Value of Travel Time Reliability in Transportation Decision Making: Proof of Concept—Portland, Oregon, Metro (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

2Travel time reliability is a significant aspect of transportation system performance. Recent work has shown that reliability has value to travelers and that their behavior is influenced by it. A wealth of research findings related to the second Strategic Highway Research Program’s (SHRP 2) Reliability Program has provided numerous instances of theoretical and empirical evidence for this fact. Reliability influences decisions about where, when, and how to travel. Because of the extra economic cost of unreliable travel on users, it has been increasingly rec- ognized that transportation planners and operators need to include these costs in the project planning, programming, and selection process. SHRP 2 has launched several studies investigating theoreti- cal and practical methodologies to measure, value, and pre- dict automobile travel time reliability. In the present project, the L35A research team sought to develop a theoretically sound, practical, viable, and flexible local method based on prior related SHRP 2 project findings to incorporate travel time reliability into the project evaluation process for multi- modal project planning and development. In other words, this project does not attempt to “reinvent the wheel” and repeat the research activities undertaken by prior SHRP 2 projects but to build the knowledge formulated by these various studies into an integrated framework and procedure in a practical real-world case study that effectively engages policy makers and elevates their understanding and confi- dence that SHRP 2 Reliability research products can properly represent project performance characteristics and inform decision making. The two foundational bodies of knowledge for L35A are SHRP 2 L03 and C10B. The L03 project developed an analyti- cal framework and procedure to estimate reliability measures for a traffic network. In a nutshell, Reliability project L03 established various relationships between the mean and vari- ance of travel time. Generally speaking, the higher the mean travel time is, the higher the travel time variance is. The pro- posed method also accounts for nonrecurrent event–induced travel time variance. The advantage of this framework for regional modeling lies mainly in its computational tracta- bility because this method does not require running Monte Carlo simulation–type random processes such as those pro- posed in the SHRP 2 L04 Reliability project. The second building block for this project is the fine-grained, multimodal dynamic traffic simulation and assignment net- work models created from SHRP 2 C10B. Multimodal models capture travel time reliability for both automobile traffic and public transit services. These integrated network models allow the modeler to depict time-varying network traffic dynamics that, through the skim feedback process, can better inform the travel demand forecasting model in estimating trip generation and distribution. The agency member of this project, metropolitan planning organization Portland Metro of Portland, Oregon, is a pio- neering transportation planning agency that has adopted many innovative modeling techniques in the last few decades. Metro modeling staff once suggested a desire to assess hours of congestion and reliability measures in the project evalua- tion process, but they never implemented that wish because Metro’s existing travel demand model cannot forecast unreli- able areas in the existing framework. Traffic reliability was introduced in the Transportation Exist- ing Conditions Report using INRIX data from the past 3 years to illustrate a buffer index using the 95th percentile travel time compared to average travel time. This report marked the first time Metro addressed reliability in its reporting, and addressed in combination with a congestion map (also produced using INRIX), the reliability map was useful in identifying trans- portation needs. Leveraging the research methodologies and outcomes produced by this research, Metro was able to use reliability measures to improve the development of Metro’s overall transportation improvement program and specific project programming. In this way, projects intended to provide either a primary or a secondary benefit in improving travel time reliability can be prioritized like any other project. C h a p t e r 1 Introduction

3 through a variety of measures of performance. As a result, a critical advantage of the framework is that reliability can be cap- tured for both automobile and transit travel, capturing multi- modal measures of travel times and of travel time variability. The result of the dynamic transportation network assign- ment process is a set of metrics on automobile users’ and tran- sit passengers’ travel times and travel time variability. Using these existing and emerging methods, which have been well documented in other SHRP 2 Reliability projects, the travel times and variability measure(s) can be used to determine a resulting generalized cost of travel. In this project we chose the methodology proposed by SHRP 2 Project L03 Analyti- cal Procedures for Determining the Impacts of Reliability Mitigation Strategies (Cambridge Systematics 2013a). This method allows the recurrent- and nonrecurrent congestion– induced reliability measures to be computed analytically, accommodating the computational requirements for large- scale modeling. These reliability measures can be fed back to the travel demand modeling process for the purposes of capturing the effects of reliability in both mode split and trip distribution steps. This procedure continues in an iterative manner until the generalized cost is convergent and stable. The convergent travel times and travel time variability mea- sures from dynamic network assignment can then be converted into more direct measures of generalized cost and/or general- ized time. This conversion is important because it allows direct monetization of VTTR and the value of travel time (VOTT). These two measures, which have been extensively studied through the SHRP 2 program and in previous research, give greater economic meaning to the value of reliability. We then have the ability to convert the generalized cost (or time) into a total measure of benefit (or cost) to travelers under certain projects and under given planning scenarios. These benefits and/or costs, comprising both travel time and reliability, can be turned into criteria fitting into the exist- ing project evaluation process. Portland Metro’s existing proj- ect evaluation criteria include, but are not limited to, traffic volume, capacity ratio, vehicle miles, automobile and transit travel times, average automobile and transit work trip travel times, and in-vehicle transit travel times. 1.3 Modeling platform and Framework 1.3.1 Metro Existing Travel Demand Model The current Metro demand model is considered to be an enhanced trip-based demand model. It has gained the accred- itation of the Federal Transit Administration and the Federal Highway Administration for use in infrastructure investment projects. 1.1 research Goals and Scope The overarching goal of L35A is to establish a locally acceptable method for determining the value of travel time reliability (VTTR) and show how to apply such obtained values, as well as travel time reliability measures, to project evaluation and program development following a defined modeling process. Critical objectives serving this goal included the following: 1. Developing travel time reliability measures. The adopted method primarily follows the L03 framework and approach (in lieu of Monte Carlo simulation types of methods, such as those used in SHRP 2 L04) to facilitate the computa- tional requirements of regional modeling; 2. Developing a locally accepted method to estimate the value of reliability. Such a method was originally proposed to be adopted from the literature review, but the actual method used is a simplified stated-preference model similar to that used in a recent Dutch study (Significance et al. 2013); 3. Generating reliability measures in a multimodal, dynamic traffic assignment (DTA) framework to allow reliability measures to be captured in a time-varying manner; 4. Revising Metro’s existing project evaluation process to incorporate VTTR in an integrated feedback framework; 5. Presenting results from a real-world case study on how travel time variability affects cost-benefit results and project ranking and discussing what previously unavailable insights could be supported by the analysis afforded by the proposed integrated modeling approach; 6. Providing a case study that would be primarily conducted by the technical staff of the participating transportation agency. This requirement allowed the agency staff to obtain realistic first-hand experience and thus be able to arrive at a realistic assessment of user and market readiness of the SHRP 2 research products; and 7. Presenting the research findings to local stakeholders and learning whether the proposed overall methodology and findings elevated their understanding and willingness to continue adopting reliability measures in future project assessment and selection processes. 1.2 Overview of the research Framework The overall research framework is illustrated in Figure 1.1. The top left of the figure shows a modified representation of the traditional steps in travel demand modeling, but replac- ing the common static traffic and transit assignment with a dynamic transportation network assignment that covers both road travel and dynamic transit assignment. The dynamic traf- fic and transit assignment models used in this project are fully capable of capturing time-dependent, travel time variability

41.3.1.1 Inputs Demographic data for input into the demand model process included the number of households in each traffic analysis zone stratified by household size, income group, and age of house- hold head. Employment data were stratified by nine employ- ment categories. Several separate models were created to provide inputs into different steps of the model. These included models to estimate the number of workers, number of children, and automobile ownership by household categories. Urban form descriptors, which refer to the spatial pattern of urban activities, are an important part of the model due to their influence on trip making. An accessibility measure was calculated that reflected the proximity of households to shop- ping and job opportunities. Local intersection density was also determined. Transportation system supply data were created for the purpose of determining travel times and generalized costs. Travel time data were created for two time periods, an AM peak two-hour period and a midday one-hour period, for the auto, transit, park-and-ride, walk, and bike modes. Transit skims included in-vehicle, walk, first wait, transfer wait, and Trip Generation Dynamic Network Assignment Trip Distribution Mode Split Travel Time and Travel Time Reliability Generalized Cost of Auto & Transit Travel Time to Convergence Cost and Benefit Analysis Project Evaluation Project Prioritization/Program Economic and Demographic data Value of Time Value of Travel Time Variability Scenario: Northwest Corridor Analysis Sensitivity Analysis Figure 1.1. Overall model framework. number of transfers. Trip cost data included automobile operating costs, parking costs, and transit fares. Household survey data were used to estimate the percentage of peak versus off-peak travel for each trip purpose and were used in mode choice. 1.3.1.2 Trip Generation The trip generation step created trips for eight trip purposes: home-base work, home-base shopping, home-base recreation, home-base other, non-home-base work, non-home-base non- work, home-base school, and home-base college. For each traffic analysis zone, the number of households in specific demographic categories was multiplied by a production rate. For example, for the home-base work trips, the trips were cal- culated using the number of workers in the household, which was calculated in the worker model. For home-base work and home-base college, trip attrac- tions were calculated and productions and attractions were balanced during destination choice. For the other trip pur- poses, attractions were not calculated directly; the produc- tions were singly constrained. The magnitude of employment

5 (by type and associated parameter) served to indicate the relative attractiveness of a destination zone during destination choice. 1.3.1.3 Destination Choice The destination choice models were developed using a multi- nomial logit estimation procedure. Each of the eight trip pur- poses had a separate set of variables and coefficients. Household trips by income levels were distributed separately with different variables and coefficients for each stratum, including the employment categories at the destination zone. This feature allowed households to be linked with jobs that reflected an appropriate level of income. Logsums were used in the destina- tion choice step to provide sensitivity to changes in multimodal accessibility. These logsums were created from the variables and coefficients used in mode choice. 1.3.1.4 Mode Choice The mode choice models were developed using a multinomial logit estimation process. Household characteristics such as income level, automobile ownership, and household size were used, as well as travel time, urban form descriptor, and cost vari- ables. The modes included in the demand model were drive alone, drive with passenger, auto passenger, walk access transit, auto access transit, bike, and walk. An important model feature was the incorporation of results from a stated-preference study that quantified travelers’ percep- tions of different transit vehicles and transit stops. The model reflects the differences in how travelers view in-vehicle time by vehicle type (i.e., bus, rail) as well as the type of stop (i.e., large transit center, small shelter, and pole stop). Another notable feature was the creation of a formal bike utility that reflects the attractiveness of a bike route as measured by key indicators (e.g., gradient, bike lanes/paths/trails, ambient traffic, inter- section controls). The park-and-ride mode now has an ele- ment concerning lot choice based on a lot’s utility; this element also responds to lot capacity constraints. An important element in the model is the delineation of time periods. Trips for each type of purpose were stratified into those made during peak periods and off-peak periods. Each time stratum was populated with attribute values that reflected the characteristics of the temporal period. 1.3.1.5 Trip Assignment For the static automobile assignment, a capacity-restrained equilibrium assignment was used. Transit assignments were also performed. For most scenarios, a PM two-hour peak and a midday off-peak assignment were analyzed. Feedback mech- anisms were in place because the travel times are fed into the demand model during the iterative process. Metro has also developed DTA capabilities for use in project analysis. 1.3.2 Multimodal Simulation and Assignment Framework (DynusT and FAST-TrIPs) The University of Arizona has developed an integrated frame- work for dynamic automobile and transit assignment and simulation by using the open source software tools DynusT (Dynamic Urban Systems for Transportation) and FAST- TrIPs (Flexible Assignment and Simulation Tool for Transit and Intermodal Passengers). The interactions of these two software tools are described in Section 5.3. DynusT has been used extensively as a DTA tool in the United States, including playing leading roles in SHRP 2 Projects C10B and L11, and in many local applications and deployments of DTA. The FAST-TrIPs software tool has also been used in several recent projects, including SHRP 2 Project C10B. DynusT is used by Portland Metro, which has developed a working regionwide DTA-compatible automobile network. Network attributes include lanes, speeds, and facility types, as well as intersection-level details (number of turn bays, length of turn bays, lane-to-lane movements, signal phasing, and permitted–protected left-turn treatments). This network can be easily imported into any of several DTA packages, includ- ing DynusT. In addition, Metro has worked on improving input trip tables for use in DTA. Currently, Metro is able to produce trip tables segmented into 15-min demand periods. These trip tables come directly from the travel demand model and are easily formatted to work with any DTA package, including DynusT. Figure 1.2 illustrates the Portland Metro DTA model in DynusT. The team also adapted the existing transit data at Metro to develop and implement FAST-TrIPs for Metro. This involved converting Portland Trimet files corresponding to Google’s General Transit Feed Specification (GTFS) into a network format that matched the DynusT network. Three main parts of the integrated model are differentiated. The demand model, in the form of a tour-based (or activity choice) travel demand model, provides the outputs, contain- ing travelers with a given origin, destination, preferred arrival time, and transportation mode. These data are separated into auto, transit, and intermodal trips and are fed into the assign- ment models in DynusT and FAST-TrIPs. As a DTA model, DynusT includes transit vehicle simulation, as well as auto and truck vehicle simulation. For the initial transit run in DynusT, a fixed dwell time of the transit vehicles is assumed and fed into the first iteration of the simulation. The main outputs of the DTA model are auto skims, which are used to check the convergence of DTA and are also fed back to the travel demand model to adjust the demand for the next iteration. Another

6Figure 1.2. Portland Metro DynusT model. output of DynusT is the vehicle trajectories, including the departure times of the transit vehicles at each stop, which are passed to FAST-TrIPs to simulate passenger movements in the transit network. The transit part of the model receives the vehicle skim by the experienced vehicles, including public transit vehicles, and simulates the passengers in the transit network using the path choices coming from the transit assignment. Pas- sengers are assigned on a transit schedule network accord- ing to their stochastic behavior along a given hyperpath in which every passenger will have an elementary (or single) path. Passengers are then simulated along the assigned paths, on which it is possible to have some missing-a-bus cases caused by late arrival or capacity constraint. As iterating the integration model continues, the passengers with missing- a-bus will disappear as link travel times are converged. Ulti- mately, the transit simulation in FAST-TrIPs produces a transit skim with access, waiting, in-vehicle, and transfer times. The boarding and alighting results from the simulation are used to generate vehicle dwell times, which are used in the next iteration of the vehicle simulation in DynusT. Finally, the transit skim data are fed back into the travel demand model. 1.4 report Structure This report is structured as follows: • Chapter 2 provides an overview of the integrated model. • Chapter 3 presents the agreed local method for determining reliability measures and VTTR. • Chapter 4 discusses the process for prioritizing operational and capital improvement. • Chapters 5 through 7 describe the modeling and scenario analysis details. • Chapter 8 describes the second workshop, in which the research findings were presented to the policy group and feedback was collected. • Chapter 9 concludes this report. • Chapter 10 lists all cited literature. • Appendix A presents the Southwest Corridor work plan. • Appendix B describes the DynusT–FAST-TrIPs integration run shown in DynuStudio.