Studies have documented the significant effect of horizontal curvature on operating speed on two-lane rural highways. The geometric design of these facilities emphasizes the forgiving roadside to accommodate the higher vehicular speeds. High-speed geometric design is predicated on selecting design values for geometric elements that promote speed consistency and safety. The low-speed environment has different objectives in trying to provide access and accommodate multiple roadway users, such as bicyclists and pedestrians. The goal is to maintain lower speeds and thus to achieve the functionality of the roadway and improve overall safety. Too often, the speeds on these facilities exceed the intended target speed of the roadway. A study conducted by the Pennsylvania Transportation Institute supported several research efforts in the low-speed environment. Presented is a more sophisticated analysis of low-speed urban street data using mixed models. A mixed-model statistical approach with repeated measures is used to analyze the influence of geometric elements on operating speed. The power of a mixed-model approach is that it accounts for the random effect in the database (such as the data collection sites themselves) while modeling the fixed geometric effects. Because data were collected at several points along each roadway, the analysis also applies a repeated-measures approach that addresses the geometric elements effect on the same subjects traversing a roadway. The advantages and disadvantages of applying a more sophisticated statistical approach are presented.

This study compares different statistical approaches to modeling the geometric and driver effects on operating speeds along low-speed urban streets. Linear regression on speed data obtained through data aggregation, linear regression on individual speed data, and panel analysis are discussed. Data collected from ongoing research examining operating speed on low-speed urban streets were modeled by the three techniques. The findings of the modeling techniques are compared and their influence on predicting probable operating speeds of a facility are presented. Traditionally, empirical analysis of operating speed has relied on regression models, using descriptive statistics such as 85th-percentile speed or mean speed to describe the data. This study demonstrates how the use of descriptive statistics obtained through data aggregation misleadingly reduces the total variability and nature of the variability associated with the statistical relationship. The fit of the regression function may appear to be increased, but the influence of the geometric elements may be overstated or understated. Data aggregation also affects inferential and prediction measures. Predictions from models based on aggregate data may appear to be more precise, but this does not imply that they are more reliable. Regression models of speed choice at a specific location within the roadway alignment may explain the effect of geometry but may not capture the effect of individual driver speed choice. As demonstrated in this study, the individual driver effect and geometric variable effect are important. The preliminary conclusion is that the driver's speed choice is highly dependent on roadway geometry and individual driver behavior.

No abstract