Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses (2022)

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.

85   C H A P T E R 1 0 Previous sections of this report presented analyses about the possible factors impacting transit ridership trends. While it is important to use data to understand how these factors contribute toward transit ridership declines, explaining the complex travel behavior choices an individual makes as well as capturing the inherent dynamism of the transport system is a more compli- cated process. Therefore, there is need for a tool that could be used to map individuals’ travel behavior and simultaneously handle the multimodal interactions among existing or proposed transport systems. This section describes the web-based transportation microsimulation platform CityCast, which uses a data-driven approach for quick but rigorous scenario planning. CityCast is a complemen- tary tool to travel-demand models, which are highly specialized tools that can be cumbersome to develop, calibrate, install, and use for some planning analyses. CityCast enables planners to look at both transit ridership and road traffic impacts of improvements and policies with the latest data. Through the use of passively collected data, the CityCast platform provides the ability to study and evaluate transit ridership trends for future scenarios, such as those being considered by public transit agencies. Figure 10-1 shows an example of an activity pattern for a simulated traveler in CityCast, while Figure 10-2 shows how the route and mode of trips are determined. This latter portion is important because it explicitly considers both the competition and complementarity (such as other modes serving as connections to/from transit stations) between transit and other modes. Such interactions provide the opportunity to specify various scenarios and study their likely impacts on transit ridership and on other modes. In the background, CityCast simulates travelers making mode and route choices using the open-source framework Multi-Agent Transport Simulation (MATSim), accessible at https:// www.matsim.org. 10.1 MATSim Overview MATSim is a transport simulation framework that is used to simulate large-scale scenarios. As its name suggests, central to the MATSim framework are its agents and their activity patterns (also called plans). Each agent represents a real-world person, and each person starts with a schedule of activities that includes times and locations covering an entire single day (see Fig- ure 10-1). The initial plans of each agent are created prior to MATSim, and in the CityCast eco- system this is done using a combination of third-party, passively collected data and survey-based data from the U.S. census, FHWA, and FTA. MATSim focuses on simulating the travel between these initial plans (see Figure 10-2) using coevolutionary principles, meaning that each agent adapts its travel behavior depending upon the travel of other agents and their likely impacts on the transport infrastructure. Future Strategy Evaluation

86 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses Figure 10-1. Demand-side components of CityCast platform. Figure 10-2. Supply-side components of CityCast platform.

Future Strategy Evaluation 87   The coevolutionary principles of MATSim are realized by simulating an entire day many times through. After each time through, each agent measures how well it accomplished the day’s activities given its travel choices (mode, route, congestion, crowding) and iterates the next time through using those travel choices that perform well. With each iteration of the simulation, MATSim measures each agent’s plan using a scoring function by mode that is similar to a utility function in the traditional travel demand–modeling paradigm. When optimizing travel choices through iterations, four dimensions are usually considered by MATSim: departure time, route taken, destination, and travel mode used. If agents’ scores remain constant over iterations, this indicates that the system has reached an equilibrium state. The travel choices observed in a simulated single day in this relaxed state can be analyzed to understand representative populations and how they use the transport network (either a real one or an imagined one). In our study, the simulation was carried out for a total of 50 iterations to reach a relaxed state. 10.2 Input Data CityCast runs MATSim inside its web-based ecosystem for users, but it also allows users to download MATSim-required files for running MATSim locally with finer-grain control. At a minimum, the MATSim simulation needs the following files: • Network file: describing the road and public transport networks in the form of links and nodes. Nodes are described by (X, Y) coordinate values, while links hold all the attribute values related to segments (e.g., length, capacity, number of lanes, speed, and allowed travel modes). • Travel demand file: describing the people and their daily activities. Each activity is specified by location (X, Y), purpose, and time. To evaluate the expected effectiveness of the proposed alternatives’ transit strategy, two addi- tional files were needed: • Transit schedule: contains detailed information related to lines, routes, stop locations, and public transport schedule. • Transit vehicles: defines “type” and number of vehicles, which the simulation could utilize to run across the routes as defined in the transit schedule file. This file describes various charac- teristics, like seating and standing capacity (number of passengers), maximum speed, and how many passengers can board or depart a vehicle per second. Network files representing the detailed road networks of the two case study cities (Atlanta, Georgia, and Oshkosh, Wisconsin) were downloaded from CityCast; the platform currently uses HERE Maps for network data but has also used OpenStreetMaps in the past. All relevant modes of transport (i.e., private cars, walking, and bus and rail transit services) were included. The service details from the transit services were obtained using GTFS feeds from the agencies. 10.3 Identified Cities Atlanta and Oshkosh were identified as the two case studies. These two cities were chosen based on the different population and transit system sizes. As shown in Figure 10-3, Atlanta is a large metropolitan area with a population of around 6 million in the MSA. The Oshkosh-Neenah MSA has around 170,000 population and is adjacent to the Appleton, Wisconsin, MSA, which has another 240,000 population. These differences allowed the researchers to understand how the strategies are interpreted by residents living in cities of different scales. Moreover, Atlanta differs from Oshkosh in that it has more than one public transport operator and multiple modes.

88 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses 10.4 Development of Scenarios The goal of these scenarios is to establish a general boundary on the effectiveness of certain strategies. The following scenarios were tested for each city: • Base scenario: This scenario represents 2019 conditions before COVID-19. • Low-income focus: This scenario considers the effect of improving bus service on those routes that serve the highest share of low-income riders, while decreasing service elsewhere. • High-ridership focus: This scenario considers the effect of increasing service on those bus routes with the most riders, while decreasing service elsewhere. • High-ridership focus with exclusive bus lanes: Building upon the high-ridership scenario, this scenario adds bus-only lanes to those same high-ridership routes to give them a travel time advantage. 10.5 Atlanta, Georgia Atlanta has 500,000 people living within the city and over 6 million in the metropolitan area. It is the political and economic capital of Georgia and host to the headquarters of major compa- nies including Coca-Cola, AT&T, Delta Airlines, and UPS. Several major universities are based in Atlanta, including Georgia Tech and Emory University. MARTA anchors the regional transit system, which is supplemented by Georgia Regional Transportation Authority express buses and additional suburban transit agencies. On an average weekday, MARTA’s north-south and east-west heavy rail lines carry about 200,000 passengers, with fixed-route bus service carrying about the same number of riders (see Figure 10-4). Source: Google Maps. Figure 10-3. Statistics about case study cities, Atlanta and Oshkosh.

Future Strategy Evaluation 89   The Atlanta scenarios test the potential effectiveness of various strategies in a large and very congested area. 10.5.1 Low-Income Focus The goal of this scenario is to understand how transit ridership might change if bus service were reoriented better to serve low-income riders. To approximate this, the research team identified the 20% of bus routes within each transport operator that carry the highest share of low-income households. “Low-income households” were defined as those with less than $20,000 in annual income. Then the frequency of service on these routes was doubled. To keep total operating cost of the transit agency constant, the daily services of the remaining 80% of the transit routes were uniformly reduced, such that the VRH did not change from the base scenario. None of the schedules or frequencies of any rail transit modes were changed. The results indicate that in this scenario, linked bus trips would increase 47% among low- income riders and 44% among the general population. The map in Figure 10-5 shows how this change would be distributed geographically. The ridership gains are observed largely west of I-85, with ridership losses east of I-85 but inside the perimeter (I-285). 10.5.2 High-Ridership Focus The goal of this scenario is to understand how transit ridership might change if bus service were reoriented to better serve the routes with the highest ridership levels. To approximate this, the research team identified the 20% of bus routes within each transport operator that carry the most passengers, then doubled the frequency of service on these routes. To keep total operating cost of the transit agency constant, the daily services of the remaining 80% of the transit routes were uniformly reduced, such that the VRH did not change from the base scenario. Again, neither the schedules nor the frequencies of any rail transit modes were changed. Figure 10-4. MARTA transit system.

Figure 10-5. Change in bus ridership by census tract for low-income focus in Atlanta.

Future Strategy Evaluation 91   The results indicate that in this scenario, linked bus trips would increase 70% among low- income riders and 88% among the general population. Somewhat surprisingly, this scenario increases ridership among low-income travelers more than the first scenario. This occurs because the highest ridership routes also carry many low-income riders, even though their share of low-income riders may be lower. The map in Figure 10-6 shows how this change would be distributed geographically. The ridership gains are observed to be concentrated north of I-20, with ridership losses on the near south side of Atlanta. 10.5.3 High-Ridership Focus with Exclusive Bus Lanes This scenario aims to understand the ridership effect of giving exclusive right-of-way to high- ridership bus routes. It builds upon the high-ridership focus scenario, in which the frequency of the 20% of bus routes with the highest ridership was doubled, within each operator, while the frequency of the remaining 80% was reduced. Those 20% of bus routes with the highest ridership were provided with dedicated road infrastructure in the form of exclusive bus lanes, resulting in faster travel times. The researchers assumed that those bus lanes were new capacity, such that the road capacity did not decrease. To be conservative with the results, they did not further increase the frequency to reflect that the buses would complete their routes faster. The results indicate that in this scenario, linked bus trips would increase 89% among low-income riders and 109% among the general population. The map in Figure 10-7 shows how this change would be distributed geographically. A pattern similar to the high-ridership focus scenario is observed but with ridership gains that are more widespread, such as areas on the south side of Atlanta that lose ridership in the scenario above but gain ridership here. 10.6 Oshkosh, Wisconsin Oshkosh is a city of 66,000 in Northeast Wisconsin on the shore of Lake Winnebago. The greater metropolitan area also includes Neenah and Appleton for a total population of 160,000. Major Oshkosh employers are involved in manufacturing specialty trucks, emergency vehicles, plastics, and chocolate. The area is a destination for fishing and boating. Go Transit operates 10 bus routes on 30-minute headways, including one that travels 13 miles north to Neenah. In 2019, these routes carried about 2,000 riders on an average weekday. Valley Transit operates bus service in the Fox River Valley, including Neenah and Appleton, but Valley Transit was not considered in this analysis. The map in Figure 10-8 shows the bus routes in Oshkosh, and the map in Figure 10-9 shows the ridership (boardings plus alightings) by census block group. The largest circle, indicating the most riders, is in downtown Oshkosh. The Oshkosh scenarios align with those described previously and test the potential effectiveness of those strategies for transit operators in small metropolitan areas. 10.6.1 Low-Income Focus In this scenario, the research team considered the effect of adjusting transit service to better serve low-income riders—those with an annual household income of less than $20,000. The 20% of bus routes with the highest share of low-income riders were identified, and the frequency on those routes was doubled. There are 10 routes in Oshkosh so this meant increasing the frequency of two of those routes, going from 30-minute headways to 15-minute headways. To offset the cost of these service increases, the frequency of the remaining routes was decreased such that the total VRH remained constant. For this scenario, linked bus trips would decrease by 3% among low-income riders and 1% among the general population. The map in Figure 10-9 shows how this change would be

Figure 10-6. Change in bus ridership by census tract for high-ridership focus in Atlanta.

Figure 10-7. Change in bus ridership by census tract for high-ridership focus with exclusive bus lanes in Atlanta.

94 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses Figure 10-8. Go Transit routes. distributed geographically. For the routes on which frequency was increased, it is observed that the ridership for low-income riders and general population jumped to 110% and 120%, respec- tively. However, the overall marginal reduction is the resultant of the decrease in daily frequency on the remaining routes. 10.6.2 High-Ridership Focus This scenario considers the effect of increasing service on the highest-ridership bus routes in Oshkosh. The frequency on the two bus routes with the most ridership was doubled, and the frequency on the remaining routes was decreased to maintain an approximately constant VRM. In this scenario, linked bus trips were found to increase 39% among low-income riders and 30% among the general population. The map in Figure 10-10 shows how this change would be

Figure 10-9. Change in bus ridership by census block group for low-income focus in Oshkosh.

Figure 10-10. Change in bus ridership by census block group for high-ridership focus in Oshkosh.

Future Strategy Evaluation 97   distributed geographically. Again, the high-ridership focus actually increases ridership more among low-income riders than the low-income focus. This occurs because the specific bus routes improved in this scenario serve a large number of low-income riders. The research team also observed a geo- graphic difference—the ridership gains in this scenario were more concentrated north of down- town, whereas in the low-income scenario, they were concentrated west of downtown. 10.6.3 High-Ridership Focus with Exclusive Bus Lanes This scenario considered the effect of giving exclusive right-of-way to the two bus routes with the highest ridership in Oshkosh. These bus lanes were in addition to the service changes in the high-ridership focus that doubled the frequency of those routes while decreasing frequency elsewhere. The research team found that in this scenario, linked bus trips would increase 37% among low-income riders and 31% among the general population. The map in Figure 10-11 shows how this change would be distributed geographically. In contrast to Atlanta, which shows a notable benefit to adding exclusive bus lanes, the results of this scenario are very close to those for the high-ridership focus. This result occurs because there is much less traffic congestion in Oshkosh than in Atlanta, so the travel-time benefit of giving exclusive right-of-way to buses is less. 10.7 Discussion On the preceding pages, the results for three simulated scenarios each for Atlanta and Oshkosh were presented. These scenarios are intended to provide a broad idea of the types of strate- gies that might be effective at increasing transit ridership. Table 10-1 summarizes the ridership change for each of these scenarios. These scenarios are not intended to be a detailed evaluation of specific policies—it is expected that a local transit agency would conduct a more detailed evaluation of the effects and trade- offs of specific changes before implementing them. In particular, these scenarios are based on a simulation that assumes travelers will make the choice to drive or take transit based on the relative travel time of each. While this is a sensible assumption, real people may resist switching to transit for a variety of reasons that go beyond what can be captured in a simulation. There- fore, the results should be viewed as indicative of the direction and ordering of the expected outcomes, with uncertainty related to the magnitude. (For agencies that would like to conduct their own similar analysis using MATSim or CityCast, Technische Universität Berlin maintains a fairly updated tutorial set, which covers a step-by-step installation and usage guide for creating scenarios and testing policy cases, found at https://www.matsim.org/docs/tutorials/general.) In this application, CityCast was used to provide information on where people travel as input to the MATSim Scenarios. Further details on CityCast are available at https://citycast.io/. From these scenarios, the research team was able to draw the following conclusions: • There is potential to increase transit ridership without major budget increases by reallocating existing service. This outcome is reinforced by empirical findings from Chapter 3 where it was found that bus network redesigns increased ridership by 4.7%. • Serving high-ridership corridors and serving low-income travelers are not mutually exclu- sive goals. In fact, many low-income travelers are on the routes with the highest ridership, so the goals often align. However, the spatial difference in the ridership gains and losses in the Atlanta low-income versus high-ridership scenarios highlight that there remains a risk of underserving low-income and minority neighborhoods.

Figure 10-11. Change in bus ridership by census block group for high-ridership focus with exclusive bus lanes in Oshkosh.

Future Strategy Evaluation 99   • While all the strategies tested showed positive ridership results in both locations, the most effective design depends on local conditions. Specifically, bus-only lanes offer higher benefits in more congested settings. Finally, it is worth considering that while this analysis was focused on ridership, ridership is not the only metric that matters when evaluating the value of public transit. Even when rider- ship is low, there is value in serving those travelers who have no alternative, and there is value in serving those travelers with short headways and quick travel times. Atlanta, GA Oshkosh, WI Scenario Total Riders Low-Income Riders Total Riders Low-Income Riders Low-Income Focus 44% 47% -1% -3% High-Ridership Focus 88% 70% 30% 39% High-Ridership Focus with Exclusive Bus Lanes 109% 89% 31% 37% Table 10-1. Change in bus ridership for each future scenario.