Search

Treatment selection bias leads to an inaccurate estimation of treatment effects as applied to specific sites or problem locations. Treatment selection bias is a major source of inconsistency in the results obtained from conventional before and after and cross-sectional models. One of the major expressions of treatment selection bias concerns the use of collision occurrence data in justifying intervention. For example, in highway safety field, a treatment is often introduced at a given site based on its high collision experience. Under normal conditions we would expect these collision numbers to return to a lower long term expected value, regardless of intervention. For treated sites, conventional observational models ascribe this reduction in collisions to the given treatment. This results in an overestimation of treatment effect. In this paper, a propensity score model is introduced that deals explicitly with treatment selection bias. The model is applied to Canadian highway--railway grade crossings data to estimate reductions in collision subject to upgrades in warning devices. The results of the propensity score model are compared for similar types of treatments to a number of before and after and cross-sectional models for both U.S. and Canadian data. The propensity score method is shown to reduce treatment selection bias and has probable merit that need to be further examined. DOI: 10.1061/(ASCE)0733-947X(2007)133:2(112) CE Database subject headings: Safety; Railroad grade crossings; Evaluation; Statistics.

A number of previous studies have investigated driver behaviors including speed and lateral placement in curved sections of highways. However, few studies attempted to empirically address relationships considering other factors, such as driver's eye-fixation distributions and illuminations, which are also believed to be relevant to highway safety. This study consists of field experiments to examine driver behaviors under different geometric and illumination conditions in two-lane rural highways. In the experiment, drivers' eye movements, which have been considered as surrogate of drivers' visual attention, are investigated by using a head-mounted eye-movement tracking system. Regression models are developed to show the relationship among the data, and the results show that there are significant differences in driver's eye movements under different geometric and illumination conditions. These findings give traffic engineers practical suggestions when implementing traffic control devices to increase safety on curved sections of two-lane rural highways. CE Database subject headings: Driver behavior; Traffic speed; Traffic safety; Rural areas.