Two sets of models were developed in Project L03, the data-poor and the data-rich models. The data-poor models predict a measure of reliability as a function of just the mean Travel Time Index, except for the on-time measures. Because these data-poor models have but one independent variable, these simple models provide a great deal of versatility in estimating reliability. Data-poor models enable the use of straightforward equations for predicting reliability in sketch planning and in complex modeling systems such as a trip-based demand model married to a network model. The data-rich models predict a measure of the variability of travel time as a function of a number of important variables that Project L03 found to be meaningful explanatory variables: the demand-to-capacity ratio, lane-hours lost (due to traffic incidents or work zones), and rainfall. The data-rich models can be used to predict or estimate reliability when any of these causal variables appear in an equation and data are available. Both the data-poor and data-rich models were estimated from data collected over a year from a subset of urban freeway segments in seven cities. The data-poor models apply to all time slices throughout a day, whereas four sets of data-rich models concerning different moments of the TTI distribution (mean, 99th, 95th, 90th, 80th, 50th, and 10th percentiles) were estimated for the peak hour, the peak period, the midday, and weekdays. The objectives of Project L33, Validation of Urban Freeway Models, were threefold: (1) attempt to validate the “data-poor” and “data-rich” models, (2) develop enhanced models if justified, and (3) promote acceptance and use of the L03 type of models for planning, programming, project development, design, systems operations, and further research. In conducting the validation, the research team was prohibited from using data that were used to estimate the data-poor and data-rich models. Validation data came from California, Minnesota, Utah (Salt Lake City), and Washington (Spokane) and totaled 323 segment-years covering both midday and peak periods. L03 models used data from some of these same places, but the same data were not used in the validation, as required. The project used two criteria for assessing the validity of the L03 models. The first was the difference between the predicted and measured values of the dependent variable. The second was whether the estimated models satisfied the assumptions of linear regression. This report describes the degree to which the different models perform well in terms of prediction and satisfying regression assumptions. The data-poor models predicted acceptably well as documented here but had some shortcomings in terms of satisfying the regression assumptions. Three sets of enhanced models were developed. The research team could not find satisfactory enhancements to the data-rich models. The degree to which the data-rich models predict well and satisfy the assumptions of linear regression is also described in the report. This report describes the methodology, data, conclusions, and enhanced models regarding the validation of two sets of models developed in SHRP 2 Reliability Project L03, Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies. The significance of the L03 models is they were among the first models that could be used to predict travel time reliability. Loosely speaking, reliability is defined as how travel time changes over time. More rigorously, reliability is defined as “the level of consistency in travel conditions over time . . . measured by describing the distribution of travel times that occur over a substantial period of time.”1 Specific reliability measures can be derived from the travel time distribution, such as the standard deviation and the Travel Time Index (n), or TTIn, which is the nth percentile of the travel time distribution divided by the free-flow travel time. (Author/publisher)
Abstract