Roundabout installations are becoming common practice among DOTs and other local governments due to their superior safety attributes compared to other conventional at-grade intersections, especially stop-control and uncontrolled intersections. Current U.S. national guidelines for roundabout illumination recommend systematic illumination for all roundabouts. This recommendation might become a potential hindrance to desired widespread installations due to implied financial costs, especially in rural areas because the competing stop-control and uncontrolled intersections can be kept unlit. Interestingly rural roundabouts in most countries around the world are not illuminated as indicated by a recent survey of international roundabout illumination policies and standards from 45 countries. Also, review of intersection safety literature does not identify any publication that supports a systematic illumination policy of U.S. roundabouts. In fact, despite this recommendation there is no quantitative research on influence of illumination levels on nighttime safety at roundabouts and little on conventional intersections. Conversely, the literature shows a significant number of published studies which have indicated that currently recommended illumination levels on roadways can be reduced without compromising nighttime safety. This dissertation evaluates the link between roundabout crashes and different illumination levels. At the beginning of this dissertation research, there was no available repository of quantitative intersection illumination levels which could be used in highway safety research. Also, existing protocols for measurement require expensive light meters and are extremely time consuming to follow, making them impractical to use to study a large number of intersections. Consequently, this dissertation first evaluates the relationship with the best available data. The best available intersection illumination data was obtained from the Minnesota data contained in the Highway Safety Information System (HSIS). Minnesota crash and illumination data from 2003 to 2010 were analyzed. This illumination data was a qualitative description of intersection illuminating schemes and/or luminaire arrangement. Therefore, this dissertation also developed a cost-effective, accurate, and rapid method for measurement of quantitative intersection illumination data, and applied the developed protocol to a case study in Georgia. The measured intersection illumination was analyzed together with crash data obtained from GDOT for 2009 to 2014. The results of a naive analysis on the best available data indicated among other findings that the presence of lighting can provide approximately 61 percent lower total nighttime crash rate compared to the unlit condition. Also, providing illumination to the roundabout circle alone can yield about 80 percent of the benefits (55 percent reduction from unlit condition) of illuminating both the roundabout circle and approaches (66 percent reduction from unlit condition). Field test results for the camera calibration indicate that the average intersection illuminance derived from the protocol is within 4 percent difference of the actual average intersection illumination estimated from following the existing protocols. Next, despite limited roundabout data and potential issues of selection bias which could not be addressed in this dissertation, a cautious roundabout illumination specific crash modification factor was estimated with a negative binomial regression model. The model results showed that an increase of 1 lux in average roundabout illuminance will result in a 4.72 percent reduction in expected number nighttime crashes. The results of this work are useful in creating a sound framework for DOTs and other transportation agencies to determine the most appropriate level of illumination for roundabouts. This study also makes a number of significant contributions to highway safety research. First, this work is the first quantitative study on the impact of illumination on safety at roundabouts. The status-quo for highway safety research regarding the impacts of illumination had been to treat road lighting as a binary (Lit/Unlit) variable. However, even in most places without purposely-built road lighting there is usually ambient lighting from abutting facilities such as a gas stations or a store. Second, this dissertation is the first documented application of the photographic method to roundabouts. It is also the first documented application of the photographic method’s camera specific constant calibration approach to transportation field measurements. Previous documented application of the photographic method to transportation field measurements used an exposure specific calibration approach. Unlike the camera specific constant calibration approach, the exposure specific approach is rigid and field measurements must always be done at the exposure settings used in calibrating the camera. Thirdly, this work demonstrates the first developed procedure to developing uniformity (contour) plots from the photographic method. Next, this work can serve as the basis for initial efforts to create an illumination specific quantitative crash modification factor. Currently, the Highway Safety Manual is lacking in this important safety parameter. Last, but not the least this work offers procedures for collecting luminance data from the field and also documents a database of intersection illumination levels and intersection characteristics which can be used by future research.

Master of Science / Department of Civil Engineering / Sunanda Dissanayake / Construction of by-pass lanes at rural intersections typically has been considered a low-cost safety improvement. Safety analysis utilizes two common approaches to evaluate treatment effectiveness: before-and-after study and cross-sectional study. This research performed paired sample t-test statistical analysis to estimate changes in total of crash frequency, crash rates, EPDO crash frequency, and EPDO crash rates at intersections, three to five years after adding a by-pass lane compared to identical time period before the by-pass lane was added. Crash data between 1990 and 2011 were obtained from Kansas Crash and Analysis Record System (KCARS) maintained by the Kansas Department of Transportation (KDOT). In order to perform a cross-sectional study, intersections with by-pass lanes were compared to intersections with no countermeasures; crash data were obtained for more than 1,100 intersections in the state of Kansas. According to before-and-after study, addition of by-pass lanes improves safety at unsignalized rural intersections; crashes and their severities are reduced after adding by-pass lanes. But, these reductions are not statistically significant under 95% confidence level. However, when considering intersection related crashes, a statistically significant reduction in crash rates is happened after adding by-pass lanes at 3-legged intersections. In cross-sectional study, crashes and their severities are lower at 3-legged intersections with the by-pass lanes versus 3-legged intersections without the by-pass lanes. However, these reductions are not statistically significant under 95% confidence level. When considering 300 feet intersection crash box, statistically significant reductions are happened at 4-legged intersection. In contrast, crashes and their severities increased at 4-legged intersections with the by-pass lanes, but these changes are not statistically significant under 95% confidence level. The Crash Modification Factors were calculated to evaluate safety effectiveness of adding by-pass lanes at unsignalized rural intersections. The calculated CMFs less than 1.0, indicate a reduction in crashes after implementation of by-pass lanes. Finally, this study concluded that expected crashes at intersections with by-pass lanes are lower than intersections without by-pass lanes.

As a result of high crash frequencies on roadways, transportation safety has become a high priority for the United States Department of Transportation and the Utah Department of Transportation. A large percentage of fatal and injury crashes on roadways occur at intersections and traffic signals have been implemented to reduce these severe crashes. There is a need to evaluate the effectiveness of the traffic signal improvements through the development of Crash Modification Factors (CMFs). Recent research has shown that traditional safety evaluation methods have been inadequate in developing CMFs. In recent years, Bayesian statistical methods have been utilized in traffic safety studies to more accurately analyze the effectiveness of safety improvements. The hierarchical Bayesian method is an advanced statistical technique that has the capability to account for the shortcomings of traditional methods and to more fully reflect the effectiveness of safety improvements. This report uses a hierarchical Bayesian model to analyze the effectiveness of new traffic signal installations and modified traffic signals. CMFs were developed for multiple scenarios for both new and modified traffic signals. A benefit-to-cost (B/C) analysis was also performed for each improvement to determine how long it would take to recover the cost of installation. The results showed that there was an increase in overall crashes for both new signal installations and modifications to existing signals. The severe crash analysis revealed that there was an increase in non-severe crashes and a reduction in severe crashes; the improvements are effectively reducing severe crashes and improving safety at intersections. The B/C analyses indicate that there is a safety benefit to both improvements and that new signal installation costs can be recovered in approximately 5 years while the installation of a left-turn signal modification can be recovered in approximately 9 weeks.

Two way left turn lanes (TWLTL) and raised medians are common median treatments on roadways. This research focused on evaluating the safety effectiveness of conversion of TWLTLs into raised medians using Before-After and Cross Sectional Studies. In the Before-After Studies, we evaluated the effect of this treatment using the Naive, Before-After with Comparison Group (CG), and Before-After with Empirical Bayes (EB) Methods. In order to apply these methods, a total of 33 segments of a treated group and 109 segments of a comparison group have been collected. Also, safety performance functions (SPFs) have been developed using the negative binomial model in order to calibrate crash modification factors (CMF) using the Before-After with Empirical Bayes Method. This research also evaluated the safety effectiveness of this treatment on four and six lane roads using Before-After with CG and Before-After with EB. The type of raised medians was further evaluated using Before-After with CG and EB. In sum, the results from this study show that applying the before-After and Cross Sectional studies have proved that the conversion from a TWLTL to a raised median helped to reduce total, fatal and injury, head on, angle, and left turn crashes. It significantly reduces crashes for head-on and left turn crashes, by restricting turning maneuvers. Also, this study has proved that the treatment is more effective on four rather than six lane roads. Furthermore, two types of raised medians, concrete and lawn curb, were evaluated after the conversion from TWLTLs. It was found that both medians have similar effects due to the conversion, and both median types helped in reducing the number of crashes.

Pedestrian countdown signals (PCSs) are viable traffic control devices that assist pedestrians in crossing intersections safely. Despite the fact that PCSs are meant for pedestrians, they also have an impact on drivers’ behavior at intersections. This study focuses on the evaluation of the safety effectiveness of PCSs to drivers in the cities of Jacksonville and Gainesville, Florida. The study employs two Bayesian approaches, before-and-after empirical Bayes (EB) and full Bayes (FB) with a comparison group, to quantify the safety impacts of PCSs to drivers. Specifically, crash modification factors (CMFs), which are estimated using the aforementioned two methods, were used to evaluate the safety effects of PCSs to drivers. Apart from establishing CMFs, crash modification functions (CMFunctions) were also developed to observe the relationship between CMFs and traffic volume. The CMFs were established for distinctive categories of crashes based on crash type (rear-end and angle collisions) and severity level (total, fatal and injury (FI), and property damage only (PDO) collisions). The CMFs findings, using the EB approach indicated that installing PCSs result in a significant improvement of driver’s safety, at a 95% confidence interval (CI), by a 8.8% reduction in total crashes, a 8.0% reduction in rear-end crashes, and a 7.1% reduction in PDO crashes. In addition, FI crashes and angle crashes were observed to be reduced by 4.8%, whereas a 4.6% reduction in angle crashes was observed. In the case of the FB approach, PCSs were observed to be effective and significant, at a 95% Bayesian credible interval (BCI), for a total (Mean = 0.894, 95% BCI (0.828, 0.911)), PDO (Mean = 0.908, 95% BCI (0.838, 0.953)), and rear-end (Mean = 0.920, 95% BCI (0.842, 0.942)) crashes. The results of two crash categories such as FI (Mean = 0.957, 95% BCI (0.886, 1. 020)) and angle (Mean = 0.969, 95% BCI (0.931, 1.022)) crashes are less than one but are not significant at the 95 % BCI. Also, discussed in this study are the CMFunctions, showing the relationship between the developed CMFs and total entering traffic volume, obtained by combining the total traffic on the major and the minor approaches. In addition, the CMFunctions developed using the FB indicated the relationship between the estimated CMFs with the post-treatment year. The CMFunctions developed in this study clearly show that the treatment effectiveness varies considerably with post-treatment time and traffic volume. Moreover, using the FB methodology, the results suggest the treatment effectiveness increased over time in the post-treatment years for the crash categories with two important indicators of effectiveness, i.e., total and PDO, and rear-end crashes. Nevertheless, the treatment effectiveness on rear-end crashes is observed to decline with post-treatment time, although the base value is still less than one for all the three years. In summary, the results suggest the usefulness of PCSs for drivers.

Network screening and countermeasure selection are two crucial steps in the highway improvement process. In network screening, potential improvement locations are ranked and prioritized based on a specific method with a set of criteria. The most common practice by transportation agencies has been to use a simple scoring method, which, in general, weighs and scores each criterion and then ranks the locations based on their relative overall scoring. The method does not deal well with criteria that are qualitative in nature, nor does it account for the impacts of correlation among the criteria. The introduction of Analytic Hierarchy Process (AHP) provides agencies with a method to include both quantitative and qualitative criteria. However, it does not address the issue on correlation. This dissertation explores the use of both Analytic Network Process (ANP) and Fuzzy Analytic Network Process (FANP) for their potential capabilities to address both issues. Using urban four-lane divided highways in Florida for bicycle safety improvements, both ANP and FANP were shown to provide more reasonable rankings than AHP, with FANP providing the best results among the methods. After the locations are ranked and prioritized for improvements, the next step is to evaluate the potential countermeasures for improvements at the selected top-ranked locations. In this step, the standard practice has been to use Crash Modification Factors (CMFs) to quantify the potential impacts from implementing specific countermeasures. In this research, CMFs for bicycle crashes on urban facilities in Florida were developed using the Generalized Linear Model approach with a Zero-Inflated Negative Binomial (ZINB) distribution. The CMFs were tested for their spatial and temporal transferability and the results show only limited transferability both spatially and temporally. The CMFs show that, in general, wider lanes, lower speed limits, and presence of vegetation in the median reduce bicycle crashes, while presence of sidewalk and sidewalk barrier increase bicycle crashes. The research further considered bicycle exposure using the bicycle activity data from the Strava smartphone application. It was found that increased bicycle activity reduces bicycle crash probabilities on segments but increases bicycle crash probabilities at signalized intersections. Also, presence of bus stops and use of permissive signal phasing at intersections were found to increase bicycle crash probabilities.