106 Transportation Research Record: Journal of the Transportation Research Board, No. 2316, Transportation Research Board of the National Academies, Washington, D.C., 2012, pp. 106–113. DOI: 10.3141/2316-12 A. Talebpour, Department of Civil Engineering, and H. S. Mahmassani, Transpor- tation Center, 215 Chambers Hall, Northwestern University, 600 Foster Street, Evanston, IL 60208. S. H. Hamdar, Department of Civil and Environmental Engi- neering, School of Engineering and Applied Science, George Washington Univer- sity, 20101 Academic Way, Room 201-I, Ashburn, VA 20147. Corresponding author: H. S. Mahmassani, masmah@northwestern.edu. It has been hypothesized that congestion level and congestion type are strongly related to accident type and accident severity. Accord- ingly, higher congestion levels would result in a higher accident rate, albeit with lower severity. However, accidents are rare events; thus, there is a lack of data that explicitly focus on this relationship. Alternatively, traffic sensor data have been used in combination with accident data from different sources to explore this relationship. Studies that deal with the effects of different flow regimes on accident types and severity can be categorized into two groups (5). The first group uses aggregated data over long periods (e.g., hours) to com- pare the crash rate and the congestion level at different locations and times; the second group uses aggregated data over short periods (e.g., 5 min) to analyze this relationship. Several studies in the first group used average hourly traffic vol- ume to investigate the relationship between congestion level and crash rate. The approach followed was largely statistical and gave limited consideration to the underlying behavior of the problem. Zhou and Sisiopiku presented an attempt to capture the relationship of the volume-over-capacity ratio and the accident rate on an urban motorway in Michigan and proposed a U-shaped relationship (6). Kononov et al. studied the relationship between the annual average daily traffic and the accident rate by means of neural networks (7 ). They used data from Colorado, California, and Texas freeways and proposed a sigmoid function to reflect this relationship. None of the studies provided specific quantitative information about the relation- ship between congestion level and crash rate. Furthermore, studies in this group present different and sometimes contradictory results about this relation. In the second group of studies, Oh et al. studied the relationship between the crash likelihood and the 5-min aggregated speed, flow, and occupancy on a motorway in California (8). Their results showed the importance of speed variance in crash occurrence. Hourdos et al. considered both crash and near-crash maneuvers in their work (9). They used a video camera and sensors along a 1-mi motorway section in Minneapolis to analyze the relationship between crash frequency and traffic pressure (which is calculated by multiplying density and speed variance). Their results indicated that large speed variance between lanes and high congestion increase the crash likelihood. In all of the preceding studies, loop detectors were used to deter- mine the prevailing congestion level from an observer point of view. To go beyond observed correlation between aggregate measures and to understand the mechanisms that may lead to observed safety out- comes, a behavioral perspective that recognizes risky maneuvers by drivers in different situational environments is necessary. Different perceptions of local prevailing congestion, as well as differences in risk-taking behavior, must be recognized. Heterogeneity across drivers in both their perception of the surrounding condition and Safety First Microsimulation Approach to Assessing Congestion Effects on Risk Experienced by Drivers A. Talebpour, Hani S. Mahmassani, and S. H. Hamdar Prevailing traffic conditions affect highway safety and the processes by which drivers perceive a stimulus, evaluate it, and execute a correspond- ing driving maneuver. Several efforts have been made to use microscopic traffic simulation for evaluating highway safety. However, these efforts faced serious challenges because previous acceleration and lane-changing models had been built in an accident-free environment with different layers of safety constraints. A new approach relies on a cognitive risk- based microscopic model to study the relationship between prevailing traffic conditions and the risk experienced by drivers in a traffic stream. The model can consider accidents endogenously through lane-changing logic and provide an indicator of relative roadway safety as experienced by drivers. Six scenarios are simulated. The results show the importance of lane changing to understanding accident and near-accident occurrence in simulation models. A risk value comparison reveals that work zone bottlenecks have a greater impact on drivers’ risk-taking tendencies than bottlenecks caused by uphill grades. In addition to the well-publicized economic efficiency and mobility costs of highway traffic congestion (1), motor vehicle crashes impose a significant societal cost. A recent study by the American Automobile Association estimated that for every dollar of congestion cost, a cor- responding $1.84 is incurred by society because of crashes (2). The problem is even more critical because not all accidents are reported in the NHTSA Fatality Analysis Reporting System and General Estimates System. Furthermore, actual crashes are only the more visible outcome of the safety levels and risks experienced by drivers, because not all high-risk driving maneuvers lead to accidents. Obtaining information about high-risk maneuvers, called near-crash maneuvers, is costly; probe vehicles and video cameras are some of the known methods for gathering these data. Generalizing in a statistical sense is difficult, because observations typically reflect only the equipped probe vehicles and not the entire traffic stream. Examples of use of probe vehicles to obtain near crash data are the 100-car naturalistic driving study funded by the U.S. Department of Transportation (3) and a study by Thomas et al. that evaluated the effectiveness of NHTSA’s high-visibility enforcement model in Washington State (4).