind of data collection, because of the macroscopic level of the analysis; in this case the data collected by police (or other institutional figures) allow a preliminary analysis to be carried out, and they can be used to obtain the probability that an accident will occur in a prefixed area as a function of the mobility characteristics, the transport system and the socio-economic characteristics of the area itself (see, for example, Broughton e Markey, 1996; Broughton, 1996; Aron et al., 1997; Broughton, 1997; Brouwer, 1997; Rodrigues et al., 1997; Ernvall, 1997; Persaud et al., 1997, Postorino e Sarnè, 2001).
On the other hand, the accident aggregate analysis is really important because it support the decisions and the actions addressed to assure a greater level of safety. Furthermore, it is addressed to evaluate the time-space characteristics of the road safety.
To summarize, the main aims of the aggregate analysis can be identified in the following:
* identification of the main factors related to the accidents;
* aggregate check of the attainment of specific objectives (as increase of the safety degree of the transportation system, or a part of it);
* analysis of the space distribution of different kinds of accidents in order to locate ?black points?;
* analysis of the accident time variation in order to establish the effectiveness of the actions taken to reduce the number and severity of accidents.
On the basis of the previous considerations, in this paper after an analysis of the main aggregate approaches used to study road accidents, a model has been specified and calibrated for a prefixed area, by using official sources of data. Furthermore, a comparison among safety indexes and calibrated models, by using the same data base, has been carried out.
In fact, the aggregate analysis can be carried out by using two methodologies:
* safety indexes
Indexes, given to their specific nature, can be used only to analyse the problem, while models can be used both to analyse and forecast.
Different indexes have been constructed, in order to establish the relationship between accidents and, respectively, human factors and environment.
Referring to models, different kinds of models have been proposed in literature, that can be grouped as follows:
* probabilistic models
* time series models
* Bayes models.
In this paper, following the first approach, the probability that an accident could verify at time t in location i, Xit, has been specified following a Poisson distribution and as a function of the Poisson parameter that represents the mean and the variance of the random variable 'number of accidents'. This parameter has been specified in order to take into account the relevant variables related to the accident occurrence. The variables, selected in the available data base, refer to the transportation system, both in terms of infrastructures and regulation.
Two kinds of models have been specified and calibrated, referring to roads and intersections. In fact, it has been considered that this separation can implicitly consider the specific behaviour of users in the two different conditions. The results in terms of goodness of fit of the specified models are very satisfactory and some interesting considerations have been carried out in terms of link between accidents and explanatory variables.
Similarly interesting results have been obtained in terms of indexes, particularly in terms of time-space variations.
Further developments concern the estimation of a behavioural probabilistic model to analyse and forecast the number of accidents given the transport network and the socio-economic characteristics.
AA. VV. (1998) ?DUMAS: Developing Urban Management and Safety ? Work package 4: Accident Investigation?. Progetto DUMAS.
Amis G. (1996) ?An application of generalised linear modelling to the analysis of traffic accidents?, Traffic Engineering + Control, December Aron M., Biecheler M.-B., Hakkert S., Peytavin J.-F. (1997) ?Headways, rear-end collisions and traffic: the case of French motorways?, 7th International Conference on Traffic Safety on Two Continents.
Ben-Akiva M., Ceder A., Cheng L.-H., Liss M. (1988) ?Estimation of intersection traffic safety improvements?, Final Report, MIT, Centre for Transportation Studies, Cambridge, Massachusetts.
Bergel R. (1997) ?Multivariate modelling of accident risk on the national road network?, 7th International Conference on Traffic Safety on Two Continents Broughton J., Markey, K.A. (1996) ?In-car equipment to help drivers avoid accidents?, TRL Project Report 198: Transport Research Laboratory. Crowthorne Broughton, J. (1996) ? A study of causation factors in car accidents?, Road Safety in Europe Conference, Birmingham, 9-11 September.
Brouwer M (1997) ?Road safety information system: Key information supporting traffic safety policy in the Netherlands?, 7th International Conference on Traffic Safety on Two Continents.
Ernvall T. (1997) ?Risks exposures and accident data?, 7th International Conference on Traffic Safety on Two Continents.
Hauer E. (1992) ?Empirical Bayes approach to the estimation of ?unsafety?: the multivariate regression method?, Accident Analysis and Prevention, vol.24 No5 Hauer E. (1996a) ?Identification of ?Sites with Promise??, Transportation Research Board Annual Meeting Preprint 960995 Hauer E.(1996b) ?Statistical test of the difference between expected accident frequencies?, Transportation Research Board 75th Annual Meeting Higle J.L., Witkowski J.M. (1988) ?Bayesian identification of hazardous locations?, Transportation Research Record, 1185, TRB Jarret D. (1997) ?Assessing the safety effect of treatment using data from a number of sites?, 7th International Conference on Traffic Safety on Two Continents Koornstra M.J. (1996) ?The quantifying of road safety developments?, Proceedings of the Conference Road Safety in Europe, Birmingham,U.K.
Persaud B., Cook W., Kazakov A. (1997) ?Demonstration of new approaches for identifying hazardous locations and prioritizing safety treatment?, 7th International Conference on Traffic Safety on Two Continents.
Postorino M.N., Sarnè G.M.L. (2001) ?Cluster analysis for road accident investigations?, Proceedings of Urban Transport 2001, WIT.
Postorino M.N. (2002) ?Sicurezza: analisi e modelli?, Rilievi, Modellizzazione e Controllo del traffico veicolare, Collana Trasporti, FrancoAngeli editore.
Rodrigues E., Picado-Santos L. (1997) ?A study on road accident in urban environment?, 7th International Conference on Traffic Safety on Two Continents.
Association for European Transport