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.

17   As discussed in the previous chapter, a mix of factors is contributing to recent ridership trends, several of which will push ridership in competing directions. To separate the effects of each of these factors, this project conducted statistical analyses that correlate each factor with changes in transit ridership. The research team analyzed these factors in a two-phase, top-down approach that considered ridership changes first at the system level and then at the detailed route and stop levels. This let the team both consider the diversity of transit systems in the United States and take advantage of more detailed data assembled for specific cities. The analysis compared conditions from 2012, when bus ridership in the United States reached its post-Recession peak, to 2018. The multicity analysis presented here includes estimated statistical models of the annual change in total bus ridership and rail ridership across 209 metropolitan statistical areas (MSAs) in the United States. (The data sources used in this chapter are outlined in Appendix D of TCRP Web-Only Document 74.) The system-level analysis allowed researchers to test many of the factors listed previously and laid the groundwork for the testing of other factors and strategies in Phase 2. The variation across both time and space allowed for better statistical estimates of the sensitivity to these variables because they may change at different rates in different MSAs. The resulting models give the percent change in ridership that would result from a 1% change in each descriptive variable, a relationship known as elasticity. These relationships are broad, and there will always be some portion of the real-world change that models cannot capture. For example, one might expect that changes in where and when service is scheduled affect ridership beyond changes in total VRM. Such details are difficult to capture at the system level and are instead explored in more detail at the route or stop level. The research team reported the results with a category labeled “unknown factors,” which includes all the observed changes beyond what could be described by the models. Once these relationships were known, the researchers applied them to calculate the contribution of each factor to the change in bus and rail ridership for each MSA. This chapter examines bus and rail ridership trends to understand better the similarities and differences across groups of MSAs. Second, the results of a statistical analysis that establishes the sensitivity of transit ridership with respect to changes in both internal and external factors are reported. Third, each factor’s contribution to ridership change in each group is provided. Finally, conclusions are drawn about the overall reasons for transit ridership decline. 3.1 Transit Ridership Trends by Group Before analyzing the reasons for transit ridership change, the research team first examined the patterns of how it has changed. The results in this chapter are grouped into three clusters of MSAs based on transit annual operating expenses per capita, as defined by APTA. The New York region is excluded from the main analysis because NY is an outlier in its historically high levels C H A P T E R 3 Multicity Evaluation

18 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses of transit ridership, which account for 40% of U.S. transit ridership overall. NY has witnessed moderate ridership gains over the past decade. Without data from the NY region, the long-term national transit ridership is decreasing. The three clusters of MSAs are as follows: • The high operating expenses group (greater than $300 million annually) includes 19 MSAs with populations between 2 million and 13 million—such as Atlanta, Chicago, Philadelphia, and Houston—each with both bus and rail services. • The mid operating expenses group (between $30 million and $300 million annually) includes 64 MSAs ranging with populations between 200,000 and 4.6 million, such as Bakersfield, California; Denver, Colorado; Indianapolis, Indiana; and New Haven, Connecticut. All MSAs with mid operating expenses have bus service, and 12 of them also have rail service. • The low operating expenses group (below $30 million annually) includes 126 MSAs with popu- lations ranging from 80,000 to 1 million—such as Athens, Georgia; Bridgetown, New Jersey; Morristown, Tennessee; and Yuma, Arizona—each with only bus service. Appendix C of TCRP Web-Only Document 74 shows the full list of MSAs by operating expenses group and results for each MSA, while this report includes the overall results for each group. Figure 3-1 shows the percent change in bus ridership relative to 2012 for each group, according to data from NTD. Bus ridership in the high and mid operating expenses groups peaked in 2008, then declined in the wake of the Great Recession, while bus ridership in the low operating expenses group continued to grow through the Recession. All three groups increased again before peaking in 2012, then declining steeply from 2014 to 2018. In 2018, bus ridership in all three groups was about 15% lower than its 2012 peak. Transit ridership declines are not limited to buses. Figure 3-2 shows the change in rail rider- ship relative to 2012. In the high operating expenses group, rail ridership increased until 2014, then decreased, ending 6% below its 2014 peak and 3% lower than 2012. Rail ridership in the mid operating expenses group was more dynamic, with a higher peak in 2008 and a steeper post-Recession decline and recovery. Rail ridership growth in this group is driven in part by new Figure 3-1. Percent change in bus ridership from 2012.

Multicity Evaluation 19   or expanded rail systems in places such as Charlotte, North Carolina; Denver, Colorado; and Seattle, Washington. Rail ridership in the mid operating expenses group peaks in 2013, and by 2018, it is 10% lower than its peak and 6% lower than its 2012 level. Only a handful of MSAs in the low operating expenses group have rail service; these are excluded from the analysis. Recent transit ridership declines are broad-based—they occurred for both bus and rail and across large, medium, and small cities. The change is especially steep between 2014 and 2018. The predominant causes of ridership declines must also be broad-based and concentrated in those years. In order to understand what those causes might be, a statistical model was estimated to measure how sensitive transit ridership is to a range of factors. 3.2 Sensitivity of Transit Ridership to Different Factors The research team determined the sensitivity of transit ridership to changes in each variable using a fixed-effects panel regression model. The fixed-effects model estimates coefficients based on the changes within each MSA, rather than from differences between MSAs. The team used data on transit ridership and operating characteristics as reported in the NTD and supplemented it with census data, economic statistics, news reports, and other publicly available data. The detailed methodology and assumptions in the model are presented in Appendix D, which is available in TCRP Web-Only Document 74. In developing this model, over 100 different specifications were tested before the researchers arrived at this preferred one. A number of variables were consid- ered but excluded because they were shown to be insignificant (p-value > 0.05) or because a different specification produced a better fit or more defensible result. The specific variables used in the analysis are described as follows and grouped into six broad categories. In all cases when discussing change, this report refers to net changes, with the assumption that all other factors remain constant. Figure 3-2. Percent change in rail ridership from 2012.

20 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses 3.2.1 Service The quantity and quality of transit service provided affects transit ridership. This effect is captured using three measures: • VRM of service is a strong determinant of transit ridership. The results indicate that each 1% increase in bus VRM increases bus ridership by 0.45% and each 1% increase in rail VRM increases rail ridership by 0.66%. Rail ridership may be more elastic to changes in VRM because rail tends to attract more choice transit riders than bus. • Bus network restructures are associated with a 4.7% higher bus ridership over the six-year period, but the effect is not statistically significant. In recent years, transit operators such as those in Baltimore, Maryland; Columbus, Ohio; Houston, Texas; Jacksonville, Florida; and Tallahassee, Florida, have restructured their bus networks—changing routes and the service allocation—in an effort to better serve their passengers. The operators that made these changes saw, on average, a 4.7% bus ridership increase over and above the effect of any VRM increases. However, not enough agencies have completed such a restructure to make the results statisti- cally significant. • Major line closures for maintenance work can have an important effect on rail ridership. Safety incidents in 2015 and 2016 on the Washington Metro in the District of Columbia led to line closures and major maintenance work in the following years, with disruptions lasting from late 2015 to early 2018. Rail ridership in the Washington MSA was found to be 13% lower in the affected years (with half the effect in 2015 and 2018) than would otherwise be expected. This effect was marginally significant. The researchers tested a more comprehensive measure of reliability based on the mean distance between failures, but the reporting of failures to the NTD is inconsistent, so they could not detect a meaningful effect. The research team tested or considered several other measures of transit service. It was found that average transit speed is negatively correlated with transit ridership, probably because vehicles can travel faster if they do not have to stop to pick up and drop off passengers. The team could not have a widely available measure of on-time performance, nor was there a comprehensive measure of where the service is allocated within a region. 3.2.2 Fare Higher fares lead to lower transit ridership. • Increasing average bus fare by 1% decreases bus ridership by 0.57%, and increasing average rail fare by 1% decreases rail ridership by 0.35%. The average fare is calculated by taking the total inflation-adjusted fare revenue earned by the transit agency in a year per UPT. The different elasticities for bus versus rail fare may reflect different income mixes of the passengers. The research team could not test specific fare or pass programs at the system level, but these were tested in case studies, as described in later chapters. 3.2.3 Land Use Transit connects people to activities and jobs, so the number and location of both affects transit ridership. • Each 1% increase in population plus employment is associated with 0.22% more transit ridership. These effects are correlated with each other and could not be estimated separately, but when taken together, the effect is positive and significant. • Higher density leads to more transit ridership. The researchers considered the percent of the population and employment in a region that is within a transit-supportive density, defined

Multicity Evaluation 21   as more than 10 people or employees per acre. For each percentage point increase (such as from 10% to 11%) in population plus employment living in these denser areas, transit ridership becomes 0.4% higher. Working at a national level, the researchers could not compile data on the location and size of transit-oriented developments, nor could they compile other, more detailed data on the alloca- tion of land use within transit-supportive areas. 3.2.4 Gas Price Higher gas prices make driving more expensive and encourage riders to switch to riding transit. • Each percent increase in gas price accounts for a 0.14% increase in transit ridership. The research team measured this with data from the Energy Information Administration and adjusted the measure for inflation. 3.2.5 Household and Income Characteristics Among the many factors related to the characteristics of households, their income, and the work norms that may affect their transit ridership, the research team found three to be important: • With higher per capita income, people are less likely to ride transit. Several variables were tested to establish the relation between income and transit ridership. Although mean and median values of household-level income both display the expected correlation, the research team chose per capita median income in 2018 dollars because of the better fit of the model. Each 1% increase in median per capita income results in a 0.07% decrease in transit ridership. • Higher shares of zero-vehicle households in an MSA have a small positive effect on transit ridership. People from households without a car constitute an important market for transit riders. However, the share of zero-vehicle households has been relatively stable in recent years, so the results show that it explains little about the change in transit ridership over this period. The model indicates that a 1% increase in households owning zero vehicles would result in 0.2% more transit ridership, but this effect is not statistically significant. • For each additional percent of workers telecommuting, transit ridership decreases by 0.76%. This result is based on the journey-to-work mode shares reported in the American Community Survey. This result is particularly interesting going forward considering the large percent of population working from home during the COVID-19 pandemic. The research team tested the percent of the population living in poverty, the percent of the population born in a different country, and the percent of the population in different age groups and did not find significant effects. The team also tested the distribution of poverty as measured by the percentage of poor households living in areas with transit-supportive density but did not find a significant result. 3.2.6 New Competing Modes Over the past several years, several new modes of travel have entered or proliferated in urban areas, including ride-hailing, bike-sharing, and electric scooters. It is possible that these modes could complement public transit by serving as first-mile/last-mile connectors or by serving trips at times and locations not well served by transit. However, these new modes could also compete for the same riders that transit serves, especially if they are concentrated in the densest corridors and center cities where transit ridership is highest. The fact that these new modes started operating in different MSAs in different years provides a natural experiment to test their

22 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses effect—the panel-data models empirically measure whether MSAs with ride-hailing or other modes have higher or lower transit ridership than would be expected when controlling for all other factors in the model. This analysis reveals that: • When ride-hailing enters a market, bus ridership decreases. Ride-hailing—sometimes also known as app-taxi or e-taxi—services are offered by local drivers using personal vehicles via a transportation network company (TNC) or by traditional taxis via an app. The app connects the drivers with potential passengers for a pre-determined fee and transports the customer door-to-door exactly where they would like to go. When ride-hailing enters a market, its effect is not an immediate switch. Instead, the effect builds over time as ride-hailing companies are able to recruit more drivers and serve more passengers. In the largest MSAs (those in the high operating expenses group), bus ridership decreases by a net 1.9% per year after ride-hailing enters the market. For MSAs in the mid and low operating expenses groups, bus ridership decreases by 3.4% per year after ride-hailing enters the market. Both measures are highly sig- nificant statistically. The difference may relate to a greater transit resilience in bigger, denser cities. It is also important to note that ride-hailing companies entered smaller markets later, and the companies may therefore have more resources for a faster ramp-up period in those markets. Whether this trend continues or levels off remains to be seen and is complicated by reduced travel during the COVID-19 pandemic. • When ride-hailing enters a market, rail ridership in MSAs with medium-sized transit agen- cies decreases, but rail ridership in MSAs with a larger amount of transit remains resilient. Ride-hailing’s entry affects rail ridership differently than bus ridership. In medium-sized MSAs (those in the mid operating expenses group), rail ridership decreases by a net 2.2% for each year after ride-hailing’s entry, and the effect is statistically significant. However, for MSAs in the high operating expenses group, ride-hailing has a slightly positive (0.2% per year) but statistically insignificant effect on rail ridership. Ride-hailing may affect rail ridership less than bus ridership because rail offers a travel-time advantage that bypasses congestion, or because rail tends to serve longer trips that would be more expensive to serve end-to-end with ride-hailing. Rail ridership in bigger cities could be more resilient for those same reasons, or because those rail systems tend to be older and better integrated into the urban fabric. • Bike-sharing has a small but insignificant effect on transit ridership. When a bike-sharing system starts, transit ridership in that MSA is found to be 1% lower in subsequent years than otherwise expected, but this effect is not statistically significant. This effect does not build over time in the same way that the effect of ride-hailing grows. • When dockless electric scooters enter a market, transit ridership decreases. Transit ridership is found to be net 4% lower after e-scooters enter a market. However, e-scooters only started in the last year of the analysis period, so the researchers are less confident in this result than they are in the ride-hailing and bike-sharing findings, where there is a longer record to ana- lyze. Therefore, this topic will be explored further in the route-level analysis. The researchers tested several other specifications to understand how the effects might vary by location or mode. They found similar results as they varied the specification, with the results presented here based on the best overall model. 3.3 The Contribution of Each Factor to Changes in Transit Ridership The research team applied the sensitivities calculated above to calculate the total contribution of each of these factors to the change in transit ridership between 2012 and 2018. The coefficients for each variable in the estimation represent either the direct elasticity or the percentage point increase in transit ridership for each unit percent change or unit change in the factors. These coefficients were multiplied by the observed change in each factor to calculate that factor’s effect

Multicity Evaluation 23   on transit ridership. While competing factors may offset each other, this approach can be used to calculate the net effect of each factor. Applying this approach does not capture 100% of the observed ridership change, and any remaining difference between the modeled and observed ridership is labeled as “unexplained change.” These calculations were applied separately for each MSA and transit mode (bus versus rail), then the results were aggregated by group. The results by group (as described in Section 3.1) are shown in the following section, and Appendix C of TCRP Web-Only Document 74 shows the results for each MSA. 3.3.1 Contributions to Bus Ridership Change Table 3-1 shows the change in each factor and its contribution to bus ridership change between 2012 and 2018. The rows are grouped by the six categories of factors as described previously, with results for each variable and a subtotal for each category. The columns are specific to the Change in Average Values by Operating Expenses Group Ridership Effect by Operating Expenses Group Description High Mid Low High Mid Low Service VRM 4.2% 11.9% 9.0% 2.5% 4.7% 4.0% Network Restructure 0.03 0.03 0.0 0.1% 0.1% 0.0% Subtotal 2.6% 4.9% 4.0% Fare Average Fare (2018$) 0.0% 1.6% 17.8% -0.3% -0.3% -4.0% Subtotal -0.3% -0.3% -4.0% Land Use Population + Employment 6.3% 7.9% 5.8% 1.4% 1.7% 1.1% Percent of Population + Employment in Transit Supportive Density -0.2% -1.2% -1.9% 0.0% -0.2% -0.1% Subtotal 1.4% 1.5% 1.0% Gas Price Average Gas Price (2018$) -26.4% -28.8% -29.5% -3.4% -3.8% -3.9% Subtotal -3.4% -3.8% -3.9% Household and Income Characteristics Median Per Capita Income (2018$) 12.5% 9.5% 8.4% -0.8% -0.6% -0.6% Percent of Households with 0 Vehicles -8.7% -12.8% -4.8% -0.2% -0.2% -0.1% Percent Working at Home 22.7% 32.5% 35.1% -0.8% -1.0% -0.9% Subtotal -1.7% -1.8% -1.5% New Competing Modes Years Since Ride-Hail Start 5.68 3.86 3.26 -10.2% -11.8% -9.8% Bike-Share 0.79 0.74 0.54 -0.8% -0.8% -0.5% Electric Scooters 0.54 0.41 0.07 -1.9% -1.3% -0.3% Subtotal -12.9% -13.9% -10.5% Total Modeled Ridership -14.4% -13.4% -15.0% Total Observed Ridership -14.4% -15.8% -14.6% Unexplained Change 0.1% -2.4% 0.4% Table 3-1. Contributions to bus ridership change between 2012 and 2018.

24 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses groups, which shows how ridership changes differ by group. For example, between 2012 and 2018, bus VRM for MSAs in the high operating expenses group increased by 4.2% on average, resulting in 2.5% more bus ridership in that group. The changes may be higher or lower for individual MSAs, with the values reported here representing the total change across all MSAs within the group. For some factors, the absolute change in the variable is reported rather than the percent change. For example, the bike-sharing variable takes a value of 1 if bike-sharing is present and 0 if it is not. The value of 0.79 in the high operating expenses group indicates that 79% of MSAs in this group did not have bike-sharing in 2012 but did have bike-sharing in 2018. These results show that: • More service leads to higher ridership for all groups but to different extents. • Network restructures, while noticeable for individual MSAs, have a small overall effect when added to the remaining MSAs in the group. • Fares have only a small effect on ridership in the high and mid operating expenses groups; however, the average fare increased by 18% in the low operating expenses group, leading to 4% lower bus ridership. • All groups added population and employment, leading to higher ridership, but this change was partially offset by much of that growth occurring in low-density areas. • Gas prices were 26%–30% lower in 2018 than in 2012, leading to 3%–4% lower bus ridership. • Over this period, incomes increased, a smaller share of households owned zero vehicles, and more people worked from home. • The biggest portion of bus ridership loss is attributable to ride-hailing; bus ridership is 10%–12% lower in each group due to competition with ride-hailing. • Competition with bike-sharing and electric scooters leads to slightly lower ridership, although the bike-sharing effect is statistically insignificant. When applied in this way, the model does a good job of predicting the total ridership change for each group. The modeled and observed changes are within 1% for both the high and low operating expenses groups. For the mid operating expenses group, the model predicts a 13.4% ridership decrease, but ridership actually decreased by 15.8%. This group has a 2.4% ridership decrease that cannot be explained by the factors described here. Table 3-1 shows how the model can be applied to determine the contributions to ridership change for 2012 and 2018. It can also be applied to estimate the contributions to ridership change for any year for which the researchers have data. In Appendix E of TCRP Web-Only Document 74, the model is applied to the period from 2002 through 2012 to validate that it can reasonably predict the observed ridership change for a period that extends beyond the estimation period. In Figure 3-3, Figure 3-4, and Figure 3-5, the model is applied to each year from 2012 through 2018, and the effect of each factor on bus ridership is plotted. In these plots, the black line shows the observed ridership. Shaded red areas above the black line indicate the amount by which ridership is lower due to the changes in that factor relative to its 2012 value. Shaded green areas below the black line indicate the amount by which ridership is higher due to changes in that factor relative to its 2012 value. For example, Figure 3-3 shows that bus service increases in the high operating expenses group led ridership to be higher than it otherwise would have been. In each of the figures, bus service increases relative to 2012 levels led to increased bus ridership. Land use changes led to slightly more ridership in each group, while changes to household and income characteristics led to slightly lower ridership. Fare increases had a small effect in the high and mid operating expenses groups and a larger effect in the low operating expenses group. The drop in gas prices between 2014 and 2015 resulted in a decrease in bus ridership for each subsequent year. The biggest contributor to change in bus ridership was the emergence of new competing modes—dominantly ride-hailing. For each group, bus ridership would have been almost level without the losses due to these new modes.

Multicity Evaluation 25   Figure 3-3. Contributions to bus ridership change for high operating expenses group.

26 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses Figure 3-4. Contributions to bus ridership change for mid operating expenses group.

Multicity Evaluation 27   Figure 3-5. Contributions to bus ridership change for low operating expenses group.

28 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses 3.3.2 Contributions to Rail Ridership Change Table 3-2 shows the change in each factor and its contribution to rail ridership change between 2012 and 2018, in the same format as Table 3-1. The small number of MSAs in the low operating expenses group that have rail service were excluded from the analysis, as there are not enough MSAs to draw broad conclusions. The 2012–2018 period continues a broad investment in rail service in the United States, with VRM increasing by 12% in the high operating expenses group and by 23% in the mid operating expenses group where more light rail or streetcar systems were opened or expanded. From these service changes, 10% and 18% more ridership in those groups would be expected. Given the fixed nature of rail investment, there are no rail network restructures. However, maintenance of aging rail systems over time has become a major issue. While all systems are conducting maintenance, the more drastic closures Change in Average Values by Operating Expenses Group Ridership Effect by Operating Expenses Group Description High Mid High Mid Service VRM 11.8% 22.9% 10.0% 17.9% Major Maintenance Event 0.09 0.0 0.0% 0.0% Subtotal 10.0% 17.9% Fare Average Fare (2018$) 12.7% 7.4% -2.7% -0.9% Subtotal -2.7% -0.9% Land Use Population + Employment 6.0% 6.0% 1.4% 1.5% Percent of Population + Employment in Transit Supportive Density 0.1% -2.0% 0.0% -0.3% Subtotal 1.4% 1.2% Gas Price Average Gas Price (2018$) -28.5% -28.4% -3.7% -3.9% Subtotal -3.7% -3.9% Household and Income Characteristics Median Per Capita Income (2018$) 11.5% 9.4% -0.8% -0.7% Percent of Households with 0 Vehicles -7.1% -13.5% -0.2% -0.2% Percent Working at Home 24.1% 32.5% -0.9% -1.4% Subtotal -1.8% -2.3% New Competing Modes Years Since Ride-Hail Start 5.88 4.21 1.3% -9.7% Bike-Share 0.64 0.50 -0.7% -0.6% Electric Scooters 0.64 0.55 -2.4% -2.2% Subtotal -1.8% -12.5% Total Modeled Ridership 1.3% -0.5% Total Observed Ridership -2.9% -5.9% Unexplained Change -4.2% -5.4% Table 3-2. Contributions to rail ridership change between 2012 and 2018.

Multicity Evaluation 29   for the Washington Metro system were a factor that the research team wanted to test. While it had an important effect on rail ridership locally, its relative contribution was small when grouped with other MSAs. Average rail fares increased by 13% in the high operating expenses group and by 7% in the mid operating expenses group, leading to 3% and 1% lower rail ridership, respectively. Popula- tion and employment increased 6% on average in MSAs with rail, leading to about 1% more rail ridership. Much like bus, lower gas prices led to 4% lower rail ridership, while income growth, higher car ownership and more people working from home led to 2% lower rail ridership. Bike- sharing and electric scooters contributed to lower rail ridership, although the bike-sharing effect is insignificant. The ride-hailing effect is different for the two groups. Ride-hailing has a positive but insignificant effect on ridership in the high operating expenses group while contributing significantly to a 10% decrease in rail ridership in the mid operating expenses group. The model suggests that when all of these factors are considered together, rail ridership would be expected to increase slightly in the high operating expenses group and decrease slightly in the mid operating expenses group. In comparison, observed ridership decreased 3% and 6% in the groups, respectively. This means that rail ridership decreased by 4%–5% more than this model explains. It is not surprising that this model does not fully capture the changes to rail ridership because there are fewer MSAs with rail, and rail systems in the United States are diverse—they include heavy rail systems that are many decades old, newly constructed light rail systems, com- muter rail systems, and more. The smaller number of observations makes it more difficult to capture some of the dynamics that may affect rail differently. However, it is important to note that considering the large expansion of rail service over this period, a corresponding ridership increase should be expected. The fact that rail ridership declined in spite of its expansion is quite striking, and the model does capture most of this difference. In Figure 3-6 and Figure 3-7, the model was applied to each year from 2012 through 2018, and the effect of each factor on rail ridership was plotted. These plots take the same format as Figures 3-3 to 3-5, where the black line shows the observed ridership, shaded red areas above the black line indicate the amount that ridership is lower due to changes in that factor, and shaded green areas below the black line indicate the amount that ridership is higher due to changes in that factor. In both of these figures, there are substantial net rail ridership increases attributable to service increases, with net decreases due to lower gas prices starting in 2015. Changes to land use and household and income characteristics had a small effect on rail ridership in each group. Fare increases led to lower rail ridership in both groups, with a larger effect on the high operating expenses group. The effect of new competing modes can be observed in year 2018 in the high operating expenses group. In the mid operating expenses group, competition with ride-hailing is an important contributor to lower ridership, and rail ridership in this group would have been roughly flat if not for losses to ride-hailing. 3.4 Conclusion In this analysis, the researchers examined bus and rail ridership trends for MSAs grouped by transit operating expenses per capita. They found that between 2012 and 2018, bus ridership in all three groups decreased by about 15%, with the decline steepest between 2014 and 2018. The decline in rail ridership started later than bus ridership and is less steep, but it was still most concentrated in the 2014–2018 period. By 2018, even rail ridership was 3% lower than its 2012 reference point in the high operating expenses group and 6% lower in the mid operating expenses group.

30 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses Figure 3-6. Contributions to rail ridership change for high operating expenses group.

Multicity Evaluation 31   Figure 3-7. Contributions to rail ridership change for mid operating expenses group.

32 Recent Decline in Public Transportation Ridership: Analysis, Causes, and Responses The research team identified a number of factors that affect transit ridership, some of which result in net increases and others in net decreases to transit ridership. Overall, the team found that two sets of factors pushed to increase transit ridership from 2012 to 2018: • More service. Across all groups and modes, transit operators are providing more service in the form of added VRM. These service additions vary by location. They resulted in net bus rider- ship increases ranging from 3% in the high operating expenses group to 5% in the medium operating expenses group. This period continues a period of investment in expanded rail service, resulting in net rail ridership increases of 10% in the high operating expenses group and 18% in the mid operating expenses group. • Land use. Land use also affects transit ridership, both in terms of total population and employ- ment growth and how centralized that growth is. By cluster, metro areas grew between 6% and 8% in population and employment, with that growth occurring more in less dense sub- urbs than in centralized areas of the regions. The combined effect of land use changes was net bus and rail ridership increases of between 1% and 1.5%. It makes sense that these effects are modest because while land use is an important driver of transit ridership, changes tend to occur over a long time frame. The causes of bus ridership decline between 2012 and 2018 come from a combination of four main sources. Together, these sources more than offset the factors listed above that pushed ridership up over this period. They include the following: • Income and household characteristics. Higher incomes, higher car ownership, and an increase in the percent of people working at home contributed to net bus and rail ridership declines of about 2%. • Higher fares. Fare increases are operator-specific—so the effect varies by location—but fares were, on average, higher in 2018 than in 2012 after adjusting for inflation. Average bus fares went up by a maximum of 18% in the low operating expenses group metro areas, accompanied by a 4% decrease in bus ridership. In the other two groups, the effect of little to no change in fares contributed a less than 1% change in bus ridership. Rail fares increased by 13% in the high operating expenses group and 7% in the mid operating expenses group, resulting in 3% and 1% lower ridership, respectively. • Lower gas prices. Average inflation-adjusted gas prices decreased by more than a quarter over this period, leading to between 3% and 4% lower bus and rail ridership. • New modes compete with bus. Three new modes emerged in cities over this period that compete directly with bus: ride-hailing, bike-sharing, and e-scooters. The analysis shows that the effects of bike-sharing systems and e-scooters are much smaller compared to ride-hailing services. Ride-hailing itself contributes to 10%–12% lower bus ridership, with the combined effect of all modes leading to 10%–14% lower bus ridership. For rail, the effect of ride-hailing varies by group. In the high operating expenses group, ride-hailing’s introduction increases rail ridership by an insignificant amount, but in the mid operating expenses group, ride- hailing reduces rail ridership by 10%. The combined effect of all three new modes leads to 2% lower rail ridership in the high operating expenses group and 12% lower rail ridership in the mid operating expenses group. By providing a better understanding of the reasons for recent transit ridership declines, this research puts transit operators and transportation planners in a better position to effectively respond to those declines.