Predicting Ridership and Travel Time Impacts of Bus Service Changes Using Sketch Planning Methods

Abstract

Bus service changes range in scale, and understanding their impacts on ridership and travel times can inform decision-making as changes are considered for the bus network. Budgetary limitations are at the heart of service change decisions, resulting in the need for analysts to assess different scenarios and accommodate quick turnarounds. This thesis provides a sketch planning framework for predicting ridership and travel time impacts of bus service changes, with a focus on direct demand models and the use of an open-source multimodal routing algorithm. The framework is designed to be streamlined with the use of data sources and capabilities, such as exporting a General Transit Feed Specification (GTFS) feed of a given bus network scenario, that agencies may have access to through existing transit planning tools.

Direct demand models are developed to estimate bus ridership at the level of approximately one-mile route-segments and time-of-day periods. This level of analysis provides a more disaggregated evaluation of bus ridership than past direct demand models. The models are sensitive to both route and network improvements. New variables designed to capture the relationship between bus routes, including the competitive and complementary nature of routes, are introduced and incorporated in the model development process. These models are developed for the Washington Metropolitan Area Transit Authority (WMATA). A case study analyzing two scenarios in WMATA's Better Bus Network Redesign (BBNR) is presented, with selected route examples to illustrate how the models capture different types of service changes. These routes fall under three categories: routes with no major service changes, routes with improvements in frequency, and routes with re-routing and other improvements.

An open-source multimodal routing algorithm, available through an R package called r5r, is used for travel time analysis. r5r calculates a distribution of door-to-door travel times for a given origin-destination (OD) matrix and returns a selected percentile value from the distribution for each OD pair. The percentile parameter is calibrated through a comparison of estimated travel times and actual travel times recorded in origin-destination-interchange inference (ODX) data. Low percentile values were found to provide travel times close to actual travel times. Additional guidance is provided for interpreting travel times from r5r, and use cases related to calculating travel time impacts between scenarios and evaluating rail competitiveness for a given bus network are explored.