Toolkit for Estimating Demand for Rural Intercity Bus Services (2011)

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.

59 Conclusions The purpose of the project was to develop a tool or tools to allow planners, operators, and analysts to estimate the potential demand for rural intercity bus services, and that has been accomplished. However, as in all research efforts, there are limitations to the results and potential opportunities for additional research. This chapter presents a discussion of what has been accomplished, its limitations, and directions for further research. In terms of accomplishments, the initial step in the develop- ment of these tools involved a literature review and analysis of previous demand estimation techniques. Additional key steps involved collecting data on the ridership on existing rural intercity bus routes and then classifying (or qualifying) these routes in a way that would assist in the development of the toolkit. Finally, the data was used to develop two main tech- niques for estimating demand, and these were packaged into a user-friendly CD that includes all needed data, examples, comparable route data, and adjustment techniques. Accomplishments The data collection and classification steps consumed a disproportionate amount of the overall project. In the end the study team collected ridership data on essentially all of the Section 5311(f)–funded rural intercity routes that had been operated for a year during a period of the last three fiscal years. This process was fairly arduous, because each of the states had to be contacted to determine if they had provided operating grants, if they had ridership data, or if they would provide contacts for the operators to collect the data. As this period was prior to the full implementation of the rural elements of the NTD, many of the states had only limited ridership information. In addition, for the ridership to be use- able to calibrate any type of demand model, the other charac- teristics of the service needed to be included, such as fare levels, frequencies, route length, connectivity with other modes, etc. Obtaining this information involved Internet research and often contact with the operators. In addition, extensive work was required to determine the populations served by the routes and whether the routes provided access to potential key traffic generators such as universities, correctional facilities, major medical centers, and commercial airports. The classification effort also proved to be more problem- atic than originally anticipated. The interim report to the study advisory panel proposed three categories, basically dividing the list of services according to the type of operator (because this involved key differences in service characteristics). Panel comments led to a major review of the services (includ- ing collecting additional data) to focus on routes that could more clearly be defined as intercity in nature, rather than services with long routes that were more clearly regional transit service or commuter services. The final classification involved looking at service characteristics and connectivity with the national intercity bus network. This classification step in effect became a qualification step, qualifying a route for inclusion in the calibration data set. In the end, the original data set of 139 routes was reduced to 58 when non-intercity routes were removed—either based on this more restricted definition or removed as an outlier. Using the revised data set, efforts turned to developing the toolkit. After some difficulties trying to estimate a regres- sion model, the study team succeeded in developing two approaches to estimate demand. One is a regression model, calibrated using the 58-route data set, and the other uses rural intercity long-distance trip rates from the NHTS, which are then adjusted using a factor estimated from the 58-route data set. Both of these techniques are more accurate for current rural intercity bus services than the demand models estimated for NCHRP in 1980. They represent a pragmatic approach that makes use of available data to produce initial estimates of potential ridership for new rural services. The regression C H A P T E R 7 Conclusions

model has the correct signs (e.g., ridership increases with a higher population base, etc.) and is plausible given general knowledge about travel behavior. It reflects higher ridership for intermodal connectivity to airports and for interlining. It utilizes population data as a key variable, but the impact of population is moderated by using the number of stops to calculate an average population per stop. This is plausible in that ridership is expected to be lower if the bus stops a lot to serve that population, which seems to reflect market prefer- ence for fewer stops. The use of the NHTS trip rate data also involves making maximum use of the available data. It provides ridership estimates based entirely on population served, but it is cali- brated in a sense through the selection of the mode choice factor to provide ridership estimates that most closely match the usage found in the data set. Regional variation is introduced through the use of regional trip rates. Finally, the 58-route data set was used to develop an adjustment factor that can be applied to the trip rate model results to further improve its results. The result is that the trip rate model and the regression model have comparable accuracy in terms of the percentage of time they will predict a ridership figure that is within a given percentage of the actual. However, they may not give the same answer. Finally, one goal of the toolkit was that it be easy to use. Given the degree to which both models depend on popula- tion data, it was determined that populations calculated with GIS would require users to have additional software and tech- nical skills that would take the toolkit out of the “easy to use” category. Initially the study team thought that GIS-developed population data for a 10- or 15-mile radius travel shed would improve the accuracy of the modeling effort. The difficulty of creating this data then led the study team to attempt calibration of the models using only municipal population of stops— but this was problematic because many stops serve multiple jurisdictions. Finally improved results came from using populations for urbanized areas (over 50,000 persons), urban clusters (2,500 to 50,000 persons), and Census-designated places (under 2,500 persons). These provide populations that are not necessarily limited to municipal boundaries. How- ever, because providing easy access to the populations in these three categories through hypertext links to Census websites would not be especially easy, the decision was made to include all this data on the same CD as the models. Thus the user could designate the stops on a potential route and at the same time obtain the populations and apply them in the models. With the data and the models on the CD, it seemed logical to include the instructions, qualifications, adjustments, exam- ples, and peer data all on the same disc and set the toolkit up so that it would provide links to this additional information at the appropriate places. Limitations The resulting models and the toolkit have some limitations. Regression Model Robustness. One limitation is that the regression model is not as robust as one might like. The adjusted R2 is 71, and a higher number would be better. The prediction and confi- dence intervals for a given estimate are fairly wide, implying that a given estimate could be much higher or lower than the estimated figure. Finally, the regression model has a negative intercept, which is more of a theoretical issue than a practical one, as it implies that if a route served no population at all, had no stops, did not interline, and did not serve any airports, the ridership would be negative. In fact it would be zero, and so the toolkit version does not allow a negative ridership. Adjustments Related to Populations. A second limitation, one that has been addressed to some extent in the toolkit, is the difficulty in making manual adjustments to the model estimates. Many sketch-planning techniques or modeling efforts call for such adjustments—they are sometimes called the “post-processing” step in the modeling process. However, as the example in the toolkit reveals, they require the analyst to do some additional research and computations, and it is not possible to put all the required data or formulas into the basic toolkit. Typically this post-processing step involves collecting data on the other intercity services at the points on the route and data on the institutional populations of colleges or universities, military bases, etc. The ratio of these populations to potential bus usage may vary significantly, and there is no published intercity bus trip generation rate that would link such special- ized populations to trips. The toolkit includes some plausible rules of thumb that could be applied to university popula- tions and some directions about contacting other institutions directly, but fundamentally the adjustments are going to require additional research and artful application of profes- sional judgment. Judgment Related to Airports. Similarly, professional judgment may be required in deciding whether a proposed route actually serves a commercial airport at a level or in a way that would result in the predicted ridership increase from the regression model. A proposed intercity bus route that serves a multimodal transit center with a taxi stand that allows a trip to the airport does in theory provide connectivity but is not at the same level or cost as an across-the-platform transfer from the intercity bus to a heavy rail line serving the airport. In addition, the level of service at the airport may make a significant difference. A small city airport with three or four 60

commuter flights per day is not going to generate the con- necting rural intercity ridership that might be generated by serving a major hub airport. Most of the examples in the data- base that involve airport service are related to a major hub level of service, because there is not much of a market for rural intercity service to airports with limited service. One useful approach with the toolkit is to run the model with airport service and then without, keeping all other variables the same. The difference between the two runs reflects the incremental ridership of an airport connection. Assuming an average ticket price allows the user to estimate the revenue associated with that ridership. If it is not sufficient to pay the additional costs (bus-hours or access fees) required to operate the airport service, it may affect the chosen service design. Judgment Related to Type of Carrier. Judgment may also be required in deciding whether a service is to be operated by an intercity bus carrier (ICB in the toolkit). One of the results of the classification exercise is an appreciation for the diffi- culty in defining “intercity” bus service—we know it when we see it, but setting exact guidance for classification can be difficult. The analyst using the model may decide to define the proposed route so as to provide a meaningful connection to the national intercity bus network in terms of schedule or a common terminal, with commonly available information about the connection. Whether this connection is sufficient to generate the additional ridership that the model estimates for ICB oper- ators in the absence of a formal interline agreement through the NBTA will require the application of judgment as well. A formal interline agreement between the rural intercity operator and an NBTA member may not make business sense for other reasons, but the rural route might function as a connecting service if these other service parameters (common station, schedule coordination, and information) are met at a high level. Other Limitations. Finally, the two most significant deficiencies of the model follow: • It is not sensitive to changes in fares or frequency. • It does not reflect ridership that might arise from places beyond those served by the route as a result of filling a gap in the network. The reason that variables for frequency and fare per mile were not significant enough to include in the regression model is likely because there is not a lot of variation in either variable among rural intercity routes. Typical fares are simi- lar for most intercity bus routes, and the typical frequency is one round-trip per day (or less than daily). Routes or services originally in the database with either a low fare per mile (often a long route with a flat fare) or high frequencies generally had high ridership as well; these routes were either eliminated as not being intercity or as outliers in terms of demand. Because the model is largely driven by population, it may project a very low demand for a route that does not provide service to many persons but provides a key link in the network that would have significant overhead or through traffic. Only a network model of a region or a nation could include this factor. This factor is included as a possible adjustment to the demand estimate, requiring either a call to the connecting car- riers to obtain data on how much traffic they might be feeding onto the route or other subjective estimates. Trip Rate Model Similarly, the trip rate model has some limitations. Like the regression model, it is driven by population, and so it will not predict ridership that comes from other places on the net- work. The trip rate model has no sensitivity to trip length, the number of stops, fares, frequency, or time of day. Ridership is strictly a function of population, without any other factors included (except region of the country). In addition, newer trip rate data may be available within a year or two. The population data used is from the 2000 Census and the NHTS is currently being updated; therefore, the model could probably use a recalibration when the more detailed American Community Survey Census data is avail- able for smaller places and the trip rates from the 2009 NHTS are available. Rural populations tend not to change dramat- ically, but it is possible that such changes would affect long- distance travel. Finally, one other limitation is that the mode-split infor- mation from the NHTS and previous surveys is not terribly explicit. For example, trips are not specified as being by airport limousine or intercity bus—scheduled or charter. The study team dealt with this issue by calibrating ridership predictions against sample routes, but the technique would be more use- ful with more explicit mode-of-travel data. Ideas for Future Research Additional Data One general direction for future research that is often put forward is a call for additional data to provide for better models by having more cases. In this case it is possible that in a few years improved ridership data for Section 5311(f ) operating projects will be available from the Rural NTD. Even if the data in the NTD is limited to operator and ridership, the existence of the requirement will mean that the states have to collect more information from their subgrantees, which 61

should result in better quality data. However, for the period of the past several years, this project has probably gathered ridership and service information for routes that have had enough operating time to provide annual data. Additional data on unsubsidized rural routes could likely be obtained, but a review of Greyhound routes (searching for “rural” routes) suggests that many routes with non-urbanized stops either receive Section 5311(f) funding already or serve mostly stops with over 50,000 persons in the urbanized area. Also, many of these routes are alternative schedules between endpoints served by express services, complicating the designation of them as rural. Other carriers operate unsubsidized routes in rural areas, but it is likely that this will decline over time Stop-Level Models A type of model that the study team had anticipated being able to include in the toolkit is a stop model—potentially a regression or a set of factors that would enable a prediction of the intercity bus ridership that might be generated in a particular town as a result of its population, demographic characteristics, and the presence of key generators. Resources limited this element. The study team obtained stop-by-stop ridership from the Michigan Department of Transporta- tion for its subsidized routes, and Jefferson Lines provided schedule-by-schedule ridership for each stop. Reassembling the Jefferson Lines data to provide totals for all ridership at a particular stop was time consuming, as the study team learned by developing this information for Minnesota. Even more time consuming is developing the demographic information for each stop, and then including the key potential generators and their characteristics. Given the existence of ridership data for only two states (without a significant additional investment in obtaining and expanding the data), development of a stop-level model was not pursued in this project, but it could be the basis for fur- ther research. Perhaps such an effort could be combined with additional efforts on overall rural public transit demand, which would use much of the same demographic information. Fare and Frequency Impacts The rural intercity demand models developed for this effort do not reflect the impacts of differences in fares or frequencies. In this sample of rural routes, there was not a great deal of variance, so further analysis would require work- ing with carriers to identify factors they look at when adding frequency. Is it driven by loadings—does a carrier add a trip when one scheduled trip gets full? Or will frequency drive increased ridership? New entrant carriers seem to start with a low “policy” frequency—one round-trip per day, or one morning and one afternoon/evening schedule. Additional research to address these issues would require ridership data for routes with greater variation in both fares and frequencies— data would be required on ridership in corridors connecting urbanized areas and perhaps on ridership for different carri- ers who may have different fare levels. These impacts may be more of an issue for the carriers that are trying to remain competitive in the deregulated industry than for public sec- tor planners developing rural feeders, as they are likely to be providing minimal frequencies. Terminals and Parking There is some apparent evidence from the acceptance of new “curbside” services that riders do not value bus terminal facilities or may in fact avoid them. These new services also are designed to have pickup locations close to major transit hubs and parking. Additional research on the ridership impacts of proximity to other modes, on presence or absence of ter- minal facilities, and on parking may be warranted as transit systems decide on intermodal terminal designs, park-and-ride facilities, and access policies. This type of research might well involve rider surveys rather than models of traveler behavior calibrated on ridership statistics. Intercity Service Planning and Procedures The effort in this project to provide enough background for an analyst to use the demand toolkit and to describe the use of the results suggests that there is a need for a more com- prehensive overview of an intercity planning process. This overview could include needs identification/gap analysis, consultation activities, service design (many factors need to be considered in addition to potential demand—connectivity, airports, road/terrain, hours of service, accessibility, etc.), estimating revenue, and building budgets—not just demand estimation. Section 5311(f) program requirements, policies, and procedures could also usefully be included in such a guide. National Network Model Although a network model was beyond the resources or goals of this project, the inability of these route-level model techniques to account for ridership arising from places beyond the route in question (except for the impact of operation by an intercity bus carrier) suggests that there is a need for a net- work demand model. Such a model would need to be built at least on a regional level but ideally would be a national model and, again, ideally would include intercity passenger rail ser- vice and potentially long-distance providers of airport ground transportation. Such a model could be used to develop rider- ship estimates for particular links, including those serving rural areas. 62

However its major use would be as a tool for examining policies and investments that are sure to come if current efforts to build high-speed intercity passenger rail are successful. These corridors are likely to be on routes that are currently served by unsubsidized intercity bus carriers, and it is likely that these firms will either have to change their service patterns in some way to complement the rail services or drop the services. In addition, there are many locations that will not support intercity rail service investment, and bus services linking these areas to the rail will need to be developed. All of these changes should be considered in the development of intercity passenger transportation policy and investment plans, yet there is no tool to consider the network implications for bus or rail. Such a network model would require intercity bus carrier cooperation in providing data to calibrate the model, particularly for trip generation in urbanized areas. 63