Davis, Gary A.

The goals of this project were to first review the state-of-art with regard to identifying highway sections where median barriers would be most effective in preventing median-crossing crashes (MCC), and if necessary, develop remedies for any identified deficiencies. A statistical technique was developed for estimating the frequency and rate of MCCs on each of a set of highway sections, which required the analyst to review only a subset of hard-copy accident reports. This technique was applied to Minnesota's freeways and rural expressways, and highway sections were ranked with respect to estimated frequency of MCCs. A first version of a simulation model was developed for comparing the cost-effectiveness of barrier projects on different highway sections. The model uses Monte Carlo simulation to estimate the probability that an encroaching vehicle crosses a median with a specific cross-section, and collides with another vehicle traveling in the opposite direction.

The primary objective of this project was to identify and evaluate parametric models for making default estimates of travel times on arterial links. A review of the literature revealed several candidate models, including the Bureau of Public Roads (BPR) function, Spiess's conical volume delay function, the Singapore model, the Skabardonis- Dowling model, and the Highway Capacity Manual's model. A license plate method was applied to a sample of 50 arterial links located in the Twin Cities seven county metropolitan area, to obtain measurements of average travel time. Also obtained were the lengths of each link, measurements of traffic volume, and signal timing information. Default values for model parameters were obtained from the Twin Cities planning model's database. Using network default parameters, we found that the BPR and conical volume-delay models produced mean average percent errors (MAPE) of about 25%, while the Singapore and Skabardonis-Dowling models, using maximal site-specific information, produced MAPE values of around 6.5%. As site-specific information was replaced by default information the performance of the latter two models deteriorated, but even under conditions of minimal information the models produced MAPE values of around 20%. A cross-validation study of the Skabardonis-Dowling model showed essentially similar performance when predicting travel times on links not used to estimate default parameter values.

This report describes an effort in estimating crash modification factors (CMFs) associated with different left-turn phasing schemes, at intersections where the major approach speed limit exceeds 40 mph. For installation of signals at previously thru/stop-controlled intersections, rear-end crashes increased while right-angle crashes decreased. Installation of the signal had no effect on either major or minor approach left turn crashes as long as the protected-only left turn phasing was used on the major approaches. At one intersection where a signal was originally installed with permitted/protected phasing on the major approaches, we found evidence for an increase in major approach left-turn crashes, which vanished when the major approach left-turn treatment was changed to protected-only. For several other phasing changes it was not possible to construct an after-treatment data set of sufficient size to permit reliable estimation of an effect. This report also describes a simple simulation model for left-turn cross-path crashes, where a probabilistic gap acceptance model for the turning driver is combined with a standard braking model for the opposing driver. The model characterizes left-turn crashes as resulting when the turning driver accepts a minimal gap and takes an atypically long time complete his/her turn, while the opposing driver takes an atypically long time to react before braking. R reconstruction of an actual fatal crash however was more consistent with the opposing driver reacting normally, but with the turning driver selecting an atypically short gap. Characterizing the rate at which such selection errors occur would then be necessary to accurately predict left-turn crash frequencies.

"What would it take to build our way out of congestion in the Twin Cities?" was the question posed by researchers five years ago. This previous study solved a roads-only network design problem (NDP) for the Twin Cities of Minnesota. Building on that work, another network design problem is examined for the Twin Cities metropolitan area of 3 million, to examine the tradeoff between demand side reductions and the limited access capacity expansion necessary to achieve desired levels of service. The problem is simplified by pre-determining a mode split, which allows for incorporating decreasing demand directly as an input rather than in the model formulation. The problem is solved using Sequential Linear Expansion (SLIE), a modified method of successive averages (MSA). Computation time for the large network is decreased to a reasonable length using another modification, the MSA with decreasing re-initialization (MSADR). A typical personal computer can solve this large-sized problem within 24 hours. For forecasted travel demand for 2030, it was found that if the number of trips were reduced by 20%, lanemiles needed to achieve LOS D decreases by up to 43%.

This is Report #2 in the Series: Developing Intersection Decision Support Solutions. This project was a component of the Intersection Decision Support (IDS) effort conducted at the University of Minnesota. In this project, statistical modeling was applied to crash data from 198 two-way, stop-controlled, intersections on Minnesota rural expressways, in order to: (1) identify intersections that were plausible candidates for future IDS deployment; (2) develop a method for estimating the crash-reduction effect of IDS deployment; (3) develop a method for predicting the crash-reduction potential of IDS deployment, and (4) test the hypothesis that older drivers were over-represented in intersection crashes along US Trunk Highway 52. All these objectives were accomplished using hierarchical model structures similar to that employed in the Interactive Highway Safety Design Model. Five rural expressway intersections were identified as having crash frequencies that were atypically high, and this group included the intersection of US Trunk Highway 52 and Goodhue County highway 9, the site chosen for the prototype IDS deployment. It was then determined that a 3-year count of crashes after deployment would probably be sufficient to detect any crash reduction effect due to the IDS, although a reliable estimate of the magnitude of this effect would require a longer test period. Assuming that the effect of an IDS deployment would be to make the crash frequencies at treated intersections similar to that experienced by typical intersections, it was estimated that deployment of the IDS at the five high-crash intersections would, over a 15-year period, result in a reduction of about 308 crashes. Finally, using an induced-exposure approach, twelve intersections were identified as showing over-representation of older drivers, five of these being on US Trunk Highway 52.

This project was divided into three phases. In phase 1 ten fatal run-off-road crashes were reconstructed from crash scene diagrams and investigation reports. We found evidence of excessive speed in five of these, and a failure to properly use seatbelts eight of the ten. For seven of these we found that barriers complying with Test Level 3 of NCHRP Report 350 would probably have stopped the crashing vehicle's encroachment. In phase 2 we developed a vehicle trajectory simulation model and used it reconstruct five fatal median-crossing crashes. We found clear evidence of excessive speed in one of these, and in three of the five the encroaching vehicle would probably have been restrained by Test Level 3-compliant barriers. In phase 3 five teams of traffic safety professionals reviewed accident reports from a sample of fatal rural crashes, with the aim of identifying possible causal factors and potential countermeasures. The most frequently identified causal factors were driver inexperience and failure to properly use restraints, while provision of rumble strips, improvements to roadsides or cross-slopes, and provision of guardrails or barriers were the most frequently-cited countermeasures.

This report presents an approach to assess the effect of vehicle traffic volumes and speeds on pedestrian safety. It shows that the probability of standardized pedestrian conflict resulting in a collision can be computed given data on the distribution of vehicle speeds and headways on a residential street. Researchers applied this method to data collected on a sample of 25 residential streets in the Twin Cities and found that collision rates varied between 4 and 64 collisions per 1,000 pedestrian conflicts, depending primarily on the street's traffic volume. Using a model that relates the impact speed of a vehicle to the severity of pedestrian injury, they computed the probabilities of a severe collision. Sensitive to both traffic volume and traffic speed, the severe collision rate varied between 1 and 25 collisions per 1,000 conflicts. Using the same data, researchers also computed the crash reduction factor, used to assess the potential safety effect of a 25 miles per hour speed limit on the sample of residential streets. The estimated crash reductions ranged between .2% and 45%, depending primarily on the degree to which the vehicle speeds currently exceeded 25 miles per hour. Researchers also showed how this computation assists with the reconstruction of actual vehicle/pedestrian collisions.

What would it take to build our way out of congestion in the Twin Cities? As part of this research project, researchers identified a method to answer that question and found a minimal set of highway capacity expansions that would accommodate future travel demand and guarantee mobility. The problem of identifying a set of capacity expansions that are in some sense optimal, while accounting for traveler reaction, is known as a network design problem. A literature review reveals numerous formulations and solution algorithms over the last three decades, but the problem of implementing these for large-scale networks has remained a challenge. This project presents a solution procedure that incorporates the capacity expansion as a modified step in the Method Successive Averages, providing an efficient algorithm capable of solving realistic problems of real-world complexity. Application of this method addresses the network design problem for the freeway system of the Twin Cities by providing a lower bound on the extent to which physical expansion of highway capacity can be used to accommodate future growth. The solution estimates that adding 1,844 lane-kilometers, or 1,146 lane-miles, would be needed to accommodate the demand predicted for the year 2020.

This project provides a statistically defensible estimate of bicycle-miles of travel (BMT) for at least a substantial portion of the Twin Cities region and assesses the feasibility of monitoring bicycle volumes using sampling methods similar to those used to monitor motor vehicle traffic. Researchers used an ArcView database of the Twin Cities street system for the initial sampling frame and extended the database by manually adding information about average annual daily traffic volumes and about on- and off-road bicycle facilities. A stratified random sample of roadways links in Hennepin, Ramsey, and Dakota counties was drawn, and during the months of May through June and August through October 1998, the daytime bicycle volume for one day at each sampled site was obtained using time-lapse video. Researchers then used Cochrane's combined estimator to compute an estimate of average daytime BMT for the study area. Findings show that monitoring bicycle miles of travel using methods similar to those employed for vehicle miles of travel is now technically feasible in the Twin Cities region where several permanent counters on bicycle trails provide a rudimentary continuous count element. A video-based approach appears to be more accurate and less demanding of personnel than is on-site manual counting.

This report describes a Bayesian method for estimating accident rates at individual sites, which takes into account the fact that the total traffic count usually used to measure exposure is generally not known with certainty. The first step involves deriving an approximation for the probability of distribution of total traffic conditioned on a short count sample. This approximation is then used to derive a Bayes estimator of a site's accident rate, conditioned on an accident count, a short count sample, and the total traffic approximation. The method then uses Gibbs sampling to compute accident rate estimates. Tests based on actual accident and traffic data revealed that accident rate estimates based on a two-week traffic sample area are almost as accurate as estimates based on full traffic counting, but that uncertainty in the estimated accident rates increase by 20 to 50% when using a two-day count sample.