Open Access
Cruzado, Ivette
Graduate Program:
Civil Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
March 05, 2009
Committee Members:
  • Eric Todd Donnell, Dissertation Advisor
  • Eric Todd Donnell, Committee Chair
  • Paul Peter Jovanis, Committee Member
  • Venkataraman Shankar, Committee Member
  • Rana Arnold, Committee Member
  • William D Burgos, Committee Member
  • transition zones
  • speed prediction models
  • multilevel models
Rural highways provide connections between developed areas. In many instances, two-lane rural highways that pass through undeveloped areas provide high levels of mobility that are accompanied by posted speed limits that exceed 45 mph. However, it is common for two-lane rural highways in Pennsylvania to pass through low-speed, developed areas (i.e., rural villages) with posted speed limits that are 35 mph or less. The roadway section between the high- and low-speed environments is referred to as a transition zone. In some cases, transition zone design may be accompanied by changes in roadway geometric features; however, it is hypothesized that drivers fail to adjust their speeds to comply with the change in the regulatory speed at the low-speed end of the transition zone. In other instances, drivers are only informed of the posted speed limit changes by regulatory signs with no corresponding changes in the roadway geometry. Speed data were collected at 20 two-lane rural highway transition zones in central Pennsylvania. At each study site, speed data were collected at four locations: 500 feet before the transition zone, at the beginning of the transition zone, at the end of the transition zone, and 500 feet after the transition zone. The location of the sensors permitted vehicles to be “tracked,” thus the final analysis database included four speed observations collected from 2,859 individual drivers for a total of 11,436 speed observations. Highway characteristic data were also collected at each location, including geometric design features, roadside elements, and access density, among others. The primary objective of this research was to develop speed prediction models to explain the relationship between the roadway features present along a two-lane rural highway transition zone and driver operating speeds. Two general model specifications were considered based on the available speed data. These included point speeds based on the “tracked” vehicles, and speed differentials between successive data collection points in a transition zone. In the point speed analysis, four repeated speed measurements were collected on each of the 2,859 drivers across 20 different sites. Longitudinal models were used to model these data and compared to the more traditional operating speed modeling approach, ordinary least squares (OLS) regression. Use of OLS regression violates the assumption of independent observations. The longitudinal models considered in this research were panel data models using both the fixed and random effects estimator, multilevel models, and generalized estimating equations (GEE). From the results of the analyses it was concluded that a three-level model in which speed observations were nested in drivers and drivers were nested in sites is more appropriate in explaining the influence of highway characteristics on driver speeds along two-lane rural highway transition zones. Key relationships between highway features and mean operating speeds in transition zones are as follows: - When compared to a posted speed limit of 55 mph, a speed limit of 45 mph is associated with a mean operating speed reduction of approximately 3.5 mph. A speed limit of 25 mph is associated with a mean operating speed that is approximately 10.5 mph lower than the baseline of 55 mph. Similarly, a posted speed limit of 35 or 40 mph is associated with a mean operating speed that is approximately 2.4 mph lower than the baseline of 55 mph. - Wider travel lanes and lateral clearance distances are associated with higher operating speeds along two-lane rural highway transition zones; a mean operating speed increase of 2.4 mph is expected per one-foot of lane width increase while a one-foot increase in lateral clearance is associated with a mean operating speed increase of 0.15 mph. - The presence of curb is associated with a mean speed reduction of approximately 4 mph while the analysis indicated that a mean speed reduction of 1 mph is associated with a one-unit increase in driveway density. - The presence of Intersection Ahead and School/Children warning signs were associated with 2 and 1 mph mean speed reductions, respectively, while the presence of a Curve Ahead warning sign was associated with a mean speed increase of almost 1 mph, when compared to the baseline of other warning sign types. - Finally, the presence of a horizontal curve was associated with a mean speed reduction of 1.5 mph; if the horizontal curve is combined with a warning sign, a mean speed reduction of almost 3 mph is expected when compared to the baseline of a tangent roadway section. The results from the three-level model also provided the standard deviation associated with each level of the model hierarchy. The standard deviations of the random components from the model developed were: 3.1 mph for highest level (site cluster), 2.1 mph at the second level (driver cluster), and 6.5 mph at the lowest level (speeds). A second data set was created in which the response variable was change in speed along the transition zone. By considering speed change as the response variable, only one data point per vehicle (driver) was available; however, a site cluster could still be considered in the model specification. Use of the speed differential as the dependent variable in a statistical model eliminated part of the repeated observation issue present in the point speed analysis. As such, two general modeling methods were considered. These included OLS regression and multilevel models in which speeds were nested in sites. The variables that were consistently associated with speed reductions across all models were changes in the posted speed limit, reduction in paved shoulder width (1 mph reduction per one-foot reduction in paved shoulder width), number of driveways (0.36 mph reduction per one-unit increase in driveway density), school/children related warning signs (8 mph mean speed reduction), length of transition zone (0.8 mph average speed reduction per 100 foot increase in transition zone length), and presence of horizontal curve that warrants a warning sign (3.2 mph mean speed reduction is expected with this type of horizontal curve). The presence of a Curve Ahead warning sign and tangent sections were consistently associated with a speed increase along transition zones across all models (3.2 mph average and 2 mph average, respectively). Several independent variables were not statistically significant in the multilevel speed differential model when compared to the OLS regression model. These included the change in lane width and in lateral clearance, presence of a curb, and Intersection Ahead warning sign. Although the standard errors of the parameter estimates obtained using OLS regression were smaller than those obtained using the multilevel models, the multilevel model is a better representation of the nesting structure of driver speed differential nested within data collection sites.