The treatment of flow in the area of traffic safety has a long tradition. The influence flow has on the number of accidents is, however, often considered so obvious that it tends to be trivial. In this thesis I can show that the relation between flow and accidents holds interesting information. In order to compare countermeasures, or conduct other types of traffic safety comparisons where the flow varies, it is vital to know the full shape of the relation between accidents and flow. Such relations are called Safety Performance Functions (SPF). If we focus on the situation for individual road users the equivalent Risk Performance Functions (RPF) may be used to increase the comparability. The traditional comparison of accident rates is equal to assuming the SPF is a straight line, and thus the RPF is a constant. The main aim of this thesis is to develop a transparent system for estimating SPFs and RPFs. One step towards a transparent and, thus, interpretable treatment of accident and flow data is the development of a system for aggregating approaches in order to create aggregates with equal, or at least manageable, precision. The aggregation is based on relevant flow and is made in such a way that all aggregates represent the same number of the relevant road users. In order to generate a non-parametric function, without built-in presumptions of the overall relationship between flow and accidents, moving averages line is used. A series of computer programs is developed in order to describe the accuracy of the resulting functions. Two computer intensive methods: simulation and bootstrap, are used. With both these methods, exact confidence intervals are produced. Confidence intervals are computed by a process of interpolation. The stability of and the power of the computation of confidence intervals is tested with the use of synthetic data sets produced by a random process. The method developed was applied to a data set consisting of accident records, conflict observations and traffic flow counts for different road user categories from 95 non-signalised intersections in the cities of Malmoe and Lund. The result on bicycle and pedestrian safety is illustrated by graphs: SPFs and RPs. The empirical application has confirmed that: the method of aggregating and averaging gives a function with good visual interpretability; the bootstrap method gives an accurate and useful description of the stability of the estimated non-parametric function; the relationship between conflicts and flow is complex but knowledge about such relations could improve traffic safety evaluations that include comparisons between groups of locations with different flow; analyses of the non-parametric functions may be the base for suggesting active flow manipulation as a traffic safety measure; information about the effect of flow on safety may be used to generate hypotheses about the processes underlying traffic safety problems. (Author/publisher)
Abstract