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.

10 The study team initiated telephone interviews with industry association staff to identify existing information about demand and/or information about rural intercity providers that were included in the subsequent survey efforts. The American Bus Association provided data on the state program contacts and the general status of rural intercity bus programs, which were used as a resource for contacting state program staff to obtain lists of Section 5311(f) operators that were contacted in sub- sequent tasks. Greyhound Lines suggested several particular rural intercity projects that should be contacted for data. However, there has not been the discovery of any particular model or technique for rural intercity routes that is not described in the literature review. All of the rural intercity services described in the interviews have been included in the data collection effort, and the operators have been contacted for data on ridership and service characteristics. Literature Review: Approaches to Estimating Rural Intercity Bus Demand As indicated previously, there has been a number of efforts over the past 30 years to develop demand models for rural and intercity services. However, in reality there are few existing planning tools that are reliable. In part this has been because of the reluctance of carriers to furnish data on actual ridership or revenue to researchers to allow the calibration of models. It is also because the types of services have changed (away from rural routes operated by intercity bus companies toward services provided by rural transit operators with varying degrees of connectivity to remaining intercity bus services) and because some of the models or techniques were calibrated long ago when the overall rate of intercity bus ridership was higher. There are many different questions that a bus company, a state department of transportation, or a rural transit provider would like to be able to answer, including questions about the potential financial impact of new services, and the advisability of reducing services or shifting them to a public provider. Different questions require different tools, and so there are several possible tools, all of which need development. All of the concepts suggested here would require revenue or rider- ship data (which would have to be provided by the operators of the service) to calibrate, and all address the issue of demand, which translates into ridership and revenue. Separate cost models would be needed to determine overall feasibility of service to a point or on a route or network. Discussed in the following sections are approaches that have been used in the past, along with some discussion of their potential utility. Per Capita Intercity Trip Generation Rate This approach uses available data to estimate the average number of intercity trips per year per capita for a given town, or an average based on regional, statewide, or national data. The per capita rate is then multiplied by the population of the place potentially being served to provide an annual estimated ridership. For a route, this technique can be applied to all the places on the potential route, and the results summed. This approach has been used in intercity bus studies in Nebraska, Iowa, and Minnesota, among other places. One significant difficulty is determining the appropriate rate. One can use national data to estimate a national intercity bus trip rate (total regular-route ridership divided by total U.S. population), but there is even a problem in developing that figure because of the limited data on national intercity bus ridership on scheduled services. Also, use a of national rate includes all city sizes, and it masks any regional or demographic differences. In some cases regional or statewide data has been obtained or developed; this data would provide a better trip rate factor. Again, use of this approach is insensitive to demo- graphic differences, travel patterns, frequency, or price—it implies that the characteristics of the potential service are comparable to those of the existing intercity bus service in the locations where the trip rate data was obtained. In the cases C H A P T E R 2 Review of Rural Intercity Demand Methods and Rural Intercity Services

where this approach has been applied, the rates are believed to have been developed based on ridership on conventional, privately provided scheduled intercity service—which implies a relatively low frequency (in rural areas) and a particular fare level (around $0.10 per passenger-mile). Depending on the available data, one could imagine devel- oping trip rates for a wide variety of service types, including feeder services provided by rural public transit systems, at dif- ferent frequencies (three days a week, daily, twice a day, etc.), to connecting services with different frequencies, and at dif- ferent fare levels or arrangements (high fares typical of airport ground transportation providers, medium fares typical of in- tercity providers, low fares more typical of public providers; separate fares versus interline fares, cash vs. Internet pay- ment, etc.). Developing such trip rates would require a large number of cases to draw out patterns or create a model, and it would require a fair amount of data about each example. Use of Comparable Services Another approach that is often used to estimate demand, or validate estimates, for urban fixed-route service is to identify a route or service comparable to that under study in as many ways as possible, determine the usage of the existing route, and then apply that to the proposed service. Again, there is a need to be sure that the relevant parameters are included in the analysis, which means collecting a fair amount of data about the existing services. If one could collect data on a significant number of existing services, i.e., document the significant characteristics (route length, frequency, fare, ticketing arrange- ments, information availability, demographics, characteristics of the connecting services, etc.), one could provide a classifica- tion scheme that would allow the planner of a new service to identify comparable services and then project ridership based on the existing service. The study team did not identify any ex- amples of this approach, except the use of ridership data from the Sage Stage in Modoc County, California, to validate rider- ship estimates developed for proposed service from Gunni- son, Colorado, to Denver. The Sage Stage operates from Al- turas, California, to Reno, Nevada, and the populations served, route length, density, and frequency are similar to the pro- posed service in Colorado. This validation was done as part of a study for the Colorado Department of Transportation. Use of Historical Data A related approach involves efforts to use ridership data from previous services, particularly in cases where an intercity bus company has abandoned a route or service and efforts are underway to utilize Section 5311(f) to provide for replace- ment services. However, there are several significant issues. One is that in the past intercity bus companies did not typically collect ridership data by stop, but rather by revenue. Convert- ing revenue to ridership requires assumptions about average fares per boarding that may or may not be applicable in a given region. Second, if an intercity bus company abandoned the service, it may well have destroyed existing ridership and revenue records, or the firm itself may no longer be in existence. A third issue is that if one is trying to estimate the demand for a replacement service, it may well differ if the type of service is different—a public operator with different schedules, a required transfer at the connecting service point, lower fares, lack of interline fares, lack of information in the intercity bus information system, etc. Finally, it is likely that a replacement operator would attempt to schedule services that would serve multiple markets—for example, providing a full day at the destination city to allow for medical appointments or shopping—potentially expanding the market on the rural feeder service, but reducing the demand from persons making intercity trips who would face long waits for intercity con- nections (if the local service was not scheduled to facilitate connections). This approach would clearly work best where the new service was very similar to the previous service. Examples known to the study team include the ridership estimates developed by KFH Group for replacement bus serv- ices in rural Indiana to serve Warsaw and other rural points. Data was available for the ridership experienced by the previous operator before exiting the route, and this was combined with the use of modeling techniques to develop new estimates for ridership and revenue on Section 5311(f) replacement services. Similarly, data on the previous ridership on a route in Colorado between Grand Junction and Pueblo [serviced by Texas, New Mexico, & Oklahoma Coach, Inc. (a Greyhound subsidiary at that time)] was obtained and used as a basis for ridership estimates for several proposed service options on this corridor. Difficulties in using historic data include the possible unavail- ability of intercity bus company data on a stop-by-stop basis unless a particular study was done and the likelihood that previous ridership was a function of the fare and frequency levels of that service and so cannot be assumed for different types of service. Also, intercity bus company data often includes overhead (ridership originating from and destined to places beyond the endpoints of the route in question) or connecting ridership that might not be available to a replacement service. Finally, it is not clear that previous ridership will return once a service has been discontinued for some time. Demand Model for Boardings or Revenue at a Stop This model approach would utilize U.S. Census demographic and service (frequency, perhaps fare) data to estimate the total annual revenue and the number of persons boarding at a particular stop. Models of this type have been calibrated for 11

Macon–Brunswick service using Georgia data (1) and for the Bay Transit service area in Virginia (2) using Greyhound District 2 sales data for points under 20,000 in population. As a general tool, it was calibrated for use in rural areas (under 50,000), because these are areas that are potentially eligible for federal funding under Section 5311(f). The Georgia and Virginia efforts are a good initial step, but additional work with the function form of the regression is needed to get a better model—one with a reasonable positive intercept and a higher R-square. Calibrations for cities with higher populations would be needed if this model is intended to be the trip generation component of a generalized network demand model. Data needed to calibrate the model to predict intercity bus carrier sales can be obtained from Greyhound’s agency sales data, from the Census, and from intercity bus schedules. Because the Greyhound computerized revenue accounting system (TRIPS) did not collect numbers of tickets from small population stops, in the past surveys would have been required to calibrate the model if it were used to predict boardings. However, Greyhound is now making another system avail- able for rural stops, and it may be able to collect data on both revenue and boardings by stop—if that data can be made available to future study teams, it could assist in the develop- ment of this type of model. It may be necessary to develop separate calibrations for places on routes that are served by rural transit programs. Route-Level Demand Models If there are several points that might generate sufficient ridership, a route proposal might be developed for further analysis. Edward J. Kannel of Iowa State University developed a corridor model based on data from 11 rural corridors. (3) This model was used in planning by the Iowa Department of Transportation—a particular feature was its sensitivity to frequency of service. At about the same time, a series of rural route models were developed in the 1980s by Ecosometrics, Inc., (4) as part of an NCHRP project to develop a methodology for state planning of intercity bus service. That model is still used, though it has not been recalibrated, and there are now geographic information system (GIS) mapping tools that could aggregate data on population and its characteristics in units other than municipalities—for example, the 10-mile service areas around a stop. The ridership data used to calibrate the route models dates from before regulatory reform and so is likely to be higher than one might find now. Calibrating a route-level model requires carrier data on total ridership for a route, or route segments. Some care could be required in identifying the ridership on the route. Service data in terms of average fares and frequencies on the routes would also be needed, along with Census demo- graphic data. The basic approach of the earlier models could be used again. It would still have the flaw that a single route model offers no way to address the potential revenue of over- head traffic. Such a model would be useful for predicting the ridership and revenue on potential new routes, allowing consideration of particular corridors or regions. State agencies would find it useful to identify potential new service corridors or estimate funding requirements, and Greyhound (or other carriers) would find it useful in deciding whether to submit service proposals and in pricing potential services. Another type of service that could be addressed is the rural connector service, in which the demand for local rural transit service that feeds intercity service would be estimated. City-Pair Demand Model This model, or series of models, would allow the analyst to project the demand (number of tickets or revenue) for a particular city-pair. This is the model type developed by the Texas Transportation Institute (TTI) as part of a recent study performed under contract to Greyhound and the Texas Department of Transportation. (5) This study developed two regression models, one for large destination cities, and one for small destination cities. The small destination city model could be used to estimate the number of tickets sold between the two cities based on the travel time between the cities, the destination population, the origin population, the mileage distance between the cities, and the origin median age. The TTI study was calibrated using data only from places with a popu- lation greater than 15,000, so its use for more rural locations may be inappropriate—use of this type of model for smaller cities would require calibrating new models of this type with data from manual ticket counts (or waiting until the MAX system is implemented in rural agencies). Because this model is much more data intensive (to examine 10 city-pairs from a given origin requires getting Census and service data on 10 locations plus the origin city, rather than just the origin city), its best use would be in cases where the point model suggested that there was sufficient overall demand; then the city-pair model could be applied to determine which connections would offer the highest demand, and the routes could be designed to accommodate the high-demand city-pairs. In addition, the analyst would have to assemble estimates of city-pair demand on a particular route or service to estimate its ridership. Network Models A model of this type would follow the basic urban trans- portation modeling approach used over the past 50 years, but apply it to the intercity bus network. Several states have included intercity bus as a mode in their statewide multi- 12

modal travel demand models, but the focus is generally on city-to-city travel rather than rural-to-urban travel. A single- mode bus model would include the typical steps of trip gen- eration and trip distribution to a network but would not include mode split because it is one mode. With full development, such a model could be used by bus companies and state departments of transportation to evaluate the network impacts of adding links to the network, or bypassing congested stops. The conceptual approach was presented in 1993 by William Black of the University of Indiana at the Annual Meeting of the Transportation Research Board. (6) His presentation described a network demand model of Indiana. It included major destinations outside the state as nodes, to reflect network demand for trips crossing the state, or from the state to these other points. It used a basic gravity model formulation to assign the trips, following the gravity model theory that larger population destinations attract more trips. The weakness of the Black model was that there was no good method for estimating the overall number of trips gen- erated in a particular city; efforts were made to use a trip rate factor, but the trip rate chosen included charter and tour trips, which resulted in apparent high levels of demand for intercity bus service all over Indiana. Black did not have access to any actual ridership data, so his model could not be calibrated against actual experience. The lesson of this approach is that if actual ridership (revenue) and boardings by stop were to be made available, it is possible to develop a national intercity bus network model that could be useful for examining the impact of strategies to reroute service, add new links, or eliminate routes or links. The advantage of a national model is that larger changes in strategy could be tested, as well as incremental changes. Poten- tially, it could eventually be integrated with information on station capacity to allow modeling of strategies to manage station demand, or garage demand. The proposed point demand model described previously would be required as the initial step to allow the estimation of the number of intercity bus trips generated at each stop. Then the network modeling would describe the initial network in terms of the routes, frequencies, and travel times between stops. Trip distribution would then place the trips generated onto the links, resulting in overall ridership on each link. Additional research would be needed to sub-allocate link rider- ship to particular schedules, but that would be an ultimate goal, as it would allow the analyst to test express scheduling, etc. Typically such models applied to urban regions and states (usually for highway modeling) are calibrated to actual traf- fic counts, and Greyhound’s new management information processes would facilitate calibrating the network, something not possible for Black. A number of engineering firms special- ize in this kind of modeling, and it is possible that existing software used for sketch planning could be used to develop the model. The feasibility of this approach could be tested on a statewide or regional network model initially. While this discussion provided an overview of possible approaches to rural intercity bus demand estimation, the focus of this study was on the development of models or tools to help determine the ridership on proposed feeder routes, rather than a network model. At this point in time, federal policy suggests that the private sector, with minimal regulation and no subsidy (other than gas tax reductions), is responsible for providing the overall trunk intercity network, with oper- ating subsidies limited to the Section 5311(f ) program for rural intercity bus services—i.e., services linking places with a population under 50,000 with that trunk network. Conse- quently, there is a very limited need by states or local transit operators for a network model that would provide estimates of demand on links between urbanized areas. The projects evaluated by the model or tools developed in this project are routes linking several rural points to a connection with the trunk network. A number of studies of this type have been conducted and the following section presents some of the issues encountered in these previous efforts. Recent Examples of Efforts to Estimate Ridership on Rural Intercity Services Trip Rate Model for Washington State As part of its work for the Washington Department of Transportation (WSDOT), KFH Group sought to develop a tool for estimating potential rural intercity ridership from places that currently do not have intercity service connections. These places were all identified as having higher potential needs based on a statewide ranking using Census demo- graphic data regarding typical needs characteristics for tran- sit dependency. In most cases the likely level of rural intercity bus service would have been limited to a single round trip per day, and the general approach involved using a trip rate to estimate the potential intercity ridership for each service point on the rural route for which population data was available and summing the estimated ridership. No estimate was made for additional demand at connection points that have exist- ing additional intercity bus ridership. The development of the trip rate was limited by available data. In the past one approach has been to apply a global national intercity trip rate, developed by taking an estimate of the total national regular-route ridership and dividing by the national population. At this point in time, the national regular-route ridership is not very well defined, because the only official data source, the Federal Motor Carrier Safety Administration, obtains statistics from only a partial set of the Class I (the largest) intercity passenger carriers, and the 13

most recent data is from 2002. A decade ago KFH Group developed a rate of 0.125 intercity trips per capita per year in the absence of any other data, based on the general assumption that Greyhound’s ridership was half of the total U.S. intercity ridership. However more recently KFH Group was able to use Greyhound data to develop an estimate of 0.147 trips per capita for the population within a 10-mile radius of an intercity stop, using data for stops in the Pacific Northwest. This slightly higher per capita trip rate may or may not be accurate for small towns in Washington and Oregon, because it was developed from data collected at the larger points in which Greyhound had its TRIPS ticketing system, but it seemed logical to use data that was regional for analysis in that region, and to use a trip rate that was from more recent data than the early 1990s. Table 2-1 demonstrates how one can use trip rate data and populations estimated by GIS to develop rural route ridership by developing the ridership at each point served and then summing the results. The table also reflects the need for other variables for a more accurate estimate, as the actual ridership for four existing rural routes is included in the sixth column. For the Prosser–Yakima route, the estimated daily ridership was 97, and, based on reports to WSDOT, the actual ridership was 106. In this case, it should be noted that (1) the existing service has no connectivity with intercity bus service except that it drives past the Greyhound station and (2) it has no fare. The lack of a fare would lead to higher ridership than might be expected at typical intercity bus fares of $0.10-12 per mile. 14 Population Estimated Actual Ridership within 10-mile Annual Daily (Existing Routes Place Population Radius Demand Demand Only) Points on Proposed Route: Colville 4,599 8,543 1,256 3.44 Connell 1,615 3,540 520 1.42 Deer Park 1,185 16,347 2,403 6.58 Goldendale 1,863 5,501 809 2.22 Kettle Falls 1,578 4,309 633 1.73 Newport 671 4,135 608 1.67 Oroville 1,753 3,021 444 1.22 Kettle Falls Route: 4,292 11.76 n/a Points on Existing Routes: Prosser 609 14,410 2,118 6 Grandview 8,089 35,781 5,260 14 Sunnyside 15,282 106,629 15,674 43 Granger 1,112 20,594 3,027 8 Toppenish 9,545 28,989 4,261 12 Wapato 6,492 34,464 5,066 14 35,407 97 27,020* (106 daily) Omak-Wenatchee-Ellensburg Route Omak 2,589 11,942 1,755 5 Brewster 1,493 5,251 772 2 2,527 7 6,192** Walla Walla-Pasco Route Walla Walla 35,882 45,484 6,686 18 2,096*** Port Angeles-Seattle-SeaTac Route Port Angeles 18,919 27,958 4,110 11 Sequim 4,169 21,550 3,168 9 Port Townsend 6,178 21,193 3,115 9 29,266 70,701 10,393 28 7,182 *Four times quarterly data, zero fare. **Four times quarterly data. ***Four times initial 3 months of data. Table 2-1. Washington state trip rate model.

The second route, Omak–Wenatchee–Ellensburg, is op- erated by an intercity bus operator, with interline tickets and intercity bus fare levels, so one would expect the point estimates to be more accurate. However, in this case the estimated ridership of 2,527 is less than the 6,000 or so actual riders. In part this discrepancy could be a result of the service also stopping at a number of points that are smaller than the listed towns and that do not have a Census 10-mile population figure, so the actual population served is higher. The estimated ridership for the Walla Walla–Pasco route is under 6,700, and the actual (estimated from three months of data) was about 2,100 for the initial service provider on this route. In this case the service was newly initiated and there were marketing and service issues, and so one might expect the assessment of actual versus estimated to be premature. More recent service on this route, now known as the Grape Line, is estimated to have annual ridership of 5,000 for the first year. The final route, Port Angeles–Seattle–SeaTac Airport, has an estimated ridership of about 10,400, and an actual reported ridership of a little under 7,200. Again, fares may play a role, in that the fare levels on this service are about twice the typical intercity fare level, because this service goes to the airport. So, as can be seen, the most simple tool, a trip rate, can be used to develop point estimates and route-level estimates, but its explanatory power is very limited because it does not include any adjustment for fare levels, frequency, service area anomalies, connectivity, etc. Its advantage is the ease of use. Once a rate is developed, the only data needed is population data. Therefore, it may be that carrier and survey data (from rural operators providing intercity service) could be used to develop a basic rate, and then adjustment factors could be developed to improve the basic rate’s accuracy—perhaps adjusting the results up or down depending on frequency, fare level, etc. With enough data, the rates could even be developed on a regional basis. Revenue and Ridership Estimates Based on a Statistical Model of Point Demand A more complicated type of modeling effort involves the development of a regression model to take account of more variables than just the population of service points. As part of an effort to assess the feasibility of developing replacement service in a rural region of eastern Virginia, the KFH Group developed a statistical model to estimate the revenue at a given rural service point. Greyhound provided data showing passenger revenue for all Greyhound stops in eastern North Carolina, Virginia, and Maryland, allowing the development of a linear regression model that would directly estimate revenue as a function of attributes of the location. To increase the chances of having a model that would be useful in predicting the revenue for the small towns in this area, only towns with a population of less than 20,000 were selected. A total of 41 stops were chosen using this method. For each of the 41 locations, information thought to affect intercity bus ridership was collected, including the following: • The population within the town’s boundaries • The 10-mile population around the stop • The percentage of the population with income below the poverty level • The percentage of households that are rented • The percentage of the population over 60 years of age • The percentage of the population between 18 and 24 years of age • The frequency of bus service, measured in departures per week • The presence of a four-year residential college (dummy variable) • The presence of a major medical facility (more than 150 beds) • The presence of a military installation • The presence of a state- or federal-level correctional institution • The presence of a locally operated transit system The data collected was presented in a table in the report. A number of linear regression models were tested, and the final model included the population within a 10-mile radius, frequency of service, percentage of population below the poverty level, and presence of a medical facility as significant in determining bus revenue at a rural service point. All of these appear to be plausible explanatory variables. (This is not always the case—an early demand modeling effort in Iowa found the best fitting model explained the variation in ridership based on retail sales and the number of dentists and physicians.) The town population, percentage of youth, and the presence of local transit increased the model’s standard error and decreased the adjusted R-squared value, so they were omitted from the model—probably because of correlations between the variables. The statistics of the models were presented in an appendix to the project’s final report. The regression presented the follow- ing model: Where: % bel pov = Percentage of population below poverty level freq = Weekly bus frequency, counted by departures med = Presence of a major medical facility (> 150 beds) 10-mile pop = Total population within 10 miles of an exist- ing bus stop Annual revenue bel pov= − + ( ) + 105 171 447 664 142 , , % 8 517 1 48 10 freq med -mile pop ( )+ ( ) + ( ). 15

The model was run for each point to be served on the poten- tial routes previously developed, and then the point estimates were summed to provide a route estimate. This was done for three proposed routes. The results that the model produced for all three alternatives, with varying frequencies, are presented in Table 2-2. The variables included make sense as factors likely to affect intercity demand, and the results are as expected in that it appears that increased service means increased revenue. Such a model could be used to estimate the total revenue that might be expected if service were initiated at a new small town location. It could assist in determining if the potential sales would support a commission agent, and could be used to help attract an agent. The model could be applied at several points along a proposed new route to estimate total revenue— again, if the revenue appears to be above a particular threshold, it could be analyzed in greater detail, or used as a basis for determining feasibility once operating and capital costs are developed. In the Virginia study, one concern was that there were two potential operators—Greyhound or a rural transit operator. All the data used in the model was from a national intercity bus company—it was unclear whether the rural operator would have the same demand (it could be higher because of local recognition and the ability to serve regional trips on the same buses, or lower because residents would not perceive it as offering a valid connection to the national network). How- ever, this model would need further development to deal with issues such as the negative intercept and to improve its statis- tics. Also, as calibrated it is really based on data from a partic- ular region—more data might allow for additional regional calibrations or a national model. Finally, this approach could be used on two data pools—one from a national intercity bus company (if it will provide the data) and the other collected from rural transit operators providing rural intercity service— to allow estimates for either type of service. Other Issues Considered in Developing a Model One potential issue that was considered was the need for data to calibrate a model or tool. The review of previous modeling efforts noted that several of the efforts were severely limited by the lack of data from the intercity bus companies and others for whom the available data is confidential for business purposes (revenue per location, rather than rider- ship, for example). In this effort the study team worked with the industry to obtain some company data to develop the Toolkit, but the proprietary nature of ridership and revenue data at privately owned businesses limited this source of data. In addition, industry data on ridership or revenue from rural stops (serving populations under 50,000) or routes primarily serving such rural places was limited because the industry has shifted so much of its service to serve places that have greater populations. Because there is a need to develop a model or tool for rural services to be operated by intercity bus companies, there was a need to have data on those types of services so that the model or tool could be calibrated. At the same time, in this effort much of the key data was provided by rural transit providers that are operating services they characterize as rural intercity bus service, and one focus of the project was on understanding the characteristics of the service involved in terms of frequency, connectivity, ticketing, fare levels, etc. to determine the appro- priate classification of these services. A second key issue that the study team attempted to address in this effort was the need to develop tools that are sensitive to the potential combination of markets that may be served by a rural intercity service. While it is expected that a significant proportion of the potential ridership may be making “intercity” trips—either to the destination of the rural feeder or to other intercity services for travel beyond that point—it is not clear to what degree the transit provider should 16 Daily Round- Trips Estimated Boardings Intercity Service-Northern Route $95,661 2,126 Intercity Service-Northern Route $115,653 2,570 Intercity Service-Route 17 only $62,662 1,392 Intercity Service-Route 17 only $82,654 1,837 Bay Transit-Two Routes 1 $95,661 2,126 Bay Transit-Two Routes 2 $115,653 2,570 Bay Transit-Three Routes 1 $128,162 2,848 Bay Transit-Three Routes 2 $148,152 3,292 District Two Regression Model Estimated Revenue 1 2 1 2 Table 2-2. Demand estimates using the regression model developed for Virginia using Greyhound District Two data.

develop services to also serve other markets. Such other markets could include medical trips, connections to other intercity modes, regional shopping, or even work trips. Some of the more successful rural intercity feeders combine these markets by making multiple stops in the destination city—for example, the Olympic Bus Lines service in Washington State, which serves the major hospitals, other downtown destinations, the Amtrak station, and the airport as well as the Greyhound station. Some rural operators have argued that they should not be required to provide schedule connectivity with intercity bus carriers because so few of the rural intercity passengers are making that connection. Rather, they would develop services that meet the regional needs (perhaps a morning inbound and a late afternoon return) of the majority of the riders, with the intercity passengers left to wait for hours. This issue was considered in this study, in terms of the data classification effort, which then affected the tools developed. In the end, usable models could be developed only by focusing on services that are primarily intercity in nature. Services serving multiple markets had too much variance to retain in the models. The survey of agencies did not identify additional methods or studies. However, the study team performed a search of the TRIS database, which resulted in the identification of 17 addi- tional references regarding intercity bus demand. The most directly relevant reference was an article by D. L. Dean, from 1982 (7), which directly addressed the main issues of this project. That reference called for a new approach to modeling rural intercity bus demand, finding that city-pair models (as developed primarily for estimating intercity air and rail passenger demand) have excluded small urban places and rural stops from consideration and so are not very useful for intercity bus generally, much less rural intercity routes. It also found that rural transit demand models have not been very useful for this purpose either, as they focused on a local area without reference to the larger regional, statewide, or national networks or connections that are potentially part of rural intercity demand. Dean did call for the inclusion of level of service factors in the development of rural demand models, as well as populations and demographic factors. Much of the background discussion in this article reflects the pre-deregulation era when there was substantial rural and small town intercity service provided by private for-profit firms, when the carriers provided many extra sections to respond to demand peaks on particular schedules, and when regulatory agents collected substantial amounts of data on bus ridership. The points made regarding the limitations of city-pair, rural transit, and trip rate models continue to be valid, but this reference did not provide a recommended approach or model. Many of the other references address a bus as a mode in statewide multimodal travel demand models or national intercity multimodal demand or market share models or are complex statistical models of intercity bus market shares in particular corridors. In general, these tools did not address the points that would be served by rural routes and are very complex. Consequently, they were not easily used to test alternative routes or service levels on rural routes. 17