Transportation Research Record: Journal of the Transportation Research Board

A new multivariate approach is introduced for jointly modeling data on crash counts by severity on the basis of multivariate Poisson-lognormal models. Although the data on crash frequency by severity are multivariate in nature, they have often been analyzed by modeling each severity level separately, without taking into account correlations that exist among different severity levels. The new multivariate Poisson-lognormal regression approach can cope with both overdispersion and a fully general correlation structure in the data, as opposed to the recently suggested multivariate Poisson regression approach, which allows for neither overdispersion nor a general correlation structure in the data. The new method is applied to the multivariate crash counts obtained from intersections in California for 10 years. The results show promise toward the goal of obtaining more accurate estimates by accounting for correlations in the multivariate crash counts and overdispersion.

A road safety analysis for different age and gender categories of road users is introduced. In contrast to many previous studies, the time series of road crashes by age and gender category is considered. The objective is to analyze Belgian data on the yearly number of fatalities for age and gender groups by using a decomposition of the number of fatalities in terms of exposure and risk in a time series perspective. For each category, a state space time series model is developed for the risk, which is defined in this study as the number of fatalities per capita. It was found that road risk is changing over age groups according to a U-shaped curve and that men generally have a higher risk than women. Further, the risk is decreasing over time but not at the same rate for all age and gender groups. The highest yearly reduction in risk is found for the oldest and youngest road users. The models also are useful to assess the attainability of formulated road safety targets, which makes them useful policy instruments. Especially for young men, the reduction in road risk is not in line with Belgian and Flemish policy expectations.

The objective of this study is to improve the estimation of the dispersion parameter of the negative binomial distribution for modeling motor vehicle collisions. The negative binomial distribution is widely used to model count data such as traffic crash data, which often exhibit low sample mean values and small sample sizes. Under such situations, the most commonly used methods for estimating the dispersion parameter–the method of moment and the maximum likelihood estimate–may become inaccurate and unstable. A bootstrapped maximum likelihood method is proposed to improve the estimation of the dispersion parameter. The proposed method combines the technique of bootstrap resampling with the maximum likelihood estimation method to obtain better estimates of the dispersion parameter. The performance of the bootstrapped maximum likelihood estimate is compared with the method of moment and the maximum likelihood estimates through Monte Carlo simulations. To validate the simulation results, the methods are applied to observed data collected at four-leg unsignalized intersections in Toronto, Ontario, Canada. Overall, the results show that the proposed bootstrap maximum likelihood method produces smaller biases and more stable estimates. The improvements are more pronounced with small samples and low sample means.

Different modeling philosophies are explored to explain and to forecast daily traffic counts. The main objectives of this study are the analysis of the impact of holidays on daily traffic and the forecasting of future traffic counts. Data from single inductive-loop detectors, collected in 2003, 2004, and 2005, were used for the analysis. The different models investigated showed that the variation in daily traffic counts could be explained by weekly cycles. The Box-Tiao modeling approach was applied to quantify the effect of holidays on daily traffic. The results showed that traffic counts were significantly lower for holiday periods. When the different modeling techniques were compared with a large forecast horizon, Box-Tiao modeling clearly outperformed the other modeling strategies. Simultaneous modeling of both the underlying reasons of travel and the revealed traffic patterns certainly is a challenge for further research.

Observational before-and-after safety evaluations have been commonly used to determine the effectiveness of safety improvements applied to high-crash locations. Such evaluations may typically be affected by regression-to-mean (RTM) bias. This research compares the effectiveness of low-cost safety improvements applied to several high-crash intersections in the cities of Detroit and Grand Rapids, Michigan, and examines the RTM effects by using various safety evaluation methodologies. Some previous studies suggest that before-and-after studies are always biased. It is also claimed that the observed number of crashes cannot be used to determine the effectiveness of safety improvements, and the use of the empirical Bayes (EB) method is suggested. This research examines the RTM effects by using before-and-after studies, before-and-after studies with control sites, and two different variants of the EB method. The research reveals that the expected crash frequencies computed by various evaluation methods do not differ significantly when 3 to 5 years' worth of traffic crash data are used. The deviations of the expected crash frequencies with the before-and-after methods and the EB method are computed to compare the RTM effects for each additional year of traffic crash data used in the evaluation. This research reveals that before-and-after studies produce results similar to those of the EB method, and the RTM effect becomes insignificant when 3 or more years' worth of traffic crash data are used in the evaluation of high-crash locations.

The number of roundabout intersections is quickly increasing around the world, mainly because of their good safety performance. However, roundabout performance can be degraded if precautions are not taken during either the design or the operation phase. Safety review methodologies are well suited for roundabout analysis, but their effectiveness can be improved substantially if the review recommendations are supported by a quantitative safety assessment. A roundabout safety assessment procedure is presented. Validation of the safety assessment procedure was carried out by a comparison of the results with empirical Bayes (EB) safety estimates. From the procedure, a safety index (SI) for roundabout approaches and for the whole number of roundabouts is assessed. The SI was assessed for 55 approaches of 15 urban roundabouts in Italy. For approaches, the correlation between the SI values and the EB safety estimates is highly significant (t = 15.62, p-value < 0.001), with 83% of the variation in the estimated number of crashes explained by the SI value. For all roundabouts, the correlation between SI values and EB safety estimates remains highly significant (t = 14.27, p-value < 0.001), with 94% of the variation in the estimated number of crashes explained by the SI value. The SI procedure makes it possible to support review recommendations by a quantitative safety evaluation. The SI can be used to rank both roundabouts and roundabout approaches by priority according to potential crash reduction from the safety measures. Furthermore, the SI procedure allows detection of safety issues that contribute the most to safety problems.

A study evaluated the safety impacts associated with improved signal visibility at urban signalized intersections. The improvements included one or a combination of the following upgrades: signal lens size, new backboards, reflective tapes added to existing backboards, and additional signal heads. Intersection collision data based on insurance claim records from the Insurance Corporation of British Columbia were used in the study to evaluate the effectiveness of the signal visibility improvement. These automobile insurance claim data are current, comprehensive, and considered quite reliable for intersection locations. Traffic volume and collision data were collected for treatment and comparison groups. The treatment group included 139 intersections and the comparison group included 85 intersections. The data for the comparison group were used to account for history and maturation confounding factors. An empirical Bayes analysis was used to ensure that the evaluation results were reliable and to account for the regression-to-the-mean confounding factor. The analysis was undertaken for both severe (injury + fatal) and property-damage-only (PDO) collisions and also for daytime and nighttime collisions. The evaluation results indicate statistically significant reductions of 8.5%, 5.9%, 6.6%, and 7.3% for PDO, daytime, nighttime, and total collisions, respectively. Severe collisions showed a nonsignificant reduction of 2.6%.

Traffic safety analysis has often been undertaken with historical collision data. However, well-recognized availability and quality problems are associated with collision data. In addition, the use of collision records for safety analysis is reactive: a significant number of collisions has to be recorded before action is taken. Therefore, the observation of traffic conflicts has been advocated as a complementary approach in the analysis of traffic safety. However, incomplete conceptualization and the cost of training observers and collecting conflict data have been factors inhibiting extensive application of the traffic conflict technique. The goal of this research is to develop a method for automated analysis of road safety with video sensors to address the problem of dependency on the deteriorating collision data. The method automates the extraction of traffic conflicts from video sensor data. This method should address the main shortcomings of the traffic conflict technique. A comprehensive system is described for traffic conflict detection in video data. The system is composed of a feature-based vehicle tracking algorithm adapted for intersections and a traffic conflict detection method based on the clustering of vehicle trajectories. The clustering uses a K-means approach with hidden Markov models and a simple heuristic to find the number of clusters automatically. Traffic conflicts can then be detected by identifying and adapting pairs of models of conflicting trajectories. The technique is demonstrated on real-world video sequences of traffic conflicts.

The reactive use is described of 35 recently developed macrolevel collision prediction models (CPMs) to conduct a black spot study with data from 577 urban and rural neighborhoods across Greater Vancouver in British Columbia, Canada. The research objective was to investigate macrolevel CPM use in a traditional reactive safety application (macroreactive use): identification, diagnosis, and remedy of hazardous locations. The results suggested that macroreactive use has the potential to complement traditional road safety improvement programs. Several collision-prone zones were identified and ranked for diagnosis. Two zones were analyzed in detail and revealed several potential enhancements to conventional methods. If adopted for normal use by practitioners, macrolevel CPMs could facilitate improved decisions by community planners and engineers and ultimately could facilitate improved neighborhood road safety for residents and other road users.

Nine criteria are discussed: they can be used to assess whether the effects attributed to road safety treatments in studies evaluating the effects of such treatments are causal or merely reflect randomness, confounding, or noncausal statistical association. The criteria are to a major extent based on similar criteria developed in epidemiology to assess the causality of factors associated with diseases. The criteria are related to (a) the presence of a statistical association between a treatment and changes in road safety (effect), (b) the strength of the statistical association between treatment and effect, (c) the internal consistency of the statistical association between treatment and effect, (d) the clarity of the causal direction between treatment and effect, (e) the invariance of the statistical association between treatment and effect regarding confounding factors, (f) the identification and measurement of causal mechanisms generating the association between treatment and effect, (g) the possibility of giving a theoretical explanation of the association between treatment and effect, (h) the presence of a dose-response pattern in the association between treatment and effect, and (i) the specificity of the association between treatment and effect to the intended target group of the treatment. These criteria can be used to assess the quality of road safety evaluation studies. The more criteria a study satisfies, the better is its quality. Use of the criteria is illustrated by means of examples.

In the Netherlands, the concept of sustainably safe traffic is the leading vision in road safety policy and research. The main goal of a sustainably safe road transport system is to reduce the annual number of road accident casualties to a fraction of the current levels. Important requirements resulting from this vision are that trips follow safe roads as much as possible, trips be as short as possible, and the quickest and safest routes coincide. Modeling route choice will provide answers to the planning issues of sustainably safe traffic; however, the safety effects of these requirements constitute a totally different issue, which needs to be dealt with. The focus of this study is on the design of a method that enables the planner to determine the safety effects of existing route choice and the changes in route choice. A description of road safety can be made in various ways. When a microscopic model is used, conflicts between vehicles will be an integral part of the simulation. The outcome will be used to compare the types of conflicts in a given simulation with the types of conflicts that would be acceptable in a sustainably safe road environment, for example, conflicts with opposing vehicles should be minimized at high speed differentials. A so-called route diagram of each route can be checked according to a series of criteria, each representing requirements for a sustainably safe route choice. Each criterion of the route diagram contributes to the total safety level of a route by the number of demerit points scored by the criterion. The criteria are described and tested in a microsimulation of alternative routes in a synthetic road network.

Accident prediction models are often used to estimate the number of accidents on segments and at intersections in the road network. Most often the models are developed for a total number of crashes for the facility or for crashes by severity. However, the frequency and severity of crashes of different types can be expected to vary according to underlying phenomena that cause them. To account for this variation better, modeling of accidents at intersections on rural four-lane highways in California is described separately for four different collision types: opposite-direction crashes, same-direction crashes, intersecting-direction crashes, and single-vehicle crashes. The findings from this modeling are reported with a special focus on the differences in crash types by (a) severity distribution, (b) dependence on traffic flow, and (c) variables that best explain between-site variations in the occurrence of different crash types. Evident differences exist in severity as well as the relationship of flow between several of the crash types. Intersecting and opposite-direction crashes are more severe than same-direction crashes. Same- and opposite-direction crashes exhibit similar relationships with traffic flow, but there are differences compared with intersecting-direction crashes and single-vehicle crashes. In addition, the variables that turn out to be good predictor variables differ somewhat for each crash type.

This study is based on real-time models of risk assessment for rear-end and lane-change-related crashes on freeways. These models were developed in recent studies on the basis of historical crash data and corresponding traffic data collected through underground loop detectors on Interstate-4 in Orlando, Florida. In this study the potential of these models was explored for identification of crashes that were not part of the database used to develop the models. These crashes include single-vehicle crashes that might result from evasive actions taken by drivers to avoid vehicles in front (i.e., a rear-end crash) or in a neighboring lane (i.e., a lane-change-related crash). In addition, traffic data corresponding to rear-end and lane-change-related crashes reported on the short form (because they did not involve any injuries or loss of life) also were subjected to the models. The results indicate that the models satisfactorily identify single-vehicle crashes (other than rollovers) as well as the short-form crashes. The study also demonstrates virtual real-time application of these models over complete traffic data collected for a month. A careful analysis of the models' output on these data sets is used to make critical inferences about their expected performance in a real-time application scenario. The proposed real-time application framework for these models not only is expected to improve traffic safety but also could yield significant enhancement in traffic operation by eliminating some of the incident-related congestion.

The linear programming model that is presented optimizes the crash and delay costs of highway networks by selecting safety and operational improvements with implementation costs within a constrained budget. The costs that are minimized in this study are those of crashes plus the costs of delay times. This optimization model works with crash prediction models that estimate the expected number of crashes by using base models and accident modification factors (AMFs). In these models the base model predicts the expected number of crashes for a base highway or intersection. The AMFs modify the prediction on the basis of the specific characteristics of each highway segment or intersection. Several improvement alternatives for different highways and intersections within a network would mean many improvement combinations, each with a different cost for implementation, which would cause certain crashes and delay. The proposed model uses linear optimization to find the combination of improvements with the lowest total of crash and delay costs within an acceptable range of implementation costs. The optimization model is presented in conjunction with crash prediction models borrowed from the Interactive Highway Safety Design Model software. The optimization model could be modified to work with other types of crash prediction models. A test case study containing a network of three highways and 13 intersections is presented. The corresponding optimization problem is solved with the CPLEX software. The variations of crash and delay cost savings versus improvement costs are studied, and the results are discussed.

The innovative time series methodology that is presented was used to examine the relationship between the safety impact of longitudinal pavement markings and their retroreflectivity. The need for the methodology was determined through an examination of the pavement marking safety literature, the desire to build on previous retroreflectivity research, and an understanding of how retroreflectivity is a function of the age of pavement markings and markers. The time series methodology involves solving for multipliers that represent the change in the expected number of crashes as a function of pavement marking retroreflectivity while simultaneously solving for seasonal effect multipliers. The time series methodology is explained in detail; its usefulness has been demonstrated through the analysis of 8 years of pavement marking and marker installation data, traffic volumes, and crash data. Safety effect multipliers were solved for yellow and white pavement markings separately and in combination. The multipliers covered different road types and crash severities by using retroreflectivity models and California's data for more than 118,000 nonintersection, nondaylight (night, dawn, and dusk) recorded crashes. To apply the time series methodology, it was necessary to develop retroreflectivity models as a function of age, color, marking material type or marker type, climate region, and amount of snow removal. The results showed that retroreflectivity levels have no effect to a small effect on the safety performance of those roadways maintained to a reasonable level of pavement marking.

Comparisons using raw fatality rates (per vehicle miles traveled or per population, or both) are likely to lead to biased pictures of relative traffic safety in the states. This study attempts to provide better pictures for a state-by-state comparison by using random intercept models based on the state-level panel data from 2000 through 2004. The data include demographic economic characteristics and roadway characteristics of 46 U.S. states. Predicted fatality rates were computed with the estimated models and fixed factors tuned to the values for Virginia as a reference. New rankings were generated on the basis of predicted rates and compared with raw rankings based on the raw rates. About half of the top 10 safest states in the raw rankings dropped out of the top 10 in the new rankings, which took into account control factors. However, the uncertainty of the estimated random intercepts suggests that the new rankings can be varied considerably; this implies that the rankings should not be used as stable traffic safety yardsticks for comparison between states, although they are better than the raw rankings.

The methodological approach to safety evaluation of two-lane rural highway segments that is presented uses both analytical procedures referring to alignment design consistency models and safety inspection processes. A safety index (SI) that quantitatively measures the relative safety performance of a road segment is calculated from the procedure. The SI is formulated by combining three components of risk: the exposure of road users to road hazards, the probability of a vehicle's being involved in an accident, and the resulting consequences should an accident occur. This systematic and replicable procedure integrates two different, complementary approaches–one based on design consistency evaluations and the other on safety inspections–and makes it possible to address a wide variety of safety issues effectively. A further advantage of the procedure is its applicability on highways where crash data are either not available or unreliable. Validation of the procedure was carried out on a sample of roads by a comparison of the risk rank obtained by using the SI and accident history. The SI was assessed in 30 segments chosen from a sample of two-lane rural highways in Italy, and the actual accident situation was obtained with the empirical Bayes (EB) procedure. Spearman's rank correlation was used to determine the level of agreement between the rankings obtained with the two techniques. The results from the Spearman's rank-correlation analysis validate the SI, indicating that the ranking from the SI scores and the EB estimates agrees at the 99.9% level of significance with a correlation coefficient of 0.87.

Safety-conscious planning is a relatively new concept. It was developed in response to safety-related provisions of the Transportation Equity Act for the 21st Century (TEA-21) of 1998, which required explicit consideration of safety in the transportation planning process. The problem of using accident rates in transportation planning is revisited, and a case history of applying safety-conscious planning methods by the Colorado Department of Transportation is reviewed. A two-phase process that has been used to evaluate the safety impacts of multiple design alternatives is introduced. The evaluation process is based on the available safety performance functions calibrated specifically for urban freeways in concert with diagnostic investigations, pattern recognition analysis, and detailed accident diagramming. The critical importance of accident diagramming is discussed in reference to examining safety history at complex interchange locations.

Right-angle crashes are prone to be severe at signalized intersections. To understand right-angle crash occurrence better and eventually to develop efficient countermeasures, this study investigated the effect of intersection traffic volume, geometric design features, and traffic control and operational features on right-angle crash occurrence at four-leg signalized intersections. Data from a total of 197 such intersections were collected from the Central Florida area. Right-angle crashes were modeled at the intersection, roadway, and approach levels. For the models at roadway and approach levels, crashes were assigned to a certain level, and then the disaggregated crashes were related to the specific roadway or approach features. The generalized estimating equations, which can account for site correlation among repeated observations from the same intersections, were used for the disaggregated models. The cumulative residuals method was used to evaluate the functional forms of the traffic volume and the models' overall performance. For the significant factors identified, the logarithm of the product of the conflicting through volumes, the number of through lanes, and the late-night and early-morning flashing operations is found to affect right-angle crash occurrence consistently. Left-turn offset, angle of the intersection, speed limit on the intersecting roadways, and yellow and all-red intervals also are significant in some models. The variables' relative significance was identified.

Speed limiters, which control the maximum speeds of vehicles so equipped, have been proposed in recent years as speed management tools. A simulation-based evaluation is described of the impact of a wide installation of speed limiters on traffic flow characteristics and on safety. Microscopic traffic simulation is used; it models the driving behavior of individual vehicles in detail and so captures not only the impact of speed limiters on equipped vehicles but also the interaction between equipped and unequipped vehicles and the resulting impact on the latter. In the experiment, the impact is evaluated of preset speed limiters for two limiter-set speeds, 100 km/h and 120 km/h, at various desired speed distributions and congestion levels. The simulation results show that speed limiters can reduce average traffic speeds up to 10% and that the variability of traffic speeds may also be lowered. Consequently, a significant reduction in accident rates might be achieved.

Red light cameras can be used as an alternative tool to supplement police efforts in enforcement against red light running, which is a major contributing factor to vehicle collisions at signalized intersections. The decision to install red light cameras should ensure overall improvement in traffic safety at signalized intersections. The major problem is deciding where and when to install the cameras in order to achieve the safety benefit. Therefore, guidelines are necessary for identifying and priority-ranking those intersections that have promise as sites for potential safety improvement. The primary objective of this study is to develop a safety analysis tool for estimating the safety impact of the installation of red light cameras at signalized intersections. Moreover, this research study provides a tool for identifying and priority-ranking problem intersections with respect to red light running within the entire roadway network under the jurisdiction of a particular agency. The proposed approach uses the empirical Bayes method, collision prediction models, and collision modification factors to estimate the safety changes upon installation of a red light camera at a signalized intersection. The generalized estimating equations technique with the assumption of negative binomial error distribution was used for development of the collision prediction models.

Reported animal-vehicle crashes (AVCs) and deer carcass removal have been used to define the deer-vehicle collision problem, identify its locations of concern, and evaluate its countermeasures. However, it has been shown that AVC magnitudes can be dramatically different. This research quantifies and compares the magnitude and patterns of AVC and deer carcass removal data from Iowa. Differences in these characteristics can affect the results produced by these activities. The difference in the magnitude of these two databases is confirmed, and some of the factors that may affect the size of this difference are discussed. Visual and quantitative comparisons are completed by using summary measures, geographic information system plots, and generalized linear regression models with a negative binomial error structure. This modeling approach has not been applied in the past to AVC or deer carcass removal data. AVC and deer carcass removal prediction (not causal) models for rural two-lane and multilane roadways were developed. The similarities and differences in the AVC and deer carcass removal models are discussed and the implications of these differences described. The differences found make the choice of database used critical to AVC-related roadway development decisions and policies, countermeasure location identification, and interpretation of research results. The recommendations provided focus on how AVC or carcass removal databases, as they typically exist, might be used and what improvements might be made for a more well-defined collection and application of these data.

Calibrating a safety performance function (SPF) with many years of accident data creates a temporal correlation that traditional model calibration procedures cannot deal with. It is well known that generalized estimating equations (GEE) models are able to incorporate trends into accident data and thus overcome difficulties in accounting for correlation; the usual application of GEEs to safety analysis uses robust (or sandwich) estimates of regression coefficients under the independence hypothesis for the working correlation matrix. This practice is justified by the robustness of the GEE procedure against misspecification of the response correlation structure. Nevertheless, with this method, one has to renounce the entirety of the advantages of GEE estimates, and–especially when correlation within the subject is high–significant losses in efficiency and misleading conclusions in model interpretation can occur. In such a case, losses in efficiency of the estimates will be transferred to the reliability of the final safety estimation, for example, by the empirical Bayes method. On the basis of these considerations, the main idea of this study is that, in safety modeling, additional effort to obtain the true data correlation structure will result in better precision in the estimation of SPF parameters. An example to illustrate the methodological aspects of the proposed approach is included.

Approximately 40% of all 2004 fatal crashes were single-vehicle, run-off-the-road (SVROR) crashes. The problem was even more significant on two-lane rural roads, where shoulder rumble strips were an important treatment in the prevention of these crashes and have proved effective on freeways. However, no studies have evaluated the effectiveness of shoulder rumble strips in reducing SVROR crashes on two-lane rural highways. An empirical Bayes before-and-after study was performed to evaluate the safety-effectiveness of milled-in continuous shoulder rumble strips on two-lane rural highways in Minnesota. The treatment was installed at 23 treatment sites (183 mi). This study reports a 13% reduction in all SVROR crashes and an 18% reduction in injury-producing SVROR crashes after installation of this treatment.

Modern road safety management programs are required to meet a number of objectives within the overall goal of reducing accidents on the road network. One of these objectives is the identification of hazardous locations through network screening. Often screening is focused on specific accident types to be mitigated by specific programs, for example, red light cameras for reducing right-angle accidents at signalized intersections. Recent methods developed for network screening utilize safety performance functions and the empirical Bayes methodology to overcome limitations in existing methods such as ignoring regression to the mean in using accident counts or rates as a ranking measure. These newer methods often are referred to as screening for the potential for safety improvement (PSI). In one version of the PSI method, sites are ranked according to their expected accident frequency; in another, sites are ranked by the expected excess frequency measured as the difference between the expected accident frequency and that expected at similar sites. This study focuses on a relatively untested method that screens for high proportions of specific accident types by using estimates of the probability that a site's observed proportion of an accident type is truly above a given critical proportion. The impacts on the method of selection for the critical proportion are examined and the efficiency of this method is evaluated by comparing its application with that of the two PSI methods for a data set of stop-controlled intersections. Results indicate that screening for high proportions may be a reasonable alternative where PSI-type approaches are not possible.

Roundabouts can provide substantial safety and traffic flow benefits compared with traffic signals and stop signs and as a result are increasingly used in place of traditional intersections. However, construction of roundabouts can be hampered by the negative perceptions held by some drivers. Prior research has found that public support increases soon after roundabouts are built and drivers become familiar with them. The purpose of the current study was to measure longer-term changes in public opinion in six communities where stop signs or traffic signals were replaced with roundabouts. Telephone surveys were conducted approximately 6 weeks before, 6 weeks after, and 1 to 5 years after construction of the roundabouts. The proportion of drivers in favor of roundabouts ranged from 22% to 44% before construction compared with 48% to 67% soon after roundabouts were built and 57% to 87% after roundabouts were in place for at least 1 year. The majority of drivers of all ages favored roundabouts after they were in place for 1 year or more, although support was higher among younger drivers (ages 18 to 34) and lower among older drivers (65 and older). There were small but nonsignificant differences between the opinions of male and female drivers. Drivers who said the roundabouts improved safety or traffic flow, or both, had more favorable opinions of roundabouts 1 to 5 years after construction. Results indicate that public support continued to increase with time, presumably because drivers became more familiar and comfortable with this form of traffic control.

Urban transportation planning has traditionally focused on capacity and congestion issues with some attention paid to operation and management and with the treatment of such issues typically made proactively. In contrast, road safety has received little attention in the planning process. Safety-conscious planning is a new proactive approach that incorporates safety issues into the transportation planning process. This approach requires a safety planning decision-support tool to facilitate a proactive approach to the assessment of safety implications of alternative network planning initiatives and scenarios. The objective of this research study is to develop a series of zonal-level collision prediction models that are consistent with conventional models commonly used for urban transportation planning. A generalized linear regression modeling approach with the assumption of a negative binomial error structure was employed for exploring relationships between collision frequency in a planning zone and some explanatory variables such as traffic intensity, socioeconomic and demographic factors, land use, and traffic demand measures. Planning-level safety models developed in this study with data for the city of Toronto, Canada, are presented with illustrative applications of how they can be used as decision-support tools for planners to explicitly consider safety in the transportation planning process. Macrolevel collision modification factors are presented to illustrate how the models can be used to examine the impact of each individual planning variable on the safety of an urban zone.

The use of observational data to estimate safety effectiveness for changes in shoulder width is explored. Common statistical models are applied to estimate safety effectiveness of shoulder width; however, two unique approaches, the case control and the cohort methods, are used in the design stage. Case control and cohort designs are common methods in epidemiological studies; they are used to develop relationships between risk factors and disease. In highway safety, case control and cohort designs show potential for relating geometric and traffic characteristics to crashes. Case control and cohort methods produce estimates of the odds ratio and relative risk, respectively, which indicate the expected percentage change in the probability of an outcome, given a unit change in the risk factor. In the highway safety application, the odds ratio and relative risk may provide a reasonable estimate of the crash modification factor (CMF) for the risk factor in question. Geometric, traffic, and crash data were obtained for two-lane, rural, undivided highway segments in Pennsylvania. Conditional logistic regression models were developed to estimate the odds ratio for shoulder width, and Cox proportional hazard models were developed to estimate relative risk. Confounding variables were identified in the preliminary analysis and adjustments were made for confounders during the design and final analysis stages. Estimates from the two methods indicate that crashes decrease as shoulder width increases. Results are compared with the CMF presented in the Highway Safety Manual, and confidence intervals are included to illustrate the certainty of the estimates.

Knowledge about characteristics of crash-involved drivers and their driving performance can enhance countermeasure programs designed to improve safety on roadways. However, few studies have been able to analyze detailed exposure data of individual drivers (travel by time of day and by roadway type) alongside driving performance data (speed, throttle, braking, and acceleration). Thus, researchers have been forced to base safety analyses on aggregate crash experience rather than on more frequent individual driving activities and driving performance parameters. This research evaluated driving exposure and performance differences between drivers who were involved in crashes versus drivers who were not involved in crashes during a 14-month study period, on the basis of data for vehicles with Global Positioning Systems. The objective was to verify whether the exposure of crash-involved drivers and driving performance were significantly different on the basis of disaggregated analyses by facility types and trip start times. Researchers found that the mileage exposure, speed, and acceleration patterns of drivers who were involved in crashes during the study period were significantly different from those of drivers who were not involved in crashes. Further, crash-involved drivers accumulated more mileage, consistently traveled at higher speeds, and engaged more frequently in hard deceleration events than their no-crash counterparts did. Although this study does not definitively correlate driver exposure and performance measures with crash causation, it does suggest that there is real potential to identify at-risk drivers on the basis of data obtained from vehicle-monitoring technologies. The identification and immediate correction of at-risk behaviors could obviate crash experience. If this research is successful, the surrogate measures also would reduce safety analysts' dependency on crash data.

One of the key hurdles in identifying unsafe intersections and roadways in Wisconsin is the lack of a complete crash location map, especially for crashes that occurred on local streets. Crash locations are reported in terms of relative offset from an intersection on the basis of on- and at-street name information, which identifies the intersection, and direction and distance information, which identifies the offset. For intersection crashes, the offset distance is typically set to zero. As described in this paper, the Traffic Operations and Safety Laboratory at the University of Wisconsin, Madison, has developed a system to automate the mapping of Wisconsin local road crash locations. The location mapping algorithm involves the integration of two separate Wisconsin Department of Transportation databases: the Wisconsin crash database of police traffic accident reports and the Wisconsin Information System of Local Roads (WISLR). The application of WISLR, which is an inventory of local roads with details such as traffic information, pavement condition, and roadway geometry, provides invaluable access to more comprehensive safety analysis. Although the methodology introduced is specific to these two databases, the general ideas can be applied to any similar sets of crash and geographic information system databases. The final result is a pinpoint map of all the intersection and segment crashes that occurred on local roads in Wisconsin, along with the complete crash information associated with each mapped crash. The algorithm developed with this methodology is able to map approximately 79% of the intended pool of available crashes. Quality evaluations indicate that the mapping is almost 98% accurate.