The development of the Traffic Conflicts Technique: an approach to the study of road accidents

A practical and reliable alternative or supplement to injury accident data is necessary to diagnose dangerous sites and evaluate remedial measures because available accident data is scarce, is lacking in detail about the events preceding the accident and it takes a long time to accumulate statistically reliable data. The most favoured alternative is the Traffic Conflicts Technique which satisfies most of the requirements of a supplementary measure, but has so far only been successfully validated for rural dual carriageway intersections (Spicer, 1973). To establish the technique it is necessary a) to ensure that the subjective judgements on which it is based are reliable, b) to develop the best methods of recording conflicts, and of training and selecting observers, and then c) to test the validity of the best available technique. The main part of this thesis reports three studies aimed at each one of these issues. In the first study intra observer reliability tested on filmed material varied between rs = 0.30 and 0.91 (0.65 overall for N = 42), but poor observers could be identified. By selecting the best observers an overall reliability figure of up to 0.88 could be obtained. Reliable observers remained reliable or even improved slightly on the second testing. These reliable observers also showed good agreement with expert judges who had viewed the film many times, and by selection a correlation with the criterion value of up to 0.83 could be obtained. In the second study a new recording method was developed, incorporating factors that experienced observers used to differentiate the grades of severity currently in use. This helped observers by defining the criteria for detection and grading of a conflict more objectively. This increased the overall intra observer reliability from 0.73 to 0.80, and agreement with the criterion values from 0.66 to 0.76. Transfer from laboratory to field led to a drop in the numbers of conflicts reported. From these studies and a survey of the requirements of local authority accident investigation units, a manual and training package was developed giving guidance on training and selecting observers for the purpose of obtaining reliable conflict data, such as that required for validating the technique. In the third study this package was validated in a study of a sample of eight urban T-junctions. Again the best observers were selected and found to have an overall reliability of 0.88. It was found that, when rear end conflicts were excluded (on the grounds that they led to so few reported injury accidents while occurring in large numbers), there was a high correlation between accidents per vehicle and conflicts per vehicle (rs = 0.79, p