Accid. Ano/. & Prev. Vol. 24, No. 4. pp. 421-436. 1992 cnm1-4575/!?2 55.00 + .cm Printed in Great Britain. 0 1992 Pergamon Press Ltd. zyxwvutsrqp DRIVER BEHAVIOUR AT HORIZONTAL CURVES: RISK COMPENSATION AND THE MARGIN OF SAFETY YIIK DIEW WONG School of Civil and Structural Engineering, Nanyang Technological University, Singapore ALAN NICHOLSON Department of Civil Engineering, University of Canterbury, Christchurch, New Zealand (Received 2 April 1991) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQ Abstract-A study involving unobtrusive observation of drivers at horizontal curves before and after realignment is described. The speeds and path radii adopted by drivers in the curves before and after realignment are compared, as are the levels of side friction demanded by each driver while negotiating the curves before and after realignment. The results reveal substantial variations between drivers (with respect to speed, path radius, and side friction demand) and between the path and curve radii. While vehicle speeds increased markedly, the side friction demand was reduced for all curves except one. It is concluded that the margin of safety was increased for all curves, and this is supported by the accident data. INTRODUCTION Engineering work involving modification to the road environment, particularly the geo- metric layout, is frequently undertaken in an attempt to improve road safety. The jus- tification for such actions is the generally strong association between adverse geometric elements and accident blackspots, as indicated by several studies, including those by Boughton (1975), Jorgensen et al. (1978), and Federal Highway Administration (1982). For instance, small radius curves and narrow width sections of road have been shown to be overrepresented among accident blackspots. Previous studies appear to have followed two distinct approaches. The first involves obtaining inventories of geometric and other variables (e.g. traffic variables) for a large number of road segments in a large area, and relating accident data for those segments to those variables, generally via cross-tabulation and multiple linear regression. Such studies may be called “mass data studies,” as they involve the collection and analysis of masses of data, many of which may relate to sites with relatively few accidents. Such studies have two major weaknesses. First, the set of explanatory variables may not include all the factors that are really involved in accident occurrence, as there is considerable uncertainty over which factors are important and there are practical difficulties in col- lecting some data. Second, the explanatory variables are often not independent. Indeed, the road design process is aimed at ensuring consistency of geometric standards, so that there should be correlation between geometric variables (e.g. low radius curves are often associated with low sight distances and widths), making it difficult to identify accurately the effect of each variable separately. In addition, there is the old problem that regression analysis does not necessarily establish a causal relationship. These difficulties can be overcome in part by identifying accident blackspots and comparison sites (not having large numbers of accidents) and making a detailed survey of both sets of sites and their environments, in order to identify those factors that are present at the blackspots but not at the comparison (control) sites. An example of this approach is the study by Wright and Robertson (1976). Nevertheless, the true effect of each individual geometric variable is still not known precisely in quantitative terms. Studies of the effect of road geometry on accident occurrence do not generally include direct consideration of human factors. The results of in-depth accident investi- gations (Treat 1980; Sabey and Taylor 1980) show that human factors, alone or in combination with other factors, are involved in around 95% of accidents. The treatment 425