Ars: an Adaptive Accident Reporting System

The paper presents the guidelines and the features of ARS, a project focused on the application of advanced methods and technologies for improving road accident data collection. In the first part of the paper, the most common types of accident reporting errors are described. Then the costs and benefits associated with some potential strategies to improve the data collection are compared. The ARS approach is presented as a combination of linkage from multiple data sources and computer-mediated data collection. The approach should improve the quality of the information collected, increase its quantity, and allow the execution of multiple types of validation. In the second part of the paper the Accident Reporting System is described. The system relies on a decision support system component and on a cognitively grounded information design to provide a helpful data collection environment. The system design guidelines, the usability guidelines, and the AI models are briefly described. After a short technical description, the system evaluation procedures are finally outlined. THE CHALLENGES OF ROAD ACCIDENT DATA COLLECTION Monitoring road safety constitutes, due to the social and economical consequences of the accidents, one of the fundamental goals of any traffic management policy. The achievement of such a goal depends on the availability of reliable data collection procedures. Good quality data will in fact allow the extraction of knowledge useful to evaluate the consequences and the potential side effects of any decision aimed at improving safety practices. Unfortunately, the use of this knowledge is often hindered by lack of international standards, and by significant differences in the semantics of the data collected. Even within the European Union countries, for example, there are many discrepancies in the definitions of core terms used to describe road accidents (1). Another major problem is constituted by the influence of different types of errors made in the collection of accident data. The main sources of errors are the following (2): • Non coverage error. The sample data (collected or used) is not representative of the population of road accidents. The accidents have been selected using deficient sampling frames. • Sampling error. This is the error associated with the use of sampling procedures. It can be quantified, and its estimate could be formulated by using confidence intervals. • Missing item or unit. A single item is missing (missing item) or the entire accident is not reported (under report: 3, 4). This is a systematic error (bias) related to the crash severity: the more severe the crash, the less frequent the under reporting error. The bias can be corrected post-hoc by using additional data sources, but these data can be affected by errors, too. • Measurement error. The report is affected by systematic or random errors due to instrumental or measurement failures. It is necessary to adopt measurement devices and procedures with good reliability and validity. • Coding or recoding error. This error is mainly related to the data recoding procedures and it can be reduced if these procedures are simple and clearly stated (again: they must be valid and reliable). To rectify such a situation, several authorities have already designed and implemented plans aimed at improving safety data collection and management processes (5, 6). Moreover, the Archived Data User Service programme, that is included in the U.S. National ITS Architecture and whose standards are under development, regards safety-related data as a basic issue. There are many factors that could affect the accuracy of the data collection process, yielding missing items, measurement, and coding errors. Among them, the following are significant ones: • high workload affecting the reporter (data collection does not have the highest priority on the accident scene); • difficulty of judgement under uncertainty (some accident-related judgements require complex probabilistic inference and medical or technical expertise); • defective information design of the report forms. Some “good” examples of defective information design can be found in the ISTAT-1998 (i.e., the National Institute of Statistics) report form, still used in Italy (with some minor changes) for accident data collection. For instance: • The sequence of the items is inconsistent and forces the reporter to shift attention frequently between the crash elements (motor vehicles, road, persons, and global accident context). • The spatial chunking and placement of graphical and textual elements do not mirror the conceptual organization of the information (for example, space-related elements instead of being clustered together are interleaved with time-related ones). • The format of the required information and the order of recurrent options across different fields are inconsistent (for example, the same information about the pedestrians involved in the accident vs. the injured occupants of motor vehicles is requested by using different formats). • The currently reported unit is not always clearly stated (in some cases the form does not make explicit whether the reported unit is the single person or a group of people sharing the same features). STRATEGIES TO IMPROVE ROAD ACCIDENT REPORTING Several strategies could be adopted to reduce the occurrence and the consequences of accident reporting errors. Table 1 presents a comparison of three of the most fruitful strategies that could be used to improve both the quality (i.e., accuracy) and the quantity of the information through orthogonal changes in the data collection process. Each strategy is implicitly compared with the standard procedure currently followed by Italian Police officers (i.e., filling a paper form on the accident site, with further recoding and integration procedures). The table displays the expected gains (No gain, Low gain, Moderate gain, High gain) in the information quantity and quality, and presents the costs (from Low to High) and potential drawbacks associated with each procedure.