Fixed-Route Transit Ridership Forecasting and Service Planning Methods (2006)

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.

5INTRODUCTION This chapter summarizes findings from a literature review related to transit ridership forecasting. A TRIS search was conducted to aid the review. In addition, a message was posted on the Travel Model Improvement Program e-mail list (see http://tmip.tamu.edu/email_list for more informa- tion) describing the synthesis project and requesting assis- tance. Additional studies for inclusion in the literature review were suggested by several respondents. OLDER STUDIES Groundbreaking work on transit ridership forecasting goes back more than 30 years, with considerable activity in the 1970s and 1980s. The Urban Mass Transportation Adminis- tration sponsored several studies at specific transit agencies, including the Greater Cleveland Regional Transit Authority in Ohio (1). Other researchers in the 1980s developed tech- niques to forecast route-level or system-level ridership, either through regression models using some combination of service levels, fares, population, population density, employ- ment, distance from the nearest stop, automobile ownership, gasoline price and supply, or through a modified four-step travel model (2–5). A 1983 report summarized the use of route-level ridership prediction techniques (6). The authors identified eight differ- ent types of transit changes (seven service-related plus a fare change) that use ridership prediction techniques, and charac- terized four general techniques: • Professional judgment; • Noncommittal or stated-preference surveys; • Cross-sectional models, ranging in sophistication from similar routes and rules of thumb to regression analy- ses; and • Time-series models, including elasticity-based ap- proaches and trend analysis. The various techniques were ranked subjectively on a num- ber of factors, but concluded that insufficient information was available to address accuracy and transferability. Two broader works that encompassed examples from throughout the United States and the world were published and/or updated between 1977 and 1981. Impacts of Changes in Fares and Services (7) and Traveler Response to Trans- portation System Changes (8,9) served as key source docu- ments for a generation of transit planners attempting to quantify the impacts of various types of service and fare changes. MORE RECENT STUDIES TCRP is sponsoring an update and expansion of the TCRP Report 95: Traveler Response to Transportation System Changes. Interestingly, one finding is that whereas much of the detailed information regarding transit service changes is old, it is not out of date. Several chapters have been finished and are available on the TCRP website, with the entire hand- book scheduled for completion in 2006. The 19 chapters in the final handbook address the following subject areas (10): • Multimodal/intermodal facilities, • Transit facilities and services, • Public transit operations, • Transportation pricing, • Land use and nonmotorized travel, and • Transportation demand management. Each chapter summarizes traveler responses to the spe- cific type of change addressed, discusses underlying factors contributing to the traveler response, provides related infor- mation and impacts, and presents case studies and examples. The most relevant chapters for this synthesis, Chapter 9, “Transit Scheduling and Frequency,” and Chapter, 10 “Bus Routing and Coverage,” have been released. Chapter 9, “Transit Scheduling and Frequency” (11), describes ridership response to changes in frequency in terms of service elasticity, with an average elasticity of 0.5. (Elas- ticities are generally used to estimate short-term changes in ridership in response to fare or service changes.) Higher elas- ticities are seen in cases where initial service levels are low (e.g., one bus per hour) and among higher-income riders. Recent examples show frequency elasticities grouping around either 1.0 or 0.3, with the higher elasticities seen in sub- urban systems and the lower elasticities in urban systems. It was noted that service reliability, clock face schedules that are easy to remember, the condition of the transit fleet, and timed transfers affect the response of riders to frequency changes, but are difficult to quantify. CHAPTER TWO LITERATURE REVIEW

Chapter 10, “Bus Routing and Coverage” (12), uses service elasticities as the measure of ridership response to service expansion and riders per hour or per capita as the measure of ridership response to successful new areawide transit systems. The authors report an average elasticity in the range of 0.6 to 1.0. First year ridership on new bus systems averages three to five trips per capita or 0.8 to 1.2 riders per bus mile. Service restructuring is more difficult to quantify, but several factors contributing to operating efficiencies and ridership growth are reported including high service levels on major routes, consis- tency in scheduling, enhancement of direct travel and ease of transferring, quantitative investigation of travel patterns, and favorable economic conditions. Among other findings, flexi- ble service designs such as hub-and-spoke have a slight but not universal edge over grid systems. New bus routes take 1 to 3 years to realize their full ridership potential. A study that examined service and fare changes in Europe found that long-run elasticities (from 3 to 7 years) are larger than short-term elasticities by a factor of 1.84 (13), although it is more difficult to isolate changes from a particular action over a long period of observation. RAIL-ORIENTED STUDIES Several studies published since the 1980s have addressed rid- ership forecasting for rail systems. Although the FTA’s New Starts program mandates ridership forecasts, these forecasts are done in the context of a traditional four-step travel model. The Chicago Transit Authority developed a spreadsheet ver- sion of the Chicago Area Transportation Study’s mode choice model on its West Corridor project to forecast ridership in response to service revisions (14). Model inputs included line-haul times and costs and access times and costs. Results confirmed the importance of transit access. An example of ridership forecasting for commuter rail included a methodol- ogy based on historical passenger rail travel patterns, origin/ destination surveys, and population for an extension of pas- senger train service to San Luis Obispo, California (15). Rail passenger forecasters have also developed a quick-response approach using multivariate regression to examine the effect of station-level variables, including surrounding land use and service characteristics at a given station, on heavy rail, light rail, and commuter rail ridership (16). ROUTE-LEVEL STUDIES Two interesting papers addressed the issue of route-level rid- ership forecasting. Stopher (17) developed a model to predict ridership changes at route and time-of-day levels resulting from headway changes, route extensions, new routes, route shortenings, short-lines on existing routes, service span changes, or a combination of actions. Peng et al. (18) proposed a ridership model operating at the route segment level by time of day and direction. This model incorporated transit demand, supply, and inter-route effects in a simultaneous system. The 6 study noted that, although a service improvement increases ridership on a given route, it is likely to later cause a ridership decrease on parallel or competing routes. Neither of these models, although theoretically appealing, has been widely adopted by transit agencies. Several transit agencies have attempted to develop rider- ship forecasting procedures. Three procedures that have been published (and there are likely many more that are used inter- nally) were prepared for Lane Transit District (LTD) in Eugene, Oregon; Metropolitan Transit Development Board (MTDB) in San Diego, California; and the Capital District Transportation Authority in Albany, New York. These attempts have all taken advantage of GIS programs to iden- tify demographic and employment characteristics within walking distance of a given transit route. The LTD route-level ridership forecasting model used route ridership rates as the dependent variable and buffer-area demographics, service levels, and competition from other routes as independent variables (19). This effort developed separate least-squares regression models for four weekday time periods plus Saturday, and converted the models to elas- ticity form for use in forecasting. Median household income and vehicle service hours were the only variables to appear in more than one model. LTD is not currently using this model, citing difficulties in obtaining the required input data. The second effort developed a preliminary ridership model for urban bus routes operated by the MTDB in San Diego (20). As at LTD, models were developed by time period (three daily time periods); however, an all-day model was also developed. This effort found that service-related variables tended to overwhelm demographic and employ- ment factors. Also, the model was not transferable to other route types such as feeder routes or community circulators. New transit modes can also stimulate ridership forecast- ing estimates. Capital District Transportation Authority developed ridership projections in the NY5 corridor as part of a bus rapid transit (BRT) study (21). The forecasting tech- nique involved several steps, including a determination of which trips would be likely to shift to BRT and an assessment of the impacts of headway and travel time changes and other improvements on ridership. Headway elasticities were taken from Pratt and Coople (9), while the travel time elasticity was taken from Mayworm et al. (7). Other BRT case studies were used to estimate ridership impacts of branding, image, and amenity improvements. METROPOLITAN PLANNING ORGANIZATION STUDIES Not surprisingly, transit agencies are not the only agencies to prepare ridership forecasts. Several metropolitan planning organizations (MPOs), where modeling expertise is gener- ally focused, and other regional agencies have developed

7transit ridership forecasting tools that are more user-friendly than formal four-step travel models. As part of its Regional Transit Access Plan, the Georgia Regional Transportation Authority in Atlanta developed a sketch planning tool that produced ridership forecasts for various transit improvement scenarios (22). The flexible nature of this tool allowed for an iterative forecasting process in which refinements could be introduced to improve overall performance and effectiveness. Ridership forecasting using this sketch planning tool focused on rail, BRT, bus-only lanes, and streetcars running in traffic lanes. The Maricopa Association of Governments in Phoenix, Arizona, created the Sketch Plan Model, which estimates light rail ridership (23). This model uses a set of trip rate fac- tors developed from other light rail systems in the western United States. These factors are based on the number of households and the percentage of regional jobs within a given distance from a light rail station. Four geographic cat- egories are used for access and egress distances, resulting in 16 average trip rate factors. The North Central Texas Council of Governments has a transit analysis process integrated within its four-step travel demand model (24). Its major advantages are that it is sim- pler and faster than a full model run (4 h versus 12 h) and it uses the already available coded travel system. Although closer to a four-step model than a sketch planning tool, it results in faster model runs and is somewhat simpler to use. The Knoxville Regional Transportation Planning Organi- zation developed the Knoxville Transit Analysis Tool (KTAT) as an independent sketch-planning add-on to its regional travel demand model (25). Inputs to KTAT include a traffic analysis zone (TAZ) layer with socioeconomic data and a selection set on a line layer to define the route being tested. KTAT operates in TransCAD to produce an estimate of ridership per revenue hour based on a regression model with population density, mean household income, workers per vehicle, and retail employment density as independent variables. The independent variables are calculated for a one- quarter mile buffer around the route. The model resulted in an R-squared value of 0.835. The user guide cautions that the ridership per revenue hour is best viewed relative to other routes and not necessarily as an absolute forecast. However, this tool provides a means to test various routes to determine the most promising alternatives. CURRENT STUDIES A ridership forecasting tool that is still under development is Transit Boardings Estimation and Simulation Tool (T-BEST) (26), which is a model being developed for the Florida Department of Transportation (DOT) Public Transit Office that works with ArcGIS to simulate travel demand at the individual stop level. A resource paper in support of this effort presents a framework for forecasting stop-level transit patronage (27) and also provides a good overview of previ- ous transit modeling efforts. The T-BEST model accounts for network connectivity, temporal and spatial accessibility, time-of-day variations, and competing and complementary routes through the use of a wide range of socioeconomic and service attributes. Results can be aggregated to time period, day of the week, route segment or route, sub-area, or the entire system from the individual stop level. The model distinguishes between direct and transfer boardings and therefore can quantify trip-linking and provide a means of analyzing the effects of transfer opportunities on ridership. An earlier version of this model has been documented in the literature (28) after calibration using data from Jacksonville, Florida. T-BEST is now being applied in Broward County, Florida. Florida DOT plans to use the T-BEST model statewide for transit ridership forecasting. Research related to improved ridership forecasting tech- niques is continuing, as indicated by two papers presented at the TRB 85th Annual Meeting in January 2006. Lane et al. (29) presented a sketch-level ridership forecasting tool for light rail and commuter rail. This model improved on the 1996 TCRP Report 16 (30) by taking into account reverse- commute trips to employment areas outside the central busi- ness district and by introducing service-related variables such as travel speed, fare, and midday headways. Marshall and Grady described a sketch transit planning model for the Washington, D.C., region that supports transit/land use sce- nario analysis (31). This model better matches suburban tran- sit ridership, is sensitive to land use effects, and is less costly to use than the traditional four-step model. Transferability of this model to other regions is not clear. SUMMARY There are other ridership forecasting models. Several transit agencies have developed models for internal use and might not find it worthwhile or cost-effective to publish a report on the subject. The studies cited here provide a good cross section of work done in this area. The more straightforward approaches exemplified by Mayworm et al. (7) and Pratt and Coople (9) are more user-friendly and are appropriate for ridership fore- casts resulting from small-scale changes. Efforts at the MPO or state levels to develop simpler and more usable sketch plan- ning tools show promise. Transferability across different met- ropolitan areas has not been established and is an important factor inhibiting widespread use of ridership forecasting mod- els. T-BEST development in Florida may provide insight into model transferability. The intent of this synthesis is not to recommend one approach over another, but to catalogue the various forecast- ing procedures currently used by transit agencies. The fol- lowing two chapters describe the results of a survey of more than 30 transit agencies in the United States and Canada.