Conference PaperPDF Available

Understanding the Recent Transit Ridership Decline in Major US Cities: Service Cuts or Emerging Modes?


Abstract and Figures

Public transit ridership in major US cities has been flat or declining over the past few years. Several authors have attempted both to explain this trend and to offer policy recommendations for how to respond to it. Past writing on the topic is dominated by theoretical arguments that identify possible explanations, with the few empirical analyses excluding the most recent data, from 2015-2018, where the decline is steepest. This research conducts a longitudinal analysis of the determinants of public transit ridership in major North American cities for the period 2002-2018, segmenting the analysis by mode to capture differing effects on rail versus bus. Our research finds that standard factors, such changes in service levels, gas price and auto ownership, while important, are insufficient to explain the recent ridership declines. We find that the introduction of bike share in a city is associated with increased light and heavy rail ridership, but a 1.8% decrease in bus ridership. Our results also suggest that for each year after Transportation Network Companies (TNCs) enter a market, heavy rail ridership can be expected to decrease by 1.3% and bus ridership can be expected to decrease by 1.7%. This TNC effect builds with each passing year and may be an important driver of recent ridership declines.
Content may be subject to copyright.
Understanding the Recent Transit Ridership Decline in Major US Cities: Service Cuts or 1 Emerging Modes? 2
3 Michael Graehler, Jr. 4 Department of Civil Engineering, University of Kentucky 5 216 Oliver H. Raymond Bldg., Lexington, KY 40506 6 859-492-7535, 7 8 Richard Alexander Mucci 9 Department of Civil Engineering, University of Kentucky 10 216 Oliver H. Raymond Bldg., Lexington, KY 40506 11 859-257-4856, 12 13 Gregory D. Erhardt (corresponding author) 14 Department of Civil Engineering, University of Kentucky 15 261 Oliver H. Raymond Bldg., Lexington, KY 40506 16 859-323-4856, 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 Submitted for Presentation Only 32 33 98th Annual Meeting of the Transportation Research Board 34 35 36 Word count: 5,623 Words + 4 Tables = 6,623 Total Words 37 38 39 40 Submitted: August 1, 2018 41
Revised and re-submitted: November 14, 2018 42
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 2
ABSTRACT 1 Public transit ridership in major US cities has been flat or declining over the past few years. 2 Several authors have attempted both to explain this trend and to offer policy recommendations 3 for how to respond to it. Past writing on the topic is dominated by theoretical arguments that 4 identify possible explanations, with the few empirical analyses excluding the most recent data, 5 from 2015-2018, where the decline is steepest. This research conducts a longitudinal analysis of 6 the determinants of public transit ridership in major North American cities for the period 2002-7 2018, segmenting the analysis by mode to capture differing effects on rail versus bus. 8 9 Our research finds that standard factors, such changes in service levels, gas price and auto 10 ownership, while important, are insufficient to explain the recent ridership declines. We find 11 that the introduction of bike share in a city is associated with increased light and heavy rail 12 ridership, but a 1.8% decrease in bus ridership. Our results also suggest that for each year after 13 Transportation Network Companies (TNCs) enter a market, heavy rail ridership can be expected 14 to decrease by 1.3% and bus ridership can be expected to decrease by 1.7%. This TNC effect 15 builds with each passing year and may be an important driver of recent ridership declines. 16 17 18 Key Words: Transit Ridership, Public Transportation, Ridesourcing, TNC, Uber, Bus, Rail, 19 Longitudinal Analysis 20
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 3
INTRODUCTION 1 Following strong ridership growth during much of the previous decade (1), public transit 2 ridership in major US cities has been flat or declining over the past few years (24). The 3 changes vary by mode and by agency, but can be observed using data from the National Transit 4 Database (NTD) (5), as shown in Figure 1. Figure 1 shows the percent change in transit 5 ridership, using Fiscal Year (FY) 2002 as a base, for the largest transit agencies in seven large 6 US cities: Boston, New York, Washington, DC, Chicago, Denver, San Francisco and Los 7 Angeles. Three separate graphs show the ridership on heavy rail, light rail and bus, with heavy 8 and light rail only available in a subset of cities. The graphs show that heavy rail ridership grows 9 steadily in four of five cities until about 2014, then declines, with the decline in Washington, DC 10 starting earlier. Light rail ridership is relatively flat in Boston and San Francisco, and grows 11 substantially in Denver and Los Angeles, two cities that expanded their light rail systems over 12 this period. Bus ridership is relatively flat for much of this period, with noticeable declines 13 starting between 2013 and 2016 on each of the bus systems except San Francisco, which has 14 embarked on a series of bus service improvement projects over this period (6). 15 16 A number of explanations have been offered for what might be causing this trend, including: 17 income growth combined with cheap gas (7); increased car ownership (2, 3); transit service cuts 18 (8); reliability issues associated with deferred maintenance (2, 9); increased bicycling, bike 19 sharing, and electric scooter use (3, 4); and the expansion of Transportation Network Companies 20 (TNCs) such as Uber and Lyft (3, 4). Crafting an effective policy response to this trend depends 21 upon first understanding its cause. 22 23 Two recent studies are worth considering in further detail: an analysis of ridership trends in 24 Southern California (10) and a longitudinal study of ridership in 25 North American cities (11). 25 26 Manville et al (10) considered the issue of falling transit ridership in Southern California and 27 concluded that the trend was largely due to increased auto ownership among immigrant 28 populations. Their recommended response is to convince people who rarely or never use transit 29 to do so occasionally. Their conclusion is based on data covering the period from 2000-2015, 30 and shows that much of the decrease in auto ownership occurred between 2000 and 2010. In 31 contrast, the NTD data (Figure 1) show that the steepest decline in transit ridership occurs from 32 2015-2018. Given that auto ownership is a long-term decision, it would be surprising if it 33 changed rapidly enough to explain this more recent decline. 34 35 Boisjoly et al (11) find that transit service cuts and auto ownership are the main determinants of 36 transit ridership. They argue that given this evidence, transit agencies should prioritize 37 expanding service to counteract these trends. Their method was a longitudinal analysis of the 38 determinants of transit ridership using 2002-2015 NTD for 22 US cities, plus equivalent data for 39 3 Canadian cities. Specifically, they estimated panel data regression models, in this case 40 multilevel mixed-effects models, to correlate the changes in transit ridership with changes in 41 descriptive variables such as vehicle revenue miles (VRM), average fare, the share of zero-car 42 households and population. This is a logical approach to studying the problem. Similar panel 43 data methods used previously to study the determinants of transit ridership changes (1214), with 44 those methods offering an advantage over time-series models which are sometimes used as well 45 (15, 16) because the panel models can consider data from multiple cities at once. 46
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 4
FIGURE 1. Percent Change in Transit Ridership from 2002 1
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 5
While Boisjoly’s methodology is sound, their data ends in 2015, which is about when we 1 observe some of the largest ridership declines begin (see Figure 1). This raises the possibility 2 that their models miss the most important part of the trend. In addition, their models are based 3 on the total ridership in each city, summed across modes. As can be observed by the different 4 trends between light rail and bus in Denver and Los Angeles, there is a possibility that this 5 aggregation washes out the change we are trying to detect. This paper updates Boisjoly’s 6 analysis using the most recently available data, segmented by mode. In doing so, we consider 7 whether their conclusions still hold, as well as possible implications for effective policy 8 responses by transit agencies. 9
BACKGROUND AND LITERATURE REVIEW 10 A number of studies have examined the factors that influence transit ridership (1, 1220). These 11 studies point to a core set of variables that are included across multiple studies, and can be 12 considered as well established determinants. These include: population, employment, VRM, 13 fare, car ownership and gas price. 14 15 Evaluation of the recent declines is dominated by theoretical arguments of what may have 16 changed over the past few years, often appearing in blog posts and media articles (24, 79). 17 These articles are useful in identifying potential causes, which include: 18
Income growth combined with cheap gas (7), 19
Increased car ownership (2, 3), 20
Service cuts (8), 21
Reliability issues associated with deferred maintenance (2, 9), 22
Increased bicycling, bike sharing, and more recently electric scooters (3, 4), and 23
The expansion of Transportation Network Companies (TNCs) such as Uber and Lyft (3, 24 4). 25
It is worth considering each of these factors, first by noting that the first three overlap with the 26 core variables noted above. The economy has been strong over the past few years, with 27 employment growth outpacing income growth. Income growth could lead to increased car 28 ownership and decreased transit ridership. However, it is also associated with strong 29 employment growth, and transit ridership tends to increase with employment growth because 30 more people commute to work. Gas prices have declined, hitting an average of $2.83 per gallon 31 in April 2018 compared to $3.63 per gallon five years earlier (21), so this may be a contributing 32 factor. 33 34 Car ownership is another logical determinant of transit ridership, with 0-car households 35 especially dependent upon transit. As discussed previously, Manville et al (10) attributed falling 36 transit ridership in Southern California largely to increased auto ownership among immigrant 37 populations. It is not clear whether car ownership is changing quickly enough to explain the 38 rapid transit ridership decline since 2015, but it is clearly a factor that must be considered. 39 40 Service cuts were identified by Boisjoly (11) as the driving factor, and it is logical that they 41 would affect ridership. The question is: how much? To better understand this, we can examine 42 the change in ridership versus the change in VRM. Figure 2 shows the percent change in 43 ridership per VRM for the same cities and modes shown in Figure 1. The light rail trend is the 44
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 6
most obviously different, with the large growth in total light rail ridership in Denver and Los 1 Angeles apparently driven by expanded service on those systems. However, Figure 2 also shows 2 that ridership per VRM is decreasing on most systems. In particular, we observe that the recent 3 bus service expansion in San Francisco seems to have counteracted a background trend of 4 declining ridership per VRM. These data suggest that something else has changed over the past 5 few years, beyond service provision, that is contributing to the decline in an important way. 6 7 Reliability and maintenance issues are a potential contributing factor, although their influence 8 may be limited to specific systems, such as New York and Washington heavy rail. 9 10 Bike sharing is new in many cities over this period, while bicycling broadly is experiencing a 11 “renaissance” with expanded bike lanes in many cities and increased use (22, 23). Bike share, 12 and bicycling in general, could compete with transit if transit users switch to bike, or it could 13 complement transit by providing first- and last-mile connectivity. Boijoly et al (11) include in 14 their models a flag for the presence of bike sharing, and find that it is correlated with higher 15 transit ridership, although not at a statistically significant level. Conversely, Campbell and 16 Brakewood conducted a more detailed study of the effect of bike sharing on bus ridership in New 17 York, and found that each additional 1000 bike share docks proximate to a bus route are 18 associated with a 1.7% to 2.4% decrease in bus ridership (24). It would be reasonable to expect 19 a similar effect from the introduction of electric scooters or similar new modes. 20 21 There is disagreement over the effect of TNCs on transit ridership. Some authors argue that 22 TNCs are likely to increase transit ridership by providing first- and last-mile connectivity, 23 providing service at locations and times (such as late at night) when there is less transit service 24 provided, or by reducing car ownership (25, 26), while other studies show that TNC users are 25 likely to switch from transit, reducing ridership (2729). Both may be true to varying degrees. 26 A survey of TNC users in seven US cities finds that TNCs are associated with a 6% derease in 27 bus trips, a 3% decrease in light rail trips, and a 3% increase in commuter rail trips (30). 28 29 As a proxy for TNC use, Boijoly et al (11) test the presence of Uber in their longitudinal model, 30 and find that it is associated with higher transit ridership, but that the effect is not significant. 31 They conclude from this that TNCs are not a major determinant of the recent decline in transit 32 ridership, although they do also note that there is a general lack of TNC use data. Similarly, 33 Manville et al (10) note that they have very little data to measure the effect of TNCs on transit 34 ridership, but go on to dismiss the importance of TNCs effect on transit using theoretical 35 arguments similar to those in (25, 26). 36 37 It is important here that we not confuse the lack of data with the lack of importance, and that we 38 consider what we can learn from locations where we do have data. One such location is San 39 Francisco, where there were 170,000 daily TNC trips in 2016, representing 15% of intra-San 40 Francisco vehicle trips (31). An analysis of these data in combination with automated passenger 41 count (APC) data found that TNCs decrease bus ridership, but not rail (32). Another location 42 where reasonably good TNC data exist is New York, where TNC trips must be reported to the 43 city’s Taxi and Limousine Commission, and a recent study found that TNC use appears to be 44 associated with decreasing transit ridership (33). 45 46
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 7
FIGURE 2. Percent Change in Transit Ridership per Vehicle Revenue Mile from 2002 1
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 8
The New York data are particularly useful because they are available by month. Figure 3 shows 1 the total daily Uber and Lyft trips in New York (34), which grow from about 60,000 to nearly 2 600,000 between 2015 and 2018. This rapid TNC growth corresponds to a period of declining 3 transit ridership (daily subway and bus ridership in New York decrease by 580,000 boardings 4 between April 2015 and April 2018 according to the NTD), as well as to a period beyond the 5 bounds other recent studies. It further demonstrates that the presence of Uber is not a binary 6 variable, and given the dramatic change in magnitude, we would expect the quantity of trips to 7 matter. 8 9
10 FIGURE 3. Daily TNC Trips in New York 11 12 This research aims to consider each of these factors, using the most recently available data. It 13 follows the methodology employed by Boijoly et al’s (11), with the following extensions: 14
It considers data from 2002 through April 2018, the most recently available in the NTD, 15
It segments the analysis by mode, to capture the possibility that the effects are different 16 for different transit modes, 17
It uses monthly data rather than annual data, which is the native resolution of the NTD, 18
It includes employment in the model in order to capture the effect of economic growth 19 over the past few years, and 20
It considers that the TNC effect is not binary, but instead increases with the growth of 21 TNCs. Because we still lack data on TNC use beyond a few specific cities, we make an 22 assumption that TNC use grows linearly starting from the date it is introduced to a new 23 market. To capture this, we use a variable that is defined as the number of years since 24 Uber entered the market to take the place of the binary Uber presence variables. 25
A few other differences from the previous study should be noted. First, the study is limited to 22 26 US cities, excluding the three Canadian cities for which data are not publicly available. Second, 27
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 9
it uses a different econometric model: a random-effects model instead of a mixed-effects model. 1 Incorporating both would be a useful future improvement. 2
DATA AND METHODS 3 For this study, we conducted a longitudinal analysis using monthly transit ridership data from the 4 National Transit Database for the 22 transit agencies and four modes (commuter rail, heavy rail, 5 light rail and motor bus) shown in Table 1. Unlinked passenger trips are available for each mode 6 allowing a total of 51 agency-mode combinations. All NTD data were collected from January 7 2002 to April 2018. 8 9 In addition to the ridership data, this study considers the possible determinants listed as variables 10 in Table 2. NTD is also the source for vehicle revenue miles and fares, with VRM broken out by 11 mode. The average fare is calculated as the fare revenue divided by the unlinked passenger trips. 12 It is adjusted for inflation, with 2016 USD as the base rate. All dollar-based data were adjusted 13 for inflation using 2016 as the base year. 14 15 We gathered data for the metropolitan population from the American Community Survey (ACS) 16 1-year estimates, and from the 2000 Census. The ACS data were collected from 2005 to 2017. 17 We linearly interpolated the years 2000 to 2005 to come up with data for years 2002 to 2004. 18 We extrapolated the data to 2018 to extend the usefulness of the data. We also linearly 19 interpolated between years to get the data in monthly terms. The percent of households with 20 zero vehicles is from the same sources and processed in the same way. 21 22 Metropolitan land area for the 22 metropolitan areas was also sourced from the United States 23 Census Bureau’s numbers for the urban area in 2010. We assumed that the metro land area 24 remained constant throughout the time period of our research. Employment data also came from 25 the Bureau of Labor Statistics. Monthly data was given for the full array of dates in our research. 26 27 Gasoline price data were sourced from the US Energy Information Administration. The data 28 came in as a weekly figure. We took the weekly data, calculated monthly averages and adjusted 29 for inflation to 2016 US dollars. 30 31 Data for Uber’s start date in each city was found primarily from Uber’s press releases. Other 32 confirming sources include local newspaper articles. Bike share start-up dates were found from 33 local newspaper articles and from Oliver O’Brien’s bike share map (35). We split the years 34 since Uber and bike share presence variables into the four different modes used in this model to 35 account for any differences between the modes. 36 37
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 10
TABLE 1: Metropolitan Areas, Transit Agencies, and Modes Analyzed 1
Metropolitan Area
Core City
Transit Agency
Atlanta - Sandy Springs - Marietta, GA
Metropolitan Atlanta Rapid
Transit Authority (MARTA)
Heavy rail, motor bus
Baltimore - Towson, MD
Maryland Transit
Heavy rail, light rail, motor
Boston - Cambridge - Quincy, MA-NH-RI
Massachusetts Bay
Transportation Authority
Commuter rail, heavy rail,
light rail, motor bus
Chicago - Joliet - Naperville, IL-IN-WI
Chicago Transit Authority
Heavy rail, motor bus
Cleveland - Elyria - Mentor, OH
The Greater Cleveland
Regional Transit Authority
Heavy rail, light rail, motor
Dallas - Fort Worth - Arlington, TX
Dallas Area Rapid Transit
Light rail, motor bus
Denver - Aurora - Broomfield, CO
Denver Regional
Transportation District
Light rail, motor bus
Houston - Sugar Land - Baytown, TX
Metropolitan Transit Authority
of Harris County (Metro)
Light rail, motor bus
Los Angeles - Long Beach - Santa Ana, CA
Los Angeles
Los Angeles County
Metropolitan Transportation
Authority (LACMTA)
Heavy rail, light rail, motor
Miami - Ft. Lauderdale - Pompano Beach,
Miami - Dade Transit (MDT)
Heavy rail, motor bus
Minneapolis - St. Paul - Bloomington, MN-
Metro Transit
Light rail, motor bus
New York - Northern New Jersey - Long
Island, NY-NJ-PA
New York
MTA New York City Transit
Heavy rail, motor bus
Philadelphia - Camden - Wilmington, PA-
Southeastern Pennsylvania
Transportation Authority
Commuter rail, heavy rail,
light rail, motor bus
Pittsburgh, PA
Port Authority of Allegheny
Light rail, motor bus
Portland - Vancouver - Hillsboro, OR-WA
Tri-County Metropolitan
Transportation District of
Light rail, motor bus
Sacramento - Arden - Arcade - Roseville,
Sacramento Regional Transit
Light rail, motor bus
San Diego - Carlsbad - San Marcos, CA
San Diego
San Diego Metropolitan Transit
Light rail, motor bus
San Francisco - Oakland - Fremont, CA
San Francisco
San Francisco Municipal
Railway (SFMTA)
Light rail, motor bus
San Jose - Sunnyvale - Santa Clara, CA
San Jose
Santa Clara Valley
Transportation Authority
Light rail, motor bus
Seattle - Tacoma - Bellevue, WA
King County Department of
Transportation (King County
Metro - KCM)
Light rail, motor bus
St. Louis, MO-IL
St. Louis
Bi-State Development (BSD)
Light rail, motor bus
Washington - Arlington - Alexandria, DC-
Washington Metropolitan Area
Transit Authority (WMATA)
Heavy rail, motor bus
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 11
TABLE 2: Description of Available Variables 1
Date Range
Ridership (UPT)
Number of unlinked
passenger trips
Vehicle Revenue Miles
Miles that vehicles
travel while in
revenue service
Fare revenue per
USD / trip
Adjusted for
inflation. Base rate
2016 USD.
Metro population
Interpolated data
between 2000-
to capture years
Extrapolated to
2018. July data
given - linearly
interpolated to
make data monthly.
Percent of household
without a car
Percent of
households without
a car
2005 data used for
years 2002-2004.
2017 data used for
2018. July data
given - linearly
interpolated to
make data monthly.
Metro Land Area
US Census Bureau
Land area of the
metropolitan area
Bureau of Labor
Employed persons
in metropolitan area
Gas price
US Energy
Average price of gas
Weekly data given.
Averaged weeks in
each month to
come up with
monthly data.
Adjusted for
inflation. Base rate
2016 USD.
Years Since Uber
Uber press
releases and other
news outlets
Years since Uber
first appeared in
metro area
Bike Share Presence
Bike Share Map
and other news
Whether or not a
city has a bike
sharing system
1 =
0 = Not
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 12
We analyze these data using a random-effects panel data model (36). A random-effects model is 1 a form of a regression model that estimates the correlation between the dependent variable 2 (unlinked passenger trips) and a set of descriptive variables based on differences both across the 3 51 entities and through time. Such models have been applied successfully in other studies 4 transportation studies (37). We also tested a fixed-effects model, but found that it resulted in an 5 employment coefficient with an illogical sign. We specify the model using a log transformation 6 on the dependent variable, and on all descriptive variables except the Uber and bike share terms. 7 For a log-log model, the coefficients can be interpreted directly as elasticities. 8
RESULTS 9 Table 3 shows the model estimation results. The first set of variables is a set of constants, one 10 for each month,that serve to control for seasonality. 11 12 The core variables are each significant and have a logical sign. Ridership increases with an 13 increase in VRM, and decreases with fare increases, as we would expect. The coefficients show 14 that higher metropolitan area population is correlated with higher ridership. This is intuitive 15 because if more people live in the metropolitan area, then more people are bound to opt for 16 transit as a transportation option. The model indicates that increasing the percentage of 17 households that do not own a car will have a positive effect on transit ridership. The metro land 18 area has a positive coefficient, although this is not thought to be especially important. Increased 19 employment is also correlated with increased transit ridership. Similar to increasing population, 20 it is apparent that more employment in an area will mean that more people commuting to and 21 from work, thus increasing transit ridership. Higher gas prices are correlated with higher 22 ridership, as travelers look to save money by switching to transit when gas prices are high. 23 24 The effect of bike sharing varies by mode. The commuter rail coefficient is negative, but 25 insignificant, so we ignore it. Of more interest are the heavy rail, light rail and bus coefficients, 26 each of which is significant, but with different signs. The positive coefficients for rail suggest 27 that bike share is a complement to rail, perhaps because it can be linked with rail trips serving a 28 first- and last-mile role. In contrast, the bus coefficient is negative and significant, suggesting 29 that bike share reduces bus ridership. This is also logical because bus trips are on average 30 shorter than rail trips, and thus users may be more likely to switch to bike share due to the similar 31 distances served by both modes. 32 33 The TNC coefficients also vary by mode. The commuter rail coefficient is positive, suggesting 34 complementarity, but insignificant. The heavy rail and bus coefficients are negative and 35 significant. This suggests that TNCs reduce transit ridership. The effect is greater for each year 36 after TNCs enter a market, with the coefficient interpreted as a growth rate. After TNCs enter a 37 market, heavy rail ridership decreases by 1.29% per year, and bus ridership decreases by 1.70% 38 percent per year. This is reasonable to expect as TNC use grows after entering a market. The 39 light rail coefficient is also negative, but is insignificant. 40 41 42
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 13
TABLE 3: Model Estimation Results 1
Month - January
Month - February
Month March
Month April
Month May
Month June
Month July
Month - August
Month September
Month October
Month November
Month December
Core Variables
Vehicle Revenue Miles (ln)
Fare Revenue per UPT (ln)
Metro Population (ln)
Percent Households with No Vehicle (ln)
Metro Land Area (ln)
Employment (ln)
Gas Price (ln)
Bike Share Effect
Presence of Bike Sharing - Commuter Rail
Presence of Bike Sharing - Heavy Rail
Presence of Bike Sharing - Light Rail
Presence of Bike Sharing - Motor Bus
TNC Effect
Years Since Uber - Commuter Rail
Years Since Uber - Heavy Rail
Years Since Uber - Light Rail
Years Since Uber - Motor Bus
R-squared (between groups)
R-squared (within groups)
R-squared (overall)
Time Periods
* Insignificant variables are in gray italics. 2 3 Table 4 illustrates the effect of the bike share and TNC variables, relative to the effect of changes 4 in VRM. The values show that bike share is associated with a 6.9% increase in heavy rail 5 ridership, a 4.2% increase in light rail ridership, and a 1.8% decrease in bus ridership, 6 corresponding directly to the estimated coefficients. The TNC effect is a 1.3% decrease in heavy 7 rail ridership and a 1.7% decrease in bus ridership per year. In a market like San Francisco, 8 where Uber started operations in 2010, the model implies that we would expect a 12.7% decrease 9 in bus ridership, all else being equal. The estimated coefficient on VRM is 0.462, which means 10
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 14
that a 1% increase in VRM corresponds to a 0.42% increase in VRM. This is specific to the 1 mode, but the coefficient is not segmented by mode. Extending this further, Table 4 shows the 2 effect of different percent increases in VRM. Continuing with San Francisco as an example, this 3 result suggests that SFMTA would need to increase bus service by slightly more than 25% in 4 order to offset the loss of bus ridership to TNCs. 5 6 TABLE 4: Effect of Changes in Select Variables 7
Change Commuter
Heavy Rail
Light Rail
Bike Share Enters Market
Binary Effect
TNCs Enter Market
Year 1 2.0% -1.3% -0.4% -1.7%
Year 2 4.0% -2.5% -0.8% -3.3%
Year 3 6.0% -3.8% -1.1% -5.0%
Year 4 8.1% -5.0% -1.5% -6.6%
Year 5 10.2% -6.2% -1.9% -8.1%
Year 6 12.4% -7.4% -2.3% -9.7%
Year 7 14.6% -8.6% -2.6% -11.2%
Year 8 16.9% -9.8% -3.0% -12.7%
Increase VRM
* Statistically insignificant effects are in gray italics.
DISCUSSION 9 The results presented above represent provide insight into the determinants of public transit 10 ridership in 22 US cities. The core variables included in the model include service provision, 11 fares, population, employment, auto ownership, land area and gas price. The estimated 12 coefficients on these core variables are logical, and consistent with previously published research 13 (1, 1219). Most variables are consistent in sign, and often in magnitude, with the study being 14 replicated (11), with notable differences in the statistical method used and in the fact that our 15 models include employment. The inclusion of an employment term is especially important given 16 the strong economic growth over the past few years. Employment growth should result in a net 17 increase in transit ridership, making the declines observed since 2015 more stark. 18 19 The bike share term estimated in our model suggests that bike share increases heavy rail and 20 light rail ridership, but decreases bus ridership. Boisjoly et al (11) find that bike sharing has a 21 positive but insignificant effect on transit ridership. The difference between the two findings 22
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 15
may be due to averaging across modes. Our result is also consistent with Campbell and 1 Brakewood’s finding that bike share has decreased New York bus ridership (24). 2 3 Our finding suggests that TNCs reduce transit ridership, specifically on heavy rail and bus. 4 Further, we find that the effect increases as TNCs become more established in a market. This 5 finding differs from that of Boisjoly et al (11), with the difference potentially attributable to our 6 inclusion of more recent data, or specification of the variable such that it is an effect that grows 7 with time. Our finding supports related research on the effect of TNCs on transit ridership (30, 8 32, 33), and contradicts the arguments made by some shared mobility advocates (25, 26). It 9 should be noted, however, that the estimated effect of TNCs on heavy rail is likely to be heavily 10 influenced by New York subway ridership, and may differ if the study were expanded to more 11 cities. 12 13 This raises another limitation of the studyit is focused on 22 large US cities, and these effects 14 may be different for smaller and medium cities with a different composition and character. In 15 addition, certain cities may be influenced by specific conditions, such as service changes or 16 maintenance issues that are not captured here. It would be useful for future studies to both 17 expand the analysis to more cities, and to examine specific cities in further detail. 18 19 A second limitation of this study is the aggregate treatment of both bike share and TNCs. The 20 former is treated as a binary variable, and the latter as a trend starting from the date of Uber’s 21 entry into the market. Actual ridership data for both would improve the analysis, although the 22 prospects of obtaining the first without regulatory intervention may be stronger. 23
CONCLUSIONS 24 This study aimed to extend recently published research that conducted a longitudinal analysis of 25 the determinants of public transit ridership in major North American cities (11). In doing so, it 26 extended the longitudinal analysis to cover the period from 2015-2018 when notable declines in 27 public transit ridership are observed. It also segments the models by mode to capture differing 28 effects on rail versus bus. 29 30 Our results suggest that previous conclusions that reductions in bus VRM explain the reduction 31 in transit ridership in many North American cities (11) may be flawed. While we do find that 32 VRM is an important determinant of transit ridership, we also find it to be insufficient to explain 33 the recent ridership declines, particularly the decline in ridership per VRM observed since 2015 34 for both bus and rail modes. 35 36 Our research also suggests that past research findings that TNCs and other emerging modes 37 either increase or do not affect transit ridership (11, 25, 26, 38) are likely incorrect. Our results 38 show that the introduction of bike share in a city is associated with light and heavy rail ridership, 39 but a 1.8% decrease in bus ridership. Our results also suggest that for each year after TNCs enter 40 a market, heavy rail ridership can be expected to decrease by 1.3% and bus ridership can be 41 expected to decrease by 1.7%. This effect increases with time as TNCs increase in use. The 42 effect of TNCs is substantial—after 8 years this would be associated with a 12.7% decrease in 43 bus ridership. 44 45
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 16
While bike share is a sustainable mode of transport, the consequences of a shift from public 1 transit to TNCs go beyond the effect on transit agencies. Recent research suggests that this shift 2 results in a large increase in traffic congestion (33, 3942), which may result in most travelers 3 being worse off. 4 5 The implication of misdiagnosing the causes of recent ridership declines is that it may lead to 6 ineffective policy responses. Boisjoly et al (11) recommend that transit agencies should focus 7 their efforts on expanding service to attract ridership. While expanding service does result in a 8 net increase ridership, as can be observed from the recent bus service expansion in San 9 Francisco, the amount of service expansion required to offset the TNC effect is substantial. To 10 offset the expected 1.7% annual loss of bus riders to TNCs, transit agencies would need to 11 increase bus VRM by 3.7% per year. After eight years, this would result in more than a 25% 12 service expansion just to maintain existing ridership. While service expansions are clearly 13 valuable, transit agencies are fighting an uphill battle. In order to implement effective policies, it 14 may be necessary to reach beyond the bounds of the transit agencies themselves and partner with 15 cities to consider strategies such as congestion pricing, or reallocating right-of-way on urban 16 streets away from cars and to transit. 17
ACKNOWLEDGMENTS 18 This work was funded internally by the University of Kentucky. 19 20 The authors confirm contribution to the paper as follows: study conception and design: Greg 21 Erhardt; data processing: Michael Graehler; analysis and interpretation of results: all authors; 22 draft manuscript preparation: Michael Graehler, Alex Mucci; final manuscript preparation: Greg 23 Erhardt. All authors reviewed the results and approved the final version of the manuscript. 24 25
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 17
REFERENCES 1 2 1. Rosenberger, T., C. Nardi, and K. Liwag. What Is Generating Transit Ridership Increase in 3 US and Canadian Cities? Presented at the European Transport Conference, 2015. 4 2. Bliss, L. Are Americans Abandoning Transit? CityLab. 5 nationwide/517701/. Accessed Aug. 1, 2018. 7 3. Siddiqui, F. Falling Transit Ridership Poses an ‘Emergency’ for Cities, Experts Fear. 8 Washington Post. transit-ridership-poses-an-emergency-for-cities-experts-fear/2018/03/20/ffb67c28-2865-10 11e8-874b-d517e912f125_story.html. Accessed Aug. 1, 2018. 11 4. Public Transport Is in Decline in Many Wealthy Cities. The Economist, Jun 21, 2018. 12 5. The National Transit Database (NTD). FTA. Accessed Aug. 13 1, 2018. 14 6. SFMTA. Muni Forward. Accessed May 16, 15 2018. 16 7. Levinson, D. On the Predictability of the Decline of Transit Ridership in the US. David 17 Levinson, Transportist, Mar 20, 2017. 18 8. Bliss, L. Service Cuts Are Killing the Bus. CityLab. 19 Accessed 20 Jun. 29, 2018. 21 9. Di Caro, M. Numbers Show Metro May Actually Be Improving — But Will Riders Return? 22 WAMU, Sep 11, 2017. 23 10. Manville, M., B. D. Taylor, and E. Blumenberg. Falling Transit Ridership: California and 24 Southern California. UCLA Institute for Transportation Studies, 2018. 25 11. Boisjoly, G., E. Grisé, M. Maguire, M.-P. Veillette, R. Deboosere, E. Berrebi, and A. El-26 Geneidy. Invest in the Ride: A 14 Year Longitudinal Analysis of the Determinants of Public 27 Transport Ridership in 25 North American Cities. Transportation Research Part A: Policy 28 and Practice, Vol. 116, 2018, pp. 434–445. 29 12. Holmgren, J. An Analysis of the Determinants of Local Public Transport Demand Focusing 30 the Effects of Income Changes. European Transport Research Review, Vol. 5, No. 2, 2013, 31 pp. 101–107. 32 13. Kennedy, D. Panel Data Analysis of Public Transport Patronage Growth–an Innovative 33 Econometric Approach. 2013. 34 14. Iseki, H., and R. Ali. Fixed-Effects Panel Data Analysis of Gasoline Prices, Fare, Service 35 Supply, and Service Frequency on Transit Ridership in 10 U.S. Urbanized Areas. 36 Transportation Research Record: Journal of the Transportation Research Board, Vol. 2537, 37 2015, pp. 71–80. 38 15. Chen, C., D. Varley, and J. Chen. What Affects Transit Ridership? A Dynamic Analysis 39 Involving Multiple Factors, Lags and Asymmetric Behaviour. Urban Studies, Vol. 48, No. 9, 40 2011, pp. 1893–1908. 41 16. Lane, B. W. A Time-Series Analysis of Gasoline Prices and Public Transportation in US 42 Metropolitan Areas. Journal of Transport Geography, Vol. 22, 2012, pp. 221–235. 43 17. Lee, K. S., J. K. Eom, S. Y. You, J. H. Min, and K. Y. Yang. An Empirical Study on the 44 Relationship between Urban Railway Ridership and Socio-Economic Characteristics. 45 Procedia Computer Science, Vol. 52, 2015, pp. 106–112. 46
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 18
18. Kain, J. F., and Z. Liu. Secrets of Success: Assessing the Large Increases in Transit 1 Ridership Achieved by Houston and San Diego Transit Providers. Transportation Research 2 Part A: Policy and Practice, Vol. 33, No. 7, 1999, pp. 601–624. 3 19. Mucci, R. A., and G. D. Erhardt. Evaluating the Ability of Transit Direct Ridership Models 4 to Forecast Medium-Term Ridership Changes: Evidence from San Francisco. Transportation 5 Research Record, 2018. 6 20. Taylor, B. D., D. Miller, H. Iseki, and C. Fink. Nature and/or Nurture? Analyzing the 7 Determinants of Transit Ridership across US Urbanized Areas. Transportation Research 8 Part A: Policy and Practice, Vol. 43, No. 1, 2009, pp. 60–77. 9 10 21. US Energy Information Administration. U.S. All Grades All Formulations Retail Gasoline 11 Prices (Dollars per Gallon). 12 DPG&f=M. Accessed Aug. 1, 2018. 14 22. Pucher, J., R. Buehler, and M. Seinen. Bicycling Renaissance in North America? An Update 15 and Re-Appraisal of Cycling Trends and Policies. Transportation Research Part A: Policy 16 and Practice, Vol. 45, No. 6, 2011, pp. 451–475. 17 23. de Chardon, C. M., G. Caruso, and I. Thomas. Bicycle Sharing System ‘Success’ 18 Determinants. Transportation Research Part A: Policy and Practice, Vol. 100, 2017, pp. 19 202–214. 20 24. Campbell, K. B., and C. Brakewood. Sharing Riders: How Bikesharing Impacts Bus 21 Ridership in New York City. Transportation Research Part A: Policy and Practice, Vol. 22 100, 2017, pp. 264–282. 23 25. Feigon, S., and C. Murphy. Shared Mobility and the Transformation of Public Transit. 24 Publication TCRP Report 188. Transportation Research Board, Washington, D.C., 2016. 25 26. Feigon, S., and C. Murphy. Broadening Understanding of the Interplay Between Public 26 Transit, Shared Mobility, and Personal Automobiles. Publication Pre-publication draft of 27 TCRP 195. Transportation Research Board, Washington, D.C., 2018. 28 27. Henao, A. Impacts of Ridesourcing-Lyft and Uber-on Transportation Including VMT, Mode 29 Replacement, Parking, and Travel Behavior. University of Colorado at Denver, 2017. 30 28. Rayle, L., D. Dai, N. Chan, R. Cervero, and S. Shaheen. Just a Better Taxi? A Survey-Based 31 Comparison of Taxis, Transit, and Ridesourcing Services in San Francisco. Transport 32 Policy, Vol. 45, No. C, 2016, pp. 168–178. 33 29. Gehrke, S., A. Felix, and T. Reardon. Fare Choices Survey of Ride-Hailing Passengers in 34 Metro Boston. Metropolitan Area Planning Council, 2018. 35 30. Clewlow, R. R., and G. S. Mishra. Disruptive Transportation: The Adoption, Utilization, and 36 Impacts of Ride-Hailing in the United States. Publication UCD-ITS-RR-17-07. 2017. 37 31. San Francisco County Transportation Authority. TNCs Today: A Profile of San Francisco 38 Transportation Network Company Activity. 2017. 39 32. Mucci, R. Transportation Network Companies: Influencers of Transit Ridership Trends. 40 University of Kentucky, 2017. 41 33. Schaller, B. The New Automobility: Lyft, Uber and the Future of American Cities. Schaller 42 Consulting, Brooklyn, NY, 2018. 43 34. NYC Open Data. FHV Base Aggregate Report. NYC Open Data. 44 45 Accessed Aug. 1, 2018. 46
TRB 2019 Annual Meeting Paper revised from original submittal.
Graehler, Mucci and Erhardt Page 19
35. Oliver O’Brien. Bike Share Map. Bike Share Map. Accessed Aug. 1 2, 2018. 2 36. Greene, W. H. Econometric Analysis. Prentice Hall, Upper Saddle River, N.J, 2003. 3 37. Gerte, R., K. C. Konduri, and N. Eluru. Is There a Limit to Adoption of Dynamic 4 Ridesharing Systems? Evidence from Analysis of Uber Demand Data from New York City. 5 Presented at the TRB Annual Meeting, Washington, D.C., 2018. 6 38. Hall, J. D., C. Palsson, and J. Price. Is Uber a Substitute or Complement for Public Transit? 7 Journal of Urban Economics, Vol. 108, 2018, pp. 36–50. 8 39. Erhardt, G. D., S. Roy, D. Cooper, B. Sana, M. Chen, and J. Castiglione. Do Transportation 9 Network Companies Decrease or Increase Congestion? Science Advances, in review. 10 40. Schaller, B. Unsustainable? The Growth of App-Based Ride Services and Traffic, Travel 11 and the Future of New York City. Schaller Consulting, Brooklyn, NY, 2017. 12 41. Roy, S., D. Cooper, R. A. Mucci, B. Sana, M. Chen, J. Castiglione, and G. D. Erhardt. Why 13 Is Traffic Congestion Getting Worse? A Decomposition of the Contributors to Growing 14 Congestion in San Francisco. Transportation Research Part A: Policy and Practice, In-15 Review. 16 42. San Francisco County Transportation Authority. TNCs and Congestion. 2018. 17 18 19
TRB 2019 Annual Meeting Paper revised from original submittal.
... This means that transit agencies that supplied more service did not see ridership increases. 8 The existing literature did not point to one significant single factor to explain the declines in 9 ...
... Watkins et al. also notes economic growth being 28 a potential contributor to ridership declines and particularly highlights low unemployment and lower gas 29 prices as potential causes (3). Graehler, Mucci and Erhardt also point to lower gas prices and increases in 30 car ownership as factors contributing to ridership declines (8). In their national rider survey, Transit 31 ...
... Erhardt's variable analysis suggests that bikeshare reduces bus trips but may complement heavy rail 33 trips-possibly because bus trips are typically shorter than rail trips, and "users may be more likely to 34 switch to bike share due to the similar distances served by both modes" (8). Bus trips are more directly 35 applicable in our rural investigation, since there are no rail modes available in rural transit districts in 36 ...
Conference Paper
Full-text available
Since 2013, ridership has declined for rural transit districts in Texas. In this study, we analyze the contributing causes of ridership loss in rural areas by looking into five years of Texas rural ridership data and analyzed datasets specific to Texas to test if nationally identified contributing factors might be causing rural ridership loss in Texas. For this purpose, we performed three types of analysis; analyzing the past five years of data from TxDOT Public Transportation Division for rural transit districts; potential contributing national factors to determine whether these factors were also observed within Texas rural transit districts; and a qualitative analysis of select rural transit districts. Overall, the research effort found that there is not one single root cause that is driving down ridership year after year. Instead, there appear to be many more nuanced and subtle causes. Although the past five years (2014 to 2018) have yielded a net loss of ridership, the past two years have seen ridership increases after removing outlier transit districts. In the rapidly changing transportation environment, transit must continue to adapt to meet local needs. Despite ridership trends, transit continues to be a vital mobility option in Texas and across the country.
... In 2017 alone, two (2) of the leading tech-based on-demand ridesharing services in the US (i.e., Uber and Lyft) estimated over 4.2 billion annual trips (Schaller, 2018;Johana, 2018;Kerr, 2018). Consequently, this booming adoption of ride-sharing services has on the one hand, stirred critics of these services to infer the induction of newer household trips which competitively absorb travel demand and potential revenue flows from existing transit and taxi services over time (Erhardt, et al., 2021;Acheampong, Siiba, Okyere, & Tuffour, 2020;Erhardt, et al., 2019;Doppelt, 2018;Lindsay, 2017;Shaheen, 2016;Rayle et al., 2016). ...
... Results from previous studies have differed substantially by location, demographics, and period. Some studies have considered ridesharing and its mode-substitution effects on personal vehicle ownership (Acheampong, et al., 2020;Ward et al., 2019;Grahn et al., 2019;Graehler et al., 2019;Dias et al., 2019;Doppelt, 2018;Lindsay, 2017;Rayle et al., 2016). Others have tried to relate their complementary impacts on active transportation as well as the first-and-last mile dilemma (Lavieri & Bhat, 2019;Yan et al., 2019;Brown, 2018;Circella & Alemi, 2018;Hall et al., 2018;Sadowsky & Nelson, 2017); adoption by class and demographics (Young & Farber, 2019;Sikder, 2019;Gehrke et al., 2018;Clewlow & Mishra, 2017); regulatory and legal framework (Flores & Rayle, 2017;Beer et al., 2017); differences, similarities and impacts on taxis (Hall et al., 2018;Rayle et al., 2016); and vehicle mile traveled (VMT) and parking (Wadud 2020;Henao and Marshall 2019). ...
... Certain studies even suggest that ride-hailing services tend to reduce rail and bus ridership by 1.29% and 1.7%, respectively, upon entering a US city (Schmitt, 2019;Graehler, et al., 2019). ...
... For example, Stewart and El-Geneidy (6) finds that 23% over Montreal's stops could be removed without noticably removing the system's coverage area. The stakes have been raised more recently because buses have been losing riders to ridehailing (7)(8)(9), which is more costly but faster than the bus. Therefore, many American cities have recently carried out campaigns of stop consolidation (or stop balancing (10)): the systematic practice of removing large numbers of stops. ...
Full-text available
Transit agencies have been removing a large number of bus stops, but discussions around the bus stop spacings exhibit a lack of clarity and data for comparison. This paper proposes new terminology and concepts for statistical consideration of stop spacings, and introduces a python package and open-source database which uses General Transit Feed Specification data to derive real-world stop spacing distributions
... Another study in four U.S. cities showed that the bike-sharing system was largely integrated with the public transit system (Kong, Jin, and Sui 2020). A separate study across major U.S. cities, however, highlighted that bike-sharing can negatively affect bus ridership (Graehler, Mucci, and Erhardt 2019). On a regional scale and relevant to this article, the relationship between bike-sharing and public transit in New York City has not been well elucidated. ...
Full-text available
The bike-sharing system has advanced urban transportation by solving “the last mile problem,” enabling riders to better connect to public transit. There has been a paucity of knowledge, however, regarding the relationship between bike-sharing and public transit. In this article, we solicit one year of bike trip data comprising approximately 17 million trips from Citi Bike, the largest dock-based bike-sharing system in New York City. Then, we derive six bike usage clusters based on three clustering variables: the start trips, end trips, and station empty status. Finally, we propose three relationships between bike-sharing and public transit: competition, integration, and complementation. The result demonstrates that bike-sharing can largely compete with public transit in New York City. A significant portion of bike-sharing trips are more time-intensive than their public transit alternatives. The article concludes that this competition exists due to riders’ preferences for lower costs and flexibility over savings in travel time, which helps to improve transportation equity for socioeconomically disadvantaged populations. Thus, in New York City, bike-sharing primarily fulfills the need for lowcost and flexible travel rather than solving “the last mile problem.” This revelation provides new insights into the roles of bike-sharing in urban transportation.
... This feature has led to an explosion of TNC trips (Johana 2018;Kerr 2018) and ridership (Conway et al. 2018;Grahn et al. 2020) during the last decade. However, the rapid rise of TNC use has also resulted in increased vehicle miles traveled (Henao and Marshall 2019), increased congestion , reduced transit ridership (Graehler et al. 2019), and equity concerns (Ge et al. 2020). During this same time, access to transit for populations living outside of the urban core has become increasingly difficult (Allen and Farber 2021). ...
Full-text available
First-mile last-mile (FMLM) mobility services that connect riders to public transit can lead to improved transit accessibility and network efficiency if such services are convenient and reliable. However, many current FMLM services are inefficient and costly because they are inflexible (e.g., fixed supply of shuttles) and do not leverage collected data for optimized decision making. At the same time, new forms of shared mobility can provide added flexibility and real-time analytics to FMLM systems when carefully integrated. This study evaluates performance and cost implications of public/private coordination between transit shuttles and transportation network companies (TNC) in the FMLM context. A real-time operations model was developed to simulate daily operations for an existing FMLM system using real-world demand data. Three supply strategies were tested with varying levels of flexibility: (1) Status Quo (two 23-passenger on-demand shuttles), (2) Hybrid (one 23-passenger on-demand shuttle + TNC), and (3) TNC Only (exclusively use TNC services). Results indicated that the added flexibility of the Hybrid service design (using shuttles and TNCs) improved service performance (a 7.7% improvement), reduced daily operating costs (− 6.0%), and improved service reliability (95th percentile travel times decreased by up to 40% during peak periods). In addition, the Hybrid service design was more robust to variations in demand. The Hybrid service was significantly cheaper to operate (− 31.6%) at reduced demand levels (50% of normal), and improved service performance (a 10.2% improvement) when demand levels were increased (150% of normal). These findings emphasize the importance of flexibility in FMLM service designs, especially when demand is sparse and variable.
Full-text available
The emergence of ridesharing services might complement or substitute public transit systems, leading to intricate relationships between the two services. However, limited studies focused on the nonlinear effects of ridesharing use frequency on public transit usage. Therefore, this paper investigated such nonlinear effects using the hierarchical negative binomial generalized additive model (HNBGAM), with the latest publicly available National Household Travel Survey (NHTS) dataset. The negative binomial and hierarchical negative binomial generalized linear models were also developed for comparison with the HNBGAM. The NHTS data involved travel information of 928 ridesharing users within 98 census tracts in San Diego. Two-level hierarchy (individual and census tract level) was constructed in the HNBGAM. In addition, the smooth function of the HNBGAM could help identify the nonlinear effects of ridesharing use frequencies on public transit usage. Socio-demographic factors (age, gender, race, household size, etc.) and built environment factors (e.g., population density, worker density, percentage of rental houses, and house unit density) were also considered in the modeling process. The findings revealed a negligible impact on public transit usage for occasional ridesharing use (from one to eleven times per month), a complementary effect for regular ridesharing use (from eleven to thirty-two times per month), and a substitution effect for active ridesharing use (more than thirty-two times per month). Understanding such nonlinear relationships could help policymakers make more informed decisions to avoid the over-substitution of public transit usage and better complement the public transport system.
Full-text available
The information and communication technology (ICT) plays an important role in improving energy consumption efficiency and reducing the emission level in the urban transport sector. ICT-based mobility services like ridesourcing provide smart tools and algorithms for matching travel demand and supply and more convenient door to door services. However, there is a concern that the convenience and competitive service fares of these new mobility modes encourage people to make more car travel or shift from more sustainable mobility modes like public transport to car travels. Therefore, it is necessary to study the frequency use and modal shift to this new mobility mode, particularly in the cities (like Cairo), where the ICT-based mobility services contain more ridesourcing of fossil fuel cars than other modes like online bikes/scooter sharing. A survey was conducted in Cairo, and logit models were developed to analyze the associations of socioeconomic, travel behavior variables with the frequency use and modal shift to ridesourcing. The results of ordinal logistic regression indicate that people who live near a metro station, with higher income, and with more non-work trips per week are more likely to be high-frequent users of ridesourcing. Moreover, women are more likely to use ridesourcing frequently than men in Cairo. The findings indicate that the most replaced mode by ridesourcing is traditional taxis (by 33 %), and the second and third shifted modes are private cars and public transport by 30% and 24 %, respectively. The results of multinomial logistic regression show that the socioeconomic parameters have significant associations with the probability of modal shift from public transport, taxis, and private cars to ridesourcing services.
This study explores how shared micromobility services can integrate with public transit through equitable payment structures to address first and last mile issues for light rail transit riders in Seattle, WA, and increase accessibility for low-income households. Seattle has recently permitted shared micromobility services such as e-scooter companies to begin operating alongside existing bikesharing services in the city. However, equity concerns have arisen as the users of bikeshare have been disproportionately white, affluent, and well-educated. To address these concerns, efforts have been made to reduce barriers to access and make these services more equitable to encourage their use among marginalized populations. Previous research has demonstrated evidence that these services can improve accessibility for disadvantaged populations such as low-income people of color. This research consists primarily of a literature review of relevant academic and gray literature, and a jurisdictional scan of cities in the U.S., Canada, Finland, and China. The objective of this research is to identify barriers to accessing shared micromobility services and synthesize existing best practices to propose solutions to make these services more equitable. Findings from this research then inform a set of recommendations for equitable payment integration in King County, which can also be generalized to other municipalities that are striving for equitable public transit and shared micromobility integration.
Full-text available
This research examines whether transportation network companies (TNCs), such as Uber and Lyft, live up to their stated vision of reducing congestion in major cities. Existing research has produced conflicting results and has been hampered by a lack of data. Using data scraped from the application programming interfaces of two TNCs, combined with observed travel time data, we find that contrary to their vision, TNCs are the biggest contributor to growing traffic congestion in San Francisco. Between 2010 and 2016, weekday vehicle hours of delay increased by 62% compared to 22% in a counterfactual 2016 scenario without TNCs. The findings provide insight into expected changes in major cities as TNCs continue to grow, informing decisions about how to integrate TNCs into the existing transportation system.
Full-text available
Public transport ridership has been steadily increasing since the early 2000s in many urban areas in North America. However, many cities have more recently seen their transit ridership plateaued, if not decreased. This trend in transit ridership has produced a lot of discussion on which factors contributed the most to this new trend. While no recent study has been conducted on this matter, understanding the levers that can be used to sustain and/or increase transit ridership is essential. The aim of this study is, therefore, to explore the determinants of public transport ridership from 2002 to 2015 for 25 transit authorities in Canada and the United States using a longitudinal multilevel mixed-effect regression approach. Our analysis demonstrates that vehicle revenue kilometers (VRK) and car ownership are the main determinants of transit ridership. More specifically, the results suggest that the reduction in bus VRK likely explains the reduction in ridership observed in recent years in many North American cities. Furthermore, external factors such as the presence of ridesourcing services (Uber) and bicycle sharing, although not statistically significant in our models, are associated with higher levels of transit ridership, which contradicts some of the experts’ hypotheses. From a policy perspective, this research suggests that investments in public transport operations, especially bus services, can be a key factor to mitigate the decline in transit ridership or sustain and increase it. While the results of this study emphasize that fare revenues cannot support such investments without deterring ridership, additional sources of revenues are required. This study is of relevance to public transport engineers, planners, researchers, and policy-makers wishing to understand the factors leading to an increase in transit ridership.
Full-text available
The major transit systems operating in San Francisco are San Francisco Municipal (MUNI), Bay Area Rapid Transit (BART), and Caltrain. The system of interest for this paper is MUNI, in particular the bus and light rail systems. During the past decade transit ridership in the area has experienced diverging growth, with bus ridership declining while rail ridership is growing significantly (Erhardt et al. 2017). Our data show that between 2009 and 2016, MUNI rail ridership increases from 146,000 to 171,400, while MUNI bus ridership decreases from 520,000 to 450,000. Direct ridership models (DRMs) are used to determine what factors are influencing MUNI light rail and bus ridership. The DRMs predict ridership fairly well, within 10% of the observed change. However, the assumption of no multi-collinearity is voided. Variables, such as employment and housing density, are found to be collinear. Fixed-effects panel models are used to combat the multi-collinearity issue. Fixed-effects panel models assign an intercept to every stop, so that any spatial correlation is removed. A transportation network company, Uber and Lyft, variable is introduced (TNC) to the panel models, to quantify the effect they have on MUNI bus and light rail ridership. The addition of a TNC variable and elimination of multi-collinearity helps the panel models predict ridership better than the daily and time-of-day DRMs, both within 5% of the observed change. TNCs are found to complement MUNI light rail and compete with MUNI buses. TNCs contributed to a 7% growth in light rail ridership and a 10% decline in bus ridership. These findings suggest that the relationship TNCs have with transit is complex and that the modes cannot be lumped together.
Full-text available
Transit direct ridership models (DRMs) are commonly used both for descriptive analysis and for forecasting, but are rarely evaluated for their ability to predict beyond the estimation data set. This research does so, using two DRMs estimated for rail and bus ridership in San Francisco. The models are estimated from 2009 data, applied to predict 2016 conditions, and compared to actual 2016 ridership. Over this period in San Francisco, observed rail ridership increased by 9% whereas observed bus ridership decreased by 13%. The results show that the models predict 2016 ridership about as well as that for 2009. The rail model correctly predicts the direction of change, but underestimates the magnitude of change. The bus model predicts the direction of change incorrectly, with a predicted 2% increase. A series of sensitivity tests are conducted to better understand the factors driving the ridership changes. These tests produce reasonable rail sensitivities, but reveal that the bus model is too sensitive to frequency, potentially because of the difficulty of estimating the coefficient from cross-sectional data when high-frequency transit also occurs in high-density locations. As the travel forecasting community increases its focus on empirically evaluating forecasts beyond a base year, DRMs must be a part of that.
Traffic congestion has worsened noticeably in San Francisco and other major cities over the past few years. This change could reasonably be explained by strong economic growth or other standard factors such as road and transit network changes. However, the worsening congestion also corresponds to the emergence of Transportation Network Companies (TNCs), such as Uber and Lyft, raising the question of whether the two trends may be related. Our research decomposes the contributors to increased congestion in San Francisco between 2010 and 2016, considering contributions from five incremental effects: road and transit network changes, population growth, employment growth, TNC volumes, and the effect of TNC pick-ups and drop-offs. We do so through a series of controlled travel demand model runs, supplemented with observed TNC data collected from the Application Programming Interfaces (APIs) of Uber and Lyft. Our results show that road and transit network changes over this period have only a small effect on congestion, population and employment growth each contribute about a quarter of the congestion increase, and that TNCs are the biggest contributor to growing congestion over this period, contributing about half of the increase in vehicle hours of delay, and adding to worsening travel time reliability. This research contradicts several studies that suggest TNCs may reduce congestion, and adds evidence in support of other recent empirical analyses showing that their net effect is to increase congestion. It is more data rich and spatially detailed than past studies, providing a better understanding of where and when TNCs add to congestion. This research gives transportation planners a better understanding of the causes of growing congestion, allowing them to more effectively craft strategies to mitigate or adapt to it.
How Uber affects public transit ridership is a relevant policy question facing cities worldwide. Theoretically, Uber's effect on transit is ambiguous: while Uber is an alternative mode of travel, it can also increase the reach and flexibility of public transit's fixed-route, fixed-schedule service. We estimate the effect of Uber on public transit ridership using a difference-in-differences design that exploits variation across U.S. metropolitan areas in both the intensity of Uber penetration and the timing of Uber entry. We find that Uber is a complement for the average transit agency, increasing ridership by five percent after two years. This average effect masks considerable heterogeneity, with Uber increasing ridership more in larger cities and for smaller transit agencies.
Recent technological advances have paved the way for new mobility alternatives within established transportation networks, including on-demand ride hailing/sharing (e.g., Uber, Lyft) and citywide bike sharing. Common across these innovative modes is a lack of direct ownership by the user; in each of these mobility offerings, a resource not owned by the end users’ is shared for fulfilling travel needs. This concept has flourished and is being hailed as a potential option for autonomous vehicle operation moving forward. However, substantial investigation into how new shared modes affect travel behaviors and integrate into existing transportation networks is lacking. This paper explores whether the growth in the adoption and usage of these modes is unbounded, or if there is a limit to their uptake. Recent trends and shifts in Uber demand usage from New York City were investigated to explore the hypothesis. Using publicly available data about Uber trips, temporal trends in the weekly demand for Uber were explored in the borough of Manhattan. A panel-based random effects model accounting for both heteroscedasticity and autocorrelation effects was estimated wherein weekly demand was expressed as a function of a variety of demographic, land use, and environmental factors. It was observed that demand appeared to initially increase after the introduction of Uber, but seemed to have stagnated and waned over time in heavily residential portions of the island, contradicting the observed macroscopic unbounded growth. The implications extend beyond already existing fully shared systems and also affect the planning of future mobility offerings.
The objective of this research is to quantify the impact that bikesharing systems have on bus ridership. We exploit a natural experiment of the phased implementation of a bikesharing system to different areas of New York City. This allows us to use a difference-in-differences identification strategy. We divide bus routes into control and treatment groups based on if they are located in areas that received bikesharing infrastructure or not. We find a significant decrease in bus ridership on treated routes compared to control routes that coincides with the implementation of the bikesharing system in New York City. The results from our preferred model indicate that every thousand bikesharing docks along a bus route is associated with a 2.42% fall in daily unlinked bus trips on routes in Manhattan and Brooklyn. A second model that also controls for the expansion of bike lanes during this time suggests that the decrease in bus ridership attributable to bikesharing infrastructure alone may be smaller (a 1.69% fall in daily unlinked bus trips). Although the magnitude of the reduction is a small proportion of total bus trips, these findings indicate that either a large proportion of overall bikeshare members are substituting bikesharing for bus trips or that bikesharing may have impacted the travel behavior of non-members, such as private bicyclists. Understanding how bikesharing and public transit systems are interrelated is vital for planning a mutually reinforcing sustainable transport network.