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.
INTRODUCTION Survey results provide a comprehensive overview of the major issues regarding transit ridership forecasting. Follow- ing a review of these results, six agencies were selected as case studies. Personnel directly involved with development and use of ridership forecasting methodologies agreed to be interviewed by telephone. In several cases, more than one person at an agency either participated in the interviews or reviewed the draft summary of the case study. The case stud- ies are intended to provide additional details on innovative and successful practices. The selection process for case studies had several criteria: (1) to include transit agencies of various sizes in different parts of the country, (2) to include a variety of approaches and methods related to ridership forecasting, and (3) to select agencies that could offer useful insights to the transit indus- try as a whole. Nearly 70% of responding agencies offered to serve as a case study and, as shown by examples from non- case study respondents in chapter three, these agencies offered very interesting responses based on their experi- ences. The six agencies chosen do not necessarily consider themselves as examples of best practices; however, together they provide a representative overview of the state of transit ridership forecasting. The six case study agencies are: ⢠VIA (VIA Metropolitan Transit), San Antonio, Texas ⢠RTD (Regional Transportation District), Denver, Colorado ⢠GRTC (Greater Richmond Transit Company) Transit System, Richmond, Virginia ⢠NYCT (Metropolitan Transit AuthorityâNew York City Transit), New York, New York ⢠OCTA (Orange County Transportation Authority), Orange, California ⢠TriMet (Tri-County Metropolitan Transportation Dis- trict of Oregon), Portland, Oregon. The case studies summarize survey responses and inter- view observations from each agency. The interviews explored issues raised by the survey responses in greater depth and also included a question regarding the value of passenger forecasting as experienced by each agency. 20 VIA METROPOLITAN TRANSIT (SAN ANTONIO, TEXAS) One person in the operations planning department has responsibility for ridership forecasts. The data collection process previously relied on a staff of nine checkers with hand-held devices that could be linked directly to an Access database. VIA has acquired 50 APCs for its bus fleet; how- ever, the APC data are not viewed as accurate. Ridership forecasts now use farebox data, available in 30-min incre- ments and summarized by time of day and day of the week. This information is supplemented with origin/destination data gathered by means of an onboard survey every 5 years; also broken down by time of day and day of the week. VIA develops a ridership forecast for any type of service change other than very minor schedule adjustments as part of its ser- vice revision form. This form is circulated internally up to the Chief Executive Officer, who at his discretion may share it with board members. Board approval is only required when initiating new routes and discontinuing current routes. VIA considers several factors in developing ridership forecasts. System ridership is used for annual forecasts tied to the budget process. Existing route or route segment rider- ship serves as input for any service change. Farebox data can- not provide ridership at the route segment level; therefore, either APC data are used to estimate the percentage of board- ings along a given segment or checkers are sent out to collect the data. For new routes or added route segments, VIA looks at land use and economic trends within the new service area and relies on analysis of similar routes serving similar areas to forecast ridership. Origin/destination information and demographic factors are used primarily for service to new areas. Typically, a new development such as a major employ- ment or retail center will generate analysis of a new route or a route extension. The input data are considered reliable; however, VIA notes that the analyst needs to understand the limitations and accuracy of the data in terms of sample size and seasonal effects. Ideally, 3 days of ridership data should be available within the past 6 months. This is always the case for farebox data, usually for APC data, and usually not for onboard ridecheck data, especially on weekends. The forecasting techniques include rules of thumb, similar route analysis, and professional judgment. A similar route CHAPTER FIVE CASE STUDIES
21 analysis would examine productivity (in terms of passenger boardings per revenue hour) by time of day and day of the week for a similar route serving a similar area and would then apply the productivity values to the planned revenue hours of service by time of day and day of the week. For a headway change from 60 to 30 min, the expected productivity would be adjusted downward depending on type of route and reason for the change. VIA categorizes its routes as major radial, minor radial, crosstown, feeder/circulator, and express/limited stop, with different productivity standards for each category. All ridership forecasts are for 6 months after the service change, recognizing that it takes time for ridership to develop. Any new service is implemented on a 180-day trial basis and can be discontinued or altered if it does not meet the produc- tivity standards for its service category. Short-range forecasts are made for a typical weekday/Saturday/Sunday, whereas long-range forecasts are at the more aggregate level of annual systemwide ridership. VIAâs goal for ridership forecasting is that all forecasts be within 10% of actual ridership at the route or system level, and this goal has been met. The professional experience of the forecaster plays a significant role in understanding how trip productions and attractions, schools (especially middle schools), and timed transfers affect ridership. Field work is essential in developing and applying this experience. Signs of good transit potential can go beyond the obvious, such as high residential density and presence of major trip genera- tors, to factors such as the presence of oil stains and trans- mission fluid leaks on residential streets. Technology has made the forecasting process faster and more reliable, but has not changed the methodology itself. A ridership forecast for a simple route realignment can usually be generated within one hour. Ridership forecasts would be developed under the scenar- ios included in the survey as follows: ⢠Half-mile rerouting of an existing route to serve a new shopping center: base the forecast on similar routes and trips generated by similar retail developments with the understanding that it may take time to develop new retail customers. ⢠Extension of an existing route for one mile to serve a new residential development: base the forecast on sim- ilar routes and trips generated per dwelling unit based on median value of homes. Residential areas must have at least 200 occupied homes before service begins. ⢠Change in headway from 12 to 10 min during peak hours: forecast would depend on load factors and over- all usage in the transit corridor. A shift from five to six trips per hour would not increase ridership unless the route is at its maximum load factor and there are more potential riders in the corridor. ⢠Implementation of a new crosstown route: evaluate existing travel times and potential savings and extract origin/destination data for current riders. Obtain employee addresses from major employers along the proposed route. ⢠Implementation of a new mode such as BRT: evaluate existing travel times and potential savings. Look at sta- tion sites and potential for reroutes for transfers from other areas. Extract origin/destination data for current riders. Evaluate all trip generators within one-quarter mile of stations. Evaluate automobile drive times and traffic volumes. ⢠Prediction of next yearâs ridership as part of the budget process: base the forecast on the service plan included in the budget and past ridership trends. ⢠A 10-year ridership forecast as part of a long-range plan: base the forecast on past ridership trends, service revisions included in the plan, and expected growth and development. VIA would not necessarily make any changes to its cur- rent methodology, but would like to obtain more data linked to GIS. Use of GIS has made it much easier to develop visual representations of ridership activity. Lessons learned include the need to understand the limitations of the data used in rid- ership forecasting. This case study provides an example of a traditional approach that relies heavily on professional judgment and an understanding gained through experience of the factors con- tributing to transit ridership. The value of ridership forecast- ing is perceived to have declined as a dedicated sales tax and other funding sources have lessened reliance on farebox revenue. REGIONAL TRANSPORTATION DISTRICT (DENVER, COLORADO) RTD prepares ridership forecasts for most service changes except for minor adjustments to schedules or route segments. There is no explicit threshold triggering the need for a rider- ship forecast; however, a change of more than 10% in service hours suggests ridership implications, and any service reduc- tion indicates a potential loss of riders. Ridership forecasting is part of the general duties of staff members in the planning, operations planning, and budget departments, depending on the type of forecast being generated. Forecasts are distributed internally, to board members, and to stakeholders. The most common change is in route frequency. RTD uses a service elasticity of 0.5 to forecast the ridership impact of frequency changes, based on the average value from TCRP Report 95 (11). RTD calibrates this average elas- ticity upward or downward based on its previous experience, depending on existing route frequency, similar routes, and setting. For example, RTD has found a greater elasticity for headway improvements to infrequent service, with dimin- ishing returns seen on headway improvements to frequent
service. Also, headway improvements in response to over- crowding show a higher elasticity than improvements for other reasons. The inelastic nature of service elasticities has not been well understood in many communities served by RTD, and staff has been aided in educating policymakers by its documentation of prior experience and by summaries of experience in other areas. RTD also relies on service standards. Although these do not forecast ridership, standards do set a minimum threshold of performance for existing routes and proposed new service. RTD also evaluates route sustainability by examining popu- lation and employment per acre. RTD considers various inputs to its ridership forecasts depending on the type of change being analyzed. System rid- ership, ridership on similar routes, and demographic factors are considered for changes in span of service; timed transfers are an important component of RTD service, therefore consis- tency in span of service is important. A route deviation requires examination of route segment ridership. Ridership on similar routes, origin/destination information, and demo- graphic and land use characteristics are important for new routes and route extensions, and most of these factors also bear on forecasts for a new mode or corridor. Economic trends are factored into annual ridership forecasts for budget purposes. Along with service elasticities and service standards, RTD uses rules of thumb and similar route analysis in fore- casting ridership impacts of most service changes. A signif- icant change, such as a new mode or new corridor, calls for the regional four-step travel model, used for rail and long- range planning. Trend analysis is used for annual budget forecasting. Professional judgment is applied to all ridership forecasts to ensure reasonableness of the results. RTD uses ridership, origin/destination, land use, and cen- sus demographic data in developing its forecasts. New tech- nologies (including the introduction of APCs along with a focused effort to establish confidence in the APC data; new software that integrates ridecheck, supervisor point check, and APC data and converts it to a usable format for service planning purposes; GIS; and new, more reliable fareboxes) have improved the quality of ridership data. Having gone through a standard debugging period, RTD is now confident in the APC data and is developing new applications. For example, stop-level boardings, alightings, and loads are exported into GIS and are mapped along with population and employment density. RTD views an optimum amount of data as a balance among data availability, methods to analyze the data appro- priately, and the ability to present results in a meaningful way to decision makers. GIS has been very important in terms of presenting results. Service planners at RTD are moving toward the use of GIS in place of Microsoft PowerPoint in making presentations to the general public and to senior 22 management. GIS can better represent the complexities not only of the transit network structure but also of the planning analysis. RTD has combined GIS with aerial photographs to develop presentations that clearly articulate its proposals and rationales, and reports very positive reception by the public. As an analytical and communications tool, GIS has helped to build support for RTD initiatives. RTD is now incorporating origin/destination data from household travel surveys con- ducted in the counties within its service area every 5 years on a rolling basis into GIS. RTD assesses its ridership forecasting methods as gener- ally adequate for short-term service planning and is now much better documented with the release of TCRP Report 95. Experience in applying these methods will result in addi- tional refinements. Desired improvements include the avail- ability and accuracy of input data at the appropriate scale, inclusion of more predictive variables, and incorporation of TCRP Report 95 into service standards and guidelines. Ridership forecasts would be developed under the scenar- ios included in the survey as follows: ⢠Half-mile rerouting of an existing route to serve a new shopping center: first assess current ridership; estimate new ridership based on similar routes and shopping centers. ⢠Extension of an existing route for one mile to serve a new residential development: estimate new ridership based on similar routes and developments. ⢠Change in headway from 12 to 10 min during peak hours: use an elasticity of 0.5 to estimate the ridership impact of the frequency improvement. This elasticity may be adjusted up and down as suggested in TCRP Report 95 based on similar routes and settings and (more broadly) on existing frequency of service. ⢠Implementation of a new crosstown route: assess cur- rent ridership on related routes. Examine origin/desti- nation data. Consider the performance of similar routes. Analyze transfer data and evaluate the setting of the proposed route. ⢠Implementation of a new mode such as BRT: run the four-step travel model. ⢠Prediction of next yearâs ridership as part of the budget process: base the forecast on a trend analysis. ⢠A 10-year ridership forecast as part of a long-range plan: run the four-step travel model. In terms of one improvement to its forecasting method- ology, RTD sees value in the adoption of written guidelines for how to do service planning, including ridership fore- casting, in a rational way. These guidelines are not viewed as limiting planners to an inflexible approach, but rather as ensuring that key elements are addressed. Although these guidelines would be valuable internally, their primary value could be in helping potential partners such as city planning agencies in service area jurisdictions to understand transit
23 planning and the importance of including the transit agency in activities such as development review and location of new sidewalks. Lessons learned include the use of proven tools to develop realistic ridership forecasts and the value of integrating rid- ership and demographic data through GIS. An example of the latter is the inclusion in individual GIS layers of employment and population density by TAZ, boarding and alighting data by stop (presented as a pie chart whose size indicates level of activity and proportions represent boarding versus alighting activity at each stop), symbols representing total employ- ment and total population by TAZ, and symbolic representa- tion of passenger loads. A potential next step is the use of GIS network analysis tools to refine ridership forecasting methodologies. In terms of value, forecasting changes to ridership as a result of changes in service brings discipline to the service development process and highlights its focus on the cus- tomer. Service changes are directed toward improving per- formance and sustainability as defined by service objectives and standards. RTDâs objective is to serve the most riders for the budget dollar. Therefore, service changes that show improvements to subsidy per boarding or boardings per hour are beneficial. The forecast data needed are boardings, hours, and unit cost (less fare revenue). This information is (or should be) readily available internally and easily communi- cated to others. This case study shows how new technologies such as APCs, integrated software, and GIS can improve the quan- tity and quality of ridership and other data, provide new methods for analyzing and forecasting ridership, and greatly enhance the ability to communicate results to stakeholders. At the same time, research projects such as TCRP Report 95 provide invaluable documentation of experience elsewhere. RTD remains in the process of blending these factors to improve several aspects of its planning efforts, including rid- ership forecasting. GREATER RICHMOND TRANSIT COMPANY (RICHMOND, VIRGINIA) This case study is representative of many mid-sized and small transit agencies that do not prepare ridership fore- casts. GRTC serves Richmond, Virginia, and the surround- ing metropolitan area with 29 fixed routes and a peak bus requirement of 149 (based on 2003 NTD data). The major- ity of the routes are long-established and unchanging, and the nature of the system does not open up the need for rid- ership forecasting. In most cases, requests for new service come from local- ities within the service area. Implementation depends on identification of a funding source. If the localities are willing to fund a demonstration project, GRTC will design and implement a new route or a route extension. At the end of the demonstration period, ridership and performance is evalu- ated and the locality makes a decision whether to continue funding the service. Two recent examples of new service involve a new mall and a package of express bus and route-deviation local ser- vice to a locality that was previously not being served. No ridership forecasts were made for the mall service, but both the county and GRTC agreed at the outset that there was a good chance that ridership would justify the service. Evalu- ation of the new service to the mall was based on the countyâs perspective that the route worked to bring people to and from the mall. The package of services was a more interesting example. A consultant for the locality had developed the service pro- posals and included ridership projections that GRTC accepted. Service proved to be successful, but was discon- tinued when the locality did not have money to fund the routes beyond the demonstration period. Funding for the express service has been restored, and GRTC was scheduled to reinstitute express service in late February 2006. GRTC recently received a request from an elected offi- cial to begin a new route serving a small area of the city. Staff developed the parameters for the proposed service (including basic route design, frequency, and span of ser- vice) to be able to estimate the cost. Implementation is dependent on identification of a funding source to begin a demonstration project. When GRTC brainstorms ways to improve service to its riders, ridership forecasts are not a significant part of the equation. Identifying the service concept and assessing how riders would respond (based on professional judgment) are the key elements. As GRTC notes, the best analysis in the world would make no difference without a local funding source. GRTC periodically undertakes a comprehensive opera- tional analysis (COA) of its transit network. The upcoming COA will dovetail with a regional mass transit needs study led by the MPO. GRTC is considering the development of a ridership forecasting tool as an element of the upcoming COA. The agency recognizes that, as it develops GIS exper- tise, it will be better able to analyze demographic factors at the route level. Thus, a ridership forecasting tool could be very useful in the future. For the present, GRTC has managed its transit system well without such a tool, noting its high ranking among all transit agencies of its size in terms of cost- efficiency. This case study shows that there may not be a real need for a ridership forecasting methodology at all transit agencies. The decision-making process at many small and mid-sized
agencies is driven more by politics and funding availability than by ridership analysis. Although many agencies can see the value of employing a forecasting methodology, it may not rank highly in terms of current needs. This is a valid assess- ment in many cases and is a useful point to keep in mind for this synthesis project. MTAâNEW YORK CITY TRANSIT (NEW YORK, NEW YORK) MTAâNew York City Transit (NYCT) prepares ridership forecasts for most service changes except for minor service adjustments and scheduling changes. There is no specific threshold triggering the need for a ridership forecast. Signif- icant subway service changes, especially during peak hours, are likely to require a forecast. Bus service changes are less likely to trigger ridership forecasts. Management decides whether a ridership forecast is needed. There is a dedicated group that generates ridership forecasts, but they work pri- marily on major projects such as the Second Avenue subway, the Manhattan Bridge reconstruction, and BRT. Forecasts for other changes are sometimes done by transit analysts as part of their regular duties. The Operations Planning Department takes the lead for preparing most ridership forecasts, although the Office of Management and Budget typically prepares annual forecasts for budget purposes. Forecasts are distributed and used inter- nally. MTAâNYCT considers a wide variety of inputs for its forecasts, although system-level ridership is used primarily for annual forecasts. MTAâNYCT maintains and uses a detailed network model of all subway and bus routes in New York City and walking links for access and transferring. This network model is used to analyze current travel patterns by assigning subway origin/destination trip tables estimated from Metro- Card farecard transactions. It is also used to model future major service changes or additions by using census-based trip tables projected into the future. Shifts between bus and subway modes are estimated using this model; however, there is no provision within this model for attracting auto- mobile or taxi users because the existing transit share is gen- erally already high. Mode share modeling as well as regional impacts is addressed by a regional transit forecasting model that is maintained by the MTA, which is MTAâNYCTâs par- ent agency. It incorporates most of MTAâNYCTâs bus and subway network model. A service elasticity of 0.2 is used primarily to estimate the impacts of contingency service reductions. The Operations Planning Department and Office of Management and Budget work together to mine Metro- Card data. Introduction of the MetroCard has proven to be very use- ful for both subway and bus ridership analysis. The Metro- Card provides a record of subway station usage by time of day and bus route usage by direction and time of day. Linked trips 24 and transfer locations can be inferred through analysis of indi- vidual MetroCard use (MetroCard captures boarding data only). MTAâNYCT recently contracted with a private firm to construct subway and bus trip tables from raw MetroCard data; a significant undertaking given that the database holds more than 7 million transaction records for each weekday. In the interim, the agency has been using a subway-only method for inferring destination stations for each station entry and then using the resulting trip table with the network model to estimate subway travel patterns and route usage. For bus trip patterns, MTAâNYCT designed an iterative probability model to predict alightings at bus stops based on total boardings and alightings from ridechecks, the travel time between stops, and total passengers alighting at a spe- cific bus stop. This model produces an acceptable result within a few iterations. The resulting stop-to-stop origin/ destination trip tables are being used to estimate ridership on proposed BRT lines. Recent efforts focus on integrating sub- way origins with feeder bus alightings and combining Metro- Card and census journey-to-work data. The analysis of expected transfer levels to the new Second Avenue subway line has been used in some station designs. MTAâNYCT uses a variety of data sources beyond the MetroCard in developing ridership forecasts. The agency does not have APCs, but uses a large contingent of trained traffic checkers to gather data through ridechecks and pointchecks at peak load points and central business district cordon points. MTAâNYCT also relies on farebox/turnstile data, origin/destination data from travel models, census and Census Transportation Planning Package demographic data, existing and forecast land use, and economic trends and fore- casts. GIS programs have helped in organizing large collec- tions of data. New York City has developed a new GIS base map (NYCMAP) with high-quality aerial photographs, and MTAâNYCT is making increasing use of this map. The city has also developed a land use database in GIS. Technology has clearly had an impact on forecasting methodology. AFC equipment provides ridership boarding data in 6-min increments, which allows for origin/destination estimation. Improved personal computers and software permit more detailed methodologies that can be applied more quickly. Input data reliability is a problem in terms of the accuracy of pointchecks for on-board train volumes, along with the labor- intensive nature of collecting enough samples to compensate. Ridechecks are practical on buses, but not on subways. Short-term forecasts are based on ridership trends and known land uses, whereas long-range forecasts use detailed socioeconomic forecasts. Short-term forecasts can be com- pleted within 1 to 5 days by service planners, including time for supplementary ridechecks. A simple long-term forecast can be completed in one week; however, more complex forecasts of alternatives can take up to a year. Typically, one-quarter to one-half the time of two analysts is needed to
25 forecast the ridership impacts of major subway service changes or additions, either line-specific or at major station complexes. Ridership forecasting models are often used as tools to test various scenarios, and this can be an open-ended process until a satisfactory service plan is selected. MTAâNYCT is satisfied with the ridership forecasting methods in use and under development, but hopes to make these methodologies faster and easier to use. Needed improvements include the availability and accuracy of input data at the appropriate scale, fewer time-intensive method- ologies, simplification, enhanced accuracy, and flexibility to address a wide variety of situations. Ridership forecasts would be developed under the scenar- ios included in the survey as follows: ⢠Half-mile rerouting of an existing route to serve a new shopping center: ridership forecasts are generally not prepared for this scale of change. ⢠Extension of an existing route for one mile to serve a new residential development: ridership forecasts are generally not prepared for this scale of change to the bus system. A proposed one-mile westward extension of the #7 subway line in Manhattan was analyzed using MTAâNYCTâs a.m. peak-hour network assignment model and MTAâs regional travel forecasting model. The former uses a stochastic user equilibrium proce- dure, whereas the latter uses a Pathfinder procedure with capacity constraint added. ⢠Change in headway from 12 to 10 min during peak hours: this change is usually too small to model. How- ever, one additional peak-hour train does add capacity. ⢠Implementation of a new crosstown route: new crosstown routes have not been implemented for at least the past 10 years. ⢠Implementation of a new mode such as BRT: MTAâNYCT is treating BRT in similar fashion to limited-stop service in its transit trip assignment model (bus, subway, and walk network); however, for the first time the agency is estimating stop-to-stop origin/ destinations using ridecheck data (see above for a description of this process). New or induced transit travel resulting from the âattractivenessâ of BRT will be based on careful and realistic quantification of expected time and reliability benefits. These will be converted to added riders using elasticity-type methods from the literature and experienced BRT planners and operators. ⢠Prediction of next yearâs ridership as part of the budget process: base the forecast on year-to-date ridership trends at the time the forecast is being prepared. The next yearâs ridership change applies forecasts of New York City employment to the current yearâs estimate, with minor adjustments to account for calendar differ- ences. Additional changes are made as needed to account for planned service or fare changes. ⢠A 10-year ridership forecast as part of a long-range plan: begin by forecasting an origin/destination trip table based on existing trips and socioeconomic forecasts by the MPO. The calibrated network model of all subway, bus, transferring, and walk options in New York City is further calibrated as needed and then modified to reflect the long-range plan. A trip assignment model (shortest path/stochastic user equilibrium) is run to predict rid- ership for the long-range plan scenario by route and station/stop. Predictions of any significant shifts from or to automobile are obtained from the MTA model. The one improvement to forecasting methodology would be to make it easier to apply. MTAâNYCT reports several lessons learned from their experience: ⢠Neither overly simplistic nor overly complex approaches work. MTAâNYCT has tried to make its model as comprehensive and realistic as possible with- out getting bogged down in unnecessary details. ⢠By having a very good representation of existing and proposed services, the model serves two purposes: (1) as a structuring tool that allows service planners to bet- ter understand the details of scenarios and interaction with existing services and (2) for the production of actual ridership forecasts by scenario. ⢠AFC (MetroCard) data are a valuable source of current transit usage and transit information, including inferred origins and destinations. This overcomes some of the limitations of survey/census-based origin/destination data, particularly their tendency to be out of date. ⢠Care should be used when applying transit trip rates. A recent study of downtown Brooklyn revealed that tran- sit trip generation rates typically used for site-specific environmental analysis needed adjustment to produce accurate results using current data. ⢠An additional factor that emerged in the case study dis- cussions is the benefit of physical proximity between modelers and service planners. Both groups are part of Operations Planning, but being housed in the same building has encouraged interaction and in the process has improved the model. The model as a tool may be more important than the model as a producer of specific results. ⢠There is a need for a modified software package that can allow smaller systems to conduct the types of analysis done at MTAâNYCT while being easy to use and understand. The value of ridership forecasting at MTAâNYCT can be seen in several ways. Modeling provides a structure for plan- ning, and this is even truer with complex projects. For exam- ple, in the ongoing BRT work, modelers need details that force service planners to think through their plans in greater detail, to the benefit of the plans and the models. The model
also provides an accounting system that can identify any inconsistencies in underlying systems. As noted earlier, the process has fostered interaction between service planners and modelers and encouraged new ideas about model uses. MTAâNYCT is often not used as a case study because of its size relative to other transit agencies. This case study shows how application of new data collection techniques (AFC) and tools such as TransCAD can improve ridership forecasting procedures. Successful exploration of new ana- lytical methods (such as inferred origins and destinations) as ridership data become more reliable is an important finding that can be applied elsewhere. Encouraging interaction between modelers and end-users through organizational structure and location of the departments ultimately results in model improvements and greatly increases the likelihood of its being trusted and used on a consistent basis. MTAâNYCT maintains and regularly uses an in-house net- work model specifically for analyzing and forecasting tran- sit usage in New York City, whereas regional transit agency and the MPO maintain larger and more complex demand forecasting models that include suburban transit services and non-transit modes. This allows MTAâNYCT to focus on its transit service planning needs while improving and updating its route coding, which is passed along to the larger models. ORANGE COUNTY TRANSPORTATION AUTHORITY (ORANGE, CALIFORNIA) OCTA prepares ridership forecasts for virtually every service change. There is no specific threshold triggering the need for a ridership forecast; however, changes in route alignment or in the number of daily trips generate a forecast. Forecasts are used internally and are part of the general planning and mod- eling duties. The Operations Planning Department prepares ridership forecasts related to short-term service changes. The Planning Department forecasts long-term ridership as well as ridership changes for major service changes. The Finance Department, in consultation with the Operations Planning Department, prepares annual forecasts for budget purposes. OCTA uses systemwide ridership in its annual forecasts. Changes to routes or service spans use route and route segment 26 data as well as ridership on similar routes. A similar routes approach is also used for new routes. Origin/destination infor- mation and demographic factors are used in the mode choice model. Table 39 summarizes OCTA forecasting methods. For short-term service planning, similar routes and pro- fessional judgment are used along with elasticities. OCTA uses a range of service elasticities depending on the extent of the service change, but within each category the analyst has some leeway. Table 40 shows guidelines for service elasticity factors. Choice of elasticity within a category can be based on knowledge of the route, reason for the service change (e.g., to add service at a major activity center), other market con- siderations, or time of day. Interestingly, OCTA used its long-range model to estimate service elasticities and found an elasticity of 0.56 for peak-period headway changes. Then, by holding service levels constant, it estimated the elasticity of ridership with respect to demographic changes at 0.19. OCTA uses an extensive array of input data, but does not currently use APC data. The Operations Department has questioned the reliability of APC data; therefore, OCTA con- tinues to rely on ridecheck and especially farebox data. The automatic vehicle location (AVL) system supplies GPS coor- dinates with every farebox transaction, thus greatly enhanc- ing the reliability of farebox data (good at the bus stop level, very accurate at the TAZ level). This has been the primary impact of technology on ridership forecasting. One advantage TABLE 39 OCTA FORECASTING METHODS Purpose Method Time Frame Geography Budgeting Trend line, group consensus Annual Countywide New starts (bus/rail) Traditional four -step Up to 20+ years Regionwide Short-term service planning Service elasticities Up to 5 years Route specific Special purpose/commuter rail Apply specific mode choice components Vari es Route specific with broader service area Special purpose/paratransit Time series regression Vari es Countywide Percent Increase in Service Level (buses/hour) Recommended Elasticity Factor 20% or less Examples: 30 min to 25 min 15 min to 12 min +0.50 to +0.70 More than 20% to 50% Examples: 60 min to 45 min 45 min to 30 min +0.50 to +0.75 More than 50% to 100% Examples: 60 min to 30 min 30 min to 15 min +0.75 to +0.90 TABLE 40 OCTA GUIDELINES FOR SERVICE ELASTICITY FACTORS
27 of relying on farebox data is that it is available every day from every bus. Getting to the point where farebox data are usable not only at the route level but also at the stop level has been the primary objective for many years. OCTA is also explor- ing the development of origin/destination trip tables through the use of the APC data. OCTA is satisfied with the accuracy of its input data, with the caveats that improvements in the reliability of APC data through field verification and calibration would be useful, and that demographic data are not always avail- able at the desired geographic scale. More accurate APC ridership data would go a long way toward ensuring that an optimal amount of data is available. Constructing an origin/ destination trip table with APC data would be a significant improvement over use of on-board surveys, which are difficult and expensive. OCTA would like to see several improvements in its rid- ership forecasting methods, including the availability and accuracy of input data at the appropriate scale, less time- intensive methodologies, inclusion of more predictive vari- ables, simplification of procedures, enhanced accuracy, and flexibility to address a wide variety of situations. The ideal next step would be to develop an automated methodology for short-range ridership forecasting. Also, the long-range model predicts ridership for 2030; however, demographic variables are available in 5-year increments. A spreadsheet model that could predict interim year ridership (say at 5-year intervals) would complement and not compete with the long-range model, which requires too much work to generate interim year forecasts. Ridership forecasts would be developed under the scenar- ios included in the survey as follows: ⢠Half-mile rerouting of an existing route to serve a new shopping center: (1) determine interest level based on public comments, (2) look at similar shopping centers, (3) analyze whether this change would affect existing customers, and (4) consider improved transfer opportu- nities and connections. ⢠Extension of an existing route for one mile to serve a new residential development: estimate the additional revenue vehicle-hours required and multiply by the minimum productivity standard to project the ridership needed to meet the minimum productivity standard. ⢠Change in headway from 12 to 10 min during peak hours: multiply the peak-hour change in service hours by the cur- rent productivity of the route by the appropriate elasticity from Table 40 (within the range of +0.5 to +0.7). ⢠Implementation of a new crosstown route: analyze rid- ership, productivity, and transfer points of similar routes with a crosstown alignment. Use the long-range model to analyze further. ⢠Implementation of a new mode such as BRT: use the mode choice model. ⢠Prediction of next yearâs ridership as part of the budget process: analyze past and current ridership trends; use the annual ridership goal as defined by the Authority. The operations planning department tracks ridership compared with the goal, and will send up red flags to senior management and other departments when the trends do not match the goal. ⢠A 10-year ridership forecast as part of a long-range plan: use a combination of the mode choice model and trend line analysis. The one improvement to forecasting methodology would be to develop a more automated approach. This would involve the use of new technologies and tools, and would result in fore- casts based on a choice of methodology from a wide variety of proven and accurate methods that best fits the goal. OCTA reports two main lessons learned from its experi- ence: ⢠Use experience and results from the past to justify rid- ership forecasts. ⢠Carefully review mode choice model results with those obtained by peers and in other corridors and to elasticity-based forecasts. OCTA sees ridership forecasting methodologies adding value in three areas: ⢠Budget: good forecasts provide more accurate informa- tion regarding ridership and revenue for budgeting pur- poses. ⢠Service planning: forecasts help to prioritize potential service improvements by quantifying the benefits (increased ridership) of each improvement. ⢠Long-range planning: forecasts also quantify benefits attributable to transit in the long-range model, includ- ing increased ridership, decreased vehicle-miles trav- eled, and net reduction in travel delay. This case study indicates that GIS programs, formal mod- eling efforts, use of elasticities, and professional judgment can together provide a menu of ridership forecasting method- ologies for use as appropriate. The various departments that require ridership forecasts are comfortable with the method- ologies and confident in the results. Additional work is ongo- ing to enhance accuracy and simplify the use of these methodologies; however, OCTA has achieved a high level of confidence in its ridership forecasts in a wide variety of situations. TRIMET (PORTLAND, OREGON) TriMet prepares ridership forecasts for virtually every ser- vice change. There is no specific threshold triggering the need for a ridership forecast. Forecasts are used internally
and distributed to interest groups and stakeholders in response to service requests, as needed. One employee han- dles the ridership forecasting for bus changes. Light rail is evaluated using the four-step travel model. The Operations Planning and Planning Departments pre- pare ridership forecasts related to short-term service changes. Ridership and demographic data, including population, employment, and retail employment are the primary inputs. APCs provide current ridership, supplemented as needed with ridecheck data. TriMet uses census data and origin/destination data gathered through on-board surveys. Employment data are available from the 2000 Metro Employment Database. TriMet first looks for similar routes and uses professional judgment to forecast ridership for most service changes. If there are no similar routes, it implements a two-step process using regression and service elasticities to predict ridership. The first step involves regression equations developed in- house for three different types of service and calibrated using TriMet routes: ⢠For regional routes: Ridership1 0.06704 * popula- tion 0.0018 * non-retail employment 0.02 * retail employment. ⢠For local routes: Ridership1 0.00984 * population 0.004 * non-retail employment 0.008 * retail employ- ment. ⢠For employer shuttles: Ridership1 0.01 * non-retail employment 0.0135 * retail employment. All population and employment values are calculated within one-quarter mile of the route using GIS. Specifically, all census blocks with a centroid within one-quarter mile of the bus route are included in the route buffer. Results of the regression model are for a âtypicalâ route; the second step of the model adjusts the regression-based forecast using service elasticities that vary based on the pro- posed level of service. The equation is in logarithmic form. For regional routes, the equation is: Ridership Exp (((LN (# daily trips) LN (62)) *Elasticity) LN (Ridership1)) For local routes and employer shuttles, the equation is: Ridership Exp (((LN (# daily trips) LN (36)) * Elasticity) LN (Ridership1)) A daily total of 62 trips for regional routes and 36 trips for other routes represent service at 30 min headways for a typ- ical service span. Service elasticities (Table 41) were taken from the Traveler Response to Transportation System Changes study (10) and calibrated using TriMet data. These service elasticities are also used to forecast ridership based on all service changes involving changes in headway only. 28 TriMet calibrated each of the three models (for regional, local, and employer shuttle routes) using data from 12 regional routes, 12 local routes, and 5 employer shuttles. Cal- ibration minimized the differences between predicted and actual ridership for the group of routes as a whole. Elasticity factors of 1.0 were used for very frequent service outside the scope of changes shown in Table 41 and for local service and employer shuttles, based on the calibration efforts and because this high number of trips is usually associated with an increase in the span of service. TriMet is satisfied with the reliability of ridership data collected through its APC/AVL system. Numerous samples are obtained for each trip and the data have proven to be both detailed and accurate down to the trip and stop levels. Census data and origin/destination data are not quite so reli- able. Census data becomes dated relatively quickly. Origin/ destination data do not provide a large enough sample to work with below the route level. Technology has had a significant effect on ridership fore- casting. The APC/AVL systems have greatly improved the accuracy and reliability of ridership data. GIS has allowed TriMet to associate census data more accurately with routes and ridership. Ready availability of more detailed data, such as vehicle ownership and income, at the stop level would be welcome. The ridership models were designed to use population and employment because data for both is readily available at the census block level. TriMet is satisfied with its current fore- casting methods. Ridership forecasts could be developed under the scenar- ios included in the survey as follows: ⢠Half-mile rerouting of an existing route to serve a new shopping center: (1) identify existing service with com- parable headways to a shopping center that is similar in terms of land use, population, retail employment, and nonretail employment, and assume similar ridership; (2) if no similar service is identified, then enter popula- tion and employment data into the ridership model; and (3) consider added travel time for existing customers as a result of the deviation and, if deemed significant, apply a travel time elasticity from Pratt (10). Change in Headway Elasticity Factor New service or new time period +1.00 60 min to 15 min +0.58 60 min to 30 min +0.80 30 min to 15 min +0.73 20 to 15 min; 15 to 10 or 12 min +0.20 12 to 10 min; 10 to 7.5 min; 7.5 to 5 min +0.10 TABLE 41 SERVICE ELASTICITIES USED BY TRIMET IN ITSâ RIDERSHIP FORECASTING MODEL
29 ⢠Extension of an existing route for one mile to serve a new residential development: (1) look at comparable existing service and (2) if no similar service is identi- fied, apply the ridership model. ⢠Change in headway from 12 to 10 min during peak hours: apply a headway elasticity of 0.1 (see Table 41). ⢠Implementation of a new crosstown route: (1) look at comparable existing service and (2) if no similar service is identified, apply the ridership model. ⢠Implementation of a new mode such as BRT: the MPO would use its travel model to forecast ridership. ⢠Prediction of next yearâs ridership as part of the budget process: generally, TriMet does not forecast next yearâs ridership. ⢠A 10-year ridership forecast as part of a long-range plan: no experience with 10-year forecasts. TriMet lists the following lessons learned from its experience: ⢠Forecast models from external sources that the agency has experimented with in the past are complicated, require substantial staff time, are data intensive, and provide results that are often inferior to a simple analy- sis of similar routes. One regression model, for example, relied heavily on service levels as an indepen- dent variable. The projections suggested unreasonable ridership response to potential service improvements in low-density areas. ⢠Using population, retail employment, and nonretail employment as the independent variables in a ridership model results in accurate estimates. Other variables such as vehicle ownership and income do not provide enough improvement in accuracy to warrant the time and difficulty in acquiring and compiling the data at the appropriate scale. ⢠If sufficient data are available, derive elasticities from local experience, not industry-wide averages. The value of ridership forecasting for TriMet is that it provides a sound basis for making decisions. In most cases, ridership is the bottom line in the evaluation of existing and proposed service. Ridership forecasts aid TriMet in making an informed choice among competing alternatives. This case study provides an example of a ridership fore- casting model in use at a transit agency. It is noteworthy that TriMetâs first choice of methodology for incremental service changes is similar-route analysis, but the model is useful in addressing unique situations. TriMet also relies heavily on service headway elasticities to assess the impact of changes in frequency. TriMet believes that its model and approach could be used at other transit agencies, once calibrated with that agencyâs ridership data.
