Best frenemies? A characterization of TNC and transit users

Abstract

The emergence of transportation network companies (TNCs) has created new options for travelers and fierce competition for taxis and public transportation (PT). While the literature focuses either on TNCs or PT users, we contrast individuals/households who use only PT, only TNCs, or both by estimating a cross-nested logit on 2017 NHTS data. We analyzed both individuals (for consistency with most of the literature) and households (to account for intrahousehold travel dependencies). Our results show that the unit of analysis (individuals vs. households) does not matter much for our dataset. We found that individuals/households who use either PT or TNCs or both share socio-economic characteristics, reside in similar areas, and differ from individuals/households who use neither transit nor TNCs. In addition, individuals/households who use both PT and TNCs tend to be composed of Millennials and Generation Z, with a higher income, more education, no children, and fewer vehicles than drivers. Our findings highlight the danger for PT of entering into outsourcing agreements with TNCs, neglecting captive riders, and further exposing choice riders to TNCs.

Full text

1
BEST FRENEMIES?
A characterization of TNC and transit users
Farzana Khatun and Jean-Daniel Saphores*
University of California, Irvine
Abstract
The emergence of transportation network companies (TNCs) has created new options for travelers and
fierce competition for taxis and public transportation (PT). While the literature focuses either on TNCs or
PT users, we contrast individuals/households who use only PT, only TNCs, or both by estimating a cross-
nested logit on 2017 NHTS data. We analyzed both individuals (for consistency with most of the
literature) and households (to account for intrahousehold travel dependencies). Our results show that
the unit of analysis (individuals vs. households) does not matter much for our dataset. We found that
individuals/households who use either PT or TNCs or both share socio-economic characteristics, reside in
similar areas, and differ from individuals/households who use neither transit nor TNCs. In addition,
individuals/households who use both PT and TNCs tend to be composed of Millennials and Generation Z,
with a higher income, more education, no children, and fewer vehicles than drivers. Our findings
highlight the danger for PT of entering into outsourcing agreements with TNCs, neglecting captive riders,
and further exposing choice riders to TNCs.
Keywords: Public Transportation; TNCs; Cross-Nested Logit (CNL); 2017 NHTS.

2
Introduction
Since 2009, the emergence of on-demand, door-to-door ride services from Transportation Network
Companies (TNCs) such as Uber and Lyft has created new and very popular mobility options that stirred
competition with other modes, especially taxis and public transportation. Building on their success, in
2014, Uber and Lyft launched UberPOOL and Lyft Line in selected metropolitan areas. These new
services allow travelers to share their rides with others at cheaper rates than UberX and Lyft Classic
(Alemi et al., 2018a). The overall expansion of their services and these additions have further fueled the
explosive growth of TNCs, which were estimated to have transported 2.61 billion passengers in 2017, up
37% from the year before (Schaller, 2018). While many have applauded the rise of TNCs, some have
raised concerns about their impact on public transportation (Malalgoda and Lim, 2019), traffic
congestion (Erhardt et al., 2019), air quality, and vehicle miles traveled (Alemi et al., 2018a; Schaller,
2018; Sperling, 2018), casting TNCs as a threat to the sustainability of urban transportation systems.
The reluctance of TNCs to share data publicly makes it difficult for policymakers and researchers to
analyze the impacts of TNCs on other modes, particularly transit. To circumvent this obstacle, we
analyzed data from the 2017 National Household Travel Survey (NHTS) to examine the claim that TNCs
are attracting riders who would have otherwise taken public transportation (or walked/biked, or not
traveled) (Alemi et al., 2018a; Schaller, 2018) by contrasting the characteristics of public transportation
users with those of TNC users.
While several papers have examined TNC users and the possible impacts of shared mobility on transit
(Blumenberg et al., 2016; Alemi et al., 2018a; Schaller, 2018), to the best of our knowledge, our paper is
the first to formally contrast transit and TNC users using multivariate models. A better understanding of
the differences between transit and TNC users should be useful to transit agencies tempted to
substitute TNCs for transit in areas where transit is declining or extend transit’s reach by contracting

3
with TNCs. Another contribution of this study is that we perform our analyses at both the individual and
the household levels. The latter accounts for intra-household dependencies of mode choice, which have
often been ignored in the transportation literature.
After reviewing selected papers that characterized transit and TNC users, we motivate our model
variables and summarize our modeling approach. We then discuss our results, summarize our
conclusions, mention some limitations of this work, and suggest future research directions.
Literature Review
In this section, we review selected papers that characterized transit users and TNC users. We focus on
studies conducted in the U.S. and Canada because of differences in context with other parts of the
world. Table 1 summarizes the papers discussed below.
Characteristics of Transit Users
One strand of the literature explored the characteristics of transit riders for different forms of transit
(bus, light rail, heavy rail, commuter rail) and the location of their residence (urban vs. suburban areas)
(Myers, 1997; Garrett and Taylor, 1999) while another strand distinguished between captive and choice
riders (Polzin et al., 2000; Krizek & El-Geneidy, 2007).
In the 1990s, researchers explored what transit service users selected based on their home location,
income, gender, and race. Garrett & Taylor (1999) reported that core city dwellers, who were primarily
low-income, female, non-Caucasian (predominantly African Americans), and young adults, relied more
on buses and light rail transit (LRT) than other demographic groups. In contrast, suburban riders chose
predominantly commuter rail, and they were primarily Caucasian, male, and members of higher-income
households (Garrett and Taylor, 1999). Other studies confirmed these findings (Myers, 1997) and
categorized riders into captive (i.e., people for whom transit is the only option) and choice groups (i.e.,
people who could use other modes, such as driving their own vehicle). For example, after analyzing data

4
from the 1995 NPTS, Polzin et al. (2000) found that captive riders were mainly composed of the elderly
and children, people with lower incomes, people with physical challenges, and families who either could
not meet their travel needs by driving or did not want to own cars. Conversely, choice riders were more
diversified and generally more affluent (Polzin et al., 2000).
The 2000s saw a plunge in transit ridership, especially in bus ridership, but passenger characteristics
remained mostly unchanged compared to the previous decade (LaChapelle, 2009; Taylor and Morris,
2015). This fall was associated with heavy investments in rail projects, which targeted more affluent
suburban choice riders, to the detriment of bus transit, which was continuing to serve primarily poorer
and minority communities (Taylor & Morris, 2015). In its investigation of transit riders during the 2000s,
the APTA’s 2007 report characterized its main patron group as adults, predominantly women, Caucasian
(for both rail and road modes but not for buses), employed, members of households with an annual
income between $25,000 and $49,999, and most likely to be composed of two-members with no motor
vehicles (Neff and Pham, 2007).
The profile of transit users has also received attention at the regional level (Kim et al., 2007; Krizek and
El-Geneidy, 2007). For example, Krizek & El-Geneidy (2007) investigated the habit and preferences of
“potential transit choice riders.” Their cluster analysis for the Twin City region led them to conclude that
choice riders care particularly about travel time, reliability, safety, convenience, and parking availability
near transit stations. After analyzing data from an on-board passenger survey in the St. Louis
Metropolitan area, Kim et al. (2007) concluded that females, African Americans, full-time students, and
middle-income people are more likely to use bus transit to reach Light Rail Transit (LRT) stations. While
these studies showed that the profile of transit users did not change much compared to the 90s, Brown
et al. (2016) reported that adults who prefer transit in their early years tend to shift to cars when they
get married and have children, which indicates a life cycle effect (Brown et al., 2016).

5
Clark's (2017) synthesis of passenger surveys from 163 transit systems spanning 2008 to 2015 provides a
profile of transit users just before and after the emergence of TNCs: during that period, transit users
were predominantly aged 25 to 54, disproportionately members of minority groups (especially for bus
users), and often (71%) employed. Moreover, many (54%) had access to at least one vehicle on a regular
basis, and they were slightly (55%) more likely to be female. Interestingly, households with annual
incomes under $15,000 or over $100,000 or more) made up a similar percentage (21% each) of transit
users. Moreover, a slight majority of transit users had a bachelor’s degree or a graduate education.
Grahn et al. (2019) reported mostly similar results from their analysis of 2017 NHTS data. Their findings
suggest that in 2017 transit users were younger, disproportionately Asian or African American, and less
likely to own a private vehicle. Moreover, those relying primarily on buses mostly had lower incomes,
while rail transit users were more likely to have higher incomes.
Although the U.S. and Canada have much in common, the profile of transit users differs in large cities on
both sides of the border because large groups of middle and upper-middle-class households still reside
in Canadian urban cores (Foth et al., 2013). Although Toronto has a transit system that strives to serve
disadvantaged communities (Foth et al., 2013), the 1996 Canadian census shows that 22% of commuters
used public transit (Kohm, 2000), partly to avoid expensive downtown parking (a feature shared with
several other large Canadian cities). Moreover, unlike in the U.S., transit ridership in Canada increased
between 2017 and 2018 (Hunt, 2019). One factor explaining the relatively good performance of transit
in some Canadian cities is that younger people use public transit to go to school. For example, Hasnine
et al. (2018) reported that female students who travel to downtown Toronto campuses use transit more
than those who travel to suburban campuses, possibly because transit services are not as convenient at
the outer edges of Toronto (Hasnine et al., 2018).

6
Characteristics of TNC Users
Several recent papers have characterized TNC users and their behavior (Alemi et al., 2017; Clewlow &
Mishra, 2017; Kooti et al., 2017; Leistner & Steiner, 2017; Alemi et al., 2018a-b; Grahn et al., 2019).
Some of these papers focused on TNC use among subgroups of the population. This is the case for Alemi
et al. (2018a), who analyzed a panel dataset of 1,191 millennials and 964 members of Generation X in
California to understand the factors that foster and hinder the use of TNCs and the impact of TNCs on
other modes. They found that millennials are more prone to using TNCs than their older counterparts
because arranging rides with TNCs is more convenient and requires less waiting, although their higher
cost may be a deterrent. Moreover, younger individuals, people in households without vehicles or with
fewer vehicles than drivers, and multimodal users tend to replace some of their transit trips with TNC
service. These findings are in line with those of other studies (Rayle et al., 2014; McDonald, 2015;
Circella et al., 2017) that focused on the travel behavior of millennials and the impact of emerging
technology on transportation. Alemi et al. (2017, 2018b) analyzed the same dataset to understand the
circumstances under which people are more likely to use TNCs (Alemi et al., 2017) as well as factors
explaining the adoption of TNC services and the frequency of their use (Alemi et al., 2018b). They
reported that land-use diversity and regional accessibility are associated with a greater likelihood of TNC
use; moreover, individuals who take more long-distance business trips (especially by plane) are more
likely to use TNCs (Alemi et al., 2018b). In addition, they found a correlation between land use diversity,
density, and the frequency of TNC use, but sociodemographic variables did not seem to matter much. As
expected, tech-oriented individuals who rely heavily on mobile apps are more likely to use TNCs, unlike
people with a strong preference for their own private vehicle (Alemi et al., 2018a).
A few other studies have explored some potential impacts of shared mobility on vehicle ownership and
mode preferences. Based on data from a survey conducted in seven major U.S. cities, Clewlow & Mishra
(2017) reported that TNC adopters have a lower level of vehicle ownership than non-adopters.

7
Moreover, they are more likely to own a private vehicle than core transit users. Overall, TNC users are
comparatively younger, more educated, have a higher income, and they tend to live in denser urban
environments. In addition, 9% of ride-hailing adopters disposed of their personal vehicle, and 26%
reduced their personal driving. Although reported changes in transit use were minimal, Clewlow &
Mishra (2017) suggested that ride-hailing can be a good substitute for bus transit and complement
commuter rail. These results are consistent with other sources (Shaheen et al., 2015; Henderson, 2017)
concerned with shared mobility and its impact on car ownership.
Similarly, Leistner & Steiner (2017) used descriptive statistics to explore the possibility of using Uber to
mitigate the travel challenges of older adults. They found that shopping and recreational trips are three
times faster on average with Uber than with transit, so they concluded that Uber may positively impact
the mobility of older adults.
Kooti et al. (2017) investigated the impact of dynamic pricing on Uber users’ participation and retention
by analyzing 59 million rides taken by 4.1 million users between October 2015 and May 2016. They
concluded that Uber riders tend to be more affluent than people who drive their own vehicles.
Moreover, younger people use this service more frequently but for shorter distances than older users,
and there appears to be gender parity among Uber riders.
Before the 2017 NHTS, the literature analyzed local, regional, or state data to characterize TNC users
(Chen, 2015; Rayle et al., 2016; Clewlow and Mishra, 2017; Alemi et al., 2018a-b; Hampshire et al.,
2019), and a nationwide understanding of these users was lacking (Sikder, 2019).
A couple of recent papers have analyzed data from the 2017 NHTS to paint a profile of TNC users
(Sikder, 2019; Grahn et al., 2019). After estimating an ordered logit model, Sikder (2019) found that
frequent TNC users (>= four rides over 30 days) are primarily male, younger, and have college/bachelor’s
degrees. They also tend to work full time but often have a flexible schedule, they have higher incomes,

8
and their households are more likely to have fewer vehicles than drivers. Conversely, African Americans
are less likely to take TNCs. Moreover, those who engage in car-sharing and bike-sharing and use public
transit are more likely to use TNCs, which suggests some complementarity between transit and TNCs.
Grahn et al. (2019) echoed these findings. To the best of our understanding, these two studies do not
appear to have considered whether the respondents analyzed had access to TNCs (they were not as
ubiquitous in 2017 as they are now), which might have impacted their results.
Another emerging strand of the literature has been exploring the impact of TNCs on public
transportation, but its conclusions are not clear-cut (Rayle et al., 2016; Clewlow and Mishra, 2017;
Sadowsky and Nelson, 2017; Hall et al., 2018; Malalgoda and Lim, 2019). Some studies, such as Rayle et
al. (2016) or Hall et al. (2018), concluded that TNC trips replaced some transit trips. For example, based
on over 2 million responses to intercept surveys, Rayle et al. (2016) found that at least half of TNC trips
in San Francisco replaced transit and driving trips. Clewlow & Mishra (2017) reported similar findings:
according to their analyses, TNCs are associated with a 6% drop in bus use and a 3% decrease in light rail
use. By contrast, Hall et al. (2018), who investigated the effect of Uber on public transit ridership in
several US metropolitan areas, reported that Uber complements transit and increased ridership by 5%
after two years. Likewise, after analyzing 2007-2017 data from the top 50 US transit agencies, Malalgoda
& Lim (2019) found that both bus and rail transit effectiveness (an index that measures transit service
quality based on the number of employees, vehicle operating hours, and fuel consumption) declined
between 2007 and 2017 and that TNC availability increased rail transit ridership in 2015. Furthermore,
rail transit effectiveness limited TNC availability, so overall, TNCs are neither complements nor
substitutes for bus transit. Finally, Grahn et al. (2019) reported that TNCs were primarily used for special
or rare events, with ~19% of TNC trips for social and recreational events, and that TNC users use public
transit at higher rates.

9
Table 1. Summary of selected studies on Transit and TNC users
Study
Data source and Method
Variables
Key findings
Characteristics of Transit Users
Garrett and
Taylor (1999)
• Review of secondary sources:
journals, reports, articles
• National Personal
Transportation Surveys (NPTS)
• American Public
Transportation Association
Demographic profile of transit
users.
Financial information about
transit (e.g., subsidies).
• Core city dwellers, who are primarily low-income,
female, non-Caucasian (mostly African Americans),
and young adults, rely more on buses and light rail.
• Suburban riders choose predominantly commuter
rail; they are primarily Caucasian and male, with
higher incomes.
Polzin et al.
(2000)
• 1995 NPTS
• Descriptive analysis
Transit captive and choice
riders’ transit use frequency,
population density, household
income, metropolitan statistical
area (MSA) categories, urban
classification, vehicle
ownership.
• Captive riders are mainly the elderly and children,
lower-income groups, people with physical
challenges, and families who either could not meet
their travel needs using cars or do not want to own
cars.
• Conversely, choice riders are diverse but are
generally more affluent.
Kim et al.
(2007)
• St. Louis Metropolitan area,
U.S.
• MNL
Socio-economic: age,
occupation, gender, race.
Mode: pick-up and drop off
option, bus, and walking
• People who take the bus to reach transit stations
are more likely to live in a commercial area and to
be female, African American, a full-time student,
and have a middle income ($15K-$24.9)
Krizek and El-
Geneidy
(2007)
• Twin cities: Minneapolis and
Saint Paul
• Transit users survey in 2001
and non-users survey in 1999
•
Factor and cluster analysis
Driver’s attitude, customer
service, transit service types,
reliability, value of travel time,
opinion about transit
cleanliness, comfort, safety.
• Choice riders value travel time, reliability, safety,
convenience, driver’s attitude, parking availability,
and other ride facilities near transit stations.
Neff and Pham
(2007)
• 2007 APTA report
• Onboard survey findings
• Descriptive Statistics
Age, race, income, gender,
education, driving license,
employment status, reasons for
choosing transit.
• Most likely public transportation stakeholders are
adult, women, Caucasian (for both rail and road
modes), households with an income between
$25,000 and $49,999, employed, and
predominantly two-members and zero-vehicle
households.

10
Taylor &
Morris (2015)
• 2009 NHTS, APTA, NTD and
primary survey of 50 transit
agencies
• Descriptive analysis
Age, race, income, vehicle
miles, number of unlinked
passenger trips, transit
subsidies.
• Lower income group, African Americans hold the
highest share among bus riders.
• Higher-income groups and Caucasians mostly
prefer commuter rail transit.
Brown et al.
(2016)
• 2001 and 2009 NHTS
• Smart Location Data (SLD)
from the U.S. EPA
• Cohort model and logistic
regression
Age, gender, race, ethnicity,
employment status, life cycle,
household size, residential
density, income, transit supply
index, birth cohort indicators.
• Young adults use transit services, but as they grow
older, they tend to shift from transit to cars due to
changes in family structure.
Clark (2017)
• APTA report (a compilation of
211 published reports of 163
transit systems)
• Descriptive statistics
Age, race, income, gender,
education, driving license,
employment status, reasons for
choosing transit
• Transit users are primarily female, 25-54,
employed, educated, minorities, and in both low
and high-income groups.
Characteristics of TNC users
Alemi et al.
(2017)
• Same dataset as (Alemi et al.,
2018b)
• One-way ANOVA and binary
logit model
65 attitudinal statements
related to land use, the
environment, technology,
government role, car
ownership, and frequency of
TNC use
• Land use diversity and centrality are positively
associated with greater TNC adoption.
• Long-distance travelers, particularly air travelers,
are more likely to use TNCs.
Leistner &
Steiner (2017)
• Pilot study conducted in
Gainesville, FL, to facilitate
the transportation needs of
older adults (60+)
• 40 adults completed 1,445
trips covering 8,119 miles.
•
Descriptive analysis
Sociodemographic: income
level, marital status, age,
gender, race, living
arrangements; travel
information: number of social,
shopping, medical and service
trips; trip cost, distance, & time.
• Primary use of traveling by Uber was shopping and
recreation.
• On average, these trips were three times faster
than similar transit trips.
• Uber may positively impact the mobility of older
adults and may be a feasible alternative to transit.
Clewlow &
Mishra (2017)
• Seven major U.S.
metropolitan areas
• 4,094 respondents: 2217
reside in dense, urban
neighborhoods and 1877 in
suburbs.
Travel attitudes, neighborhood,
technology, environment;
household demographics;
residential location; use of
shared mobility services, vehicle
ownership and preferences.
• TNC adopters have a lower level of vehicle
ownership than non-adopters, but they are more
likely to own a private vehicle than transit users.
• TNC users are younger, more educated, with a
higher income, and live in denser urban areas

11
Alemi et al.
(2018a)
• Online survey of 1,191
millennials and 964
Generation Xers.
• Quota-based sampling
approach of six major regions
in California.
Attitudes, preferences,
lifestyles, technology adoption,
residential location, commute
and non-commute travel,
vehicle ownership, frequency of
TNC use, demographic factors.
• Millennials are more likely to adopt and use TNCs.
• Uber/Lyft are user-friendly (less waiting time and
easy to arrange rides)
• Uber/Lyft can be substitutes for transit trips and
active mode trips.
Alemi et al.
(2018b)
• Same dataset as (Alemi et al.,
2018a)
• Ordered probit and zero-
inflated ordered probit model
Socio-demographic
characteristics.
Built environment; technology
adoption and use; travel
behavior; vehicle ownership.
• Land use diversity and density impact frequency of
TNC use.
• Tech-oriented individuals more likely to use TNCs.
• Individuals with a strong preference for private
vehicles are less likely to use TNCs frequently.
Hall et al.
(2018)
• 196 MSAs
• National Transit Database,
newspaper articles, press
releases, social media posts.
• Difference in differences.
Transit ridership, Uber entry
and exit, and a variety of
controls.
• Uber is complementary to transit and increases
ridership by 5%
Sikder (2019)
• 2017 NHTS
• Descriptive statistics and
ordered logit model
Personal: gender, age, student
status, ethnicity, education,
employment status, driver
status.
Household: drivers, workers,
income, vehicle ownership, size.
Land use: urban/rural; car share
and bike share programs;
transit use.
• Frequent TNC users (>= four rides over 30 days) are
mostly male, younger, college degree holders, full-
time workers with flexible schedules, and belong to
higher income and vehicle deficit households.
• African Americans are less likely to adopt TNCs.
• Those who participate in shared mobility (e.g., car
or bike share) and use public transit are more likely
to use TNCs -> complementary effect between
transit and TNCs.
Grahn et al.
(2019)
• 2017 NHTS
• Descriptive Statistics;
weighted and unweighted
linear regression
Age, education, income,
number of trips (walk, bike,
transit, TNC trips)
• TNC riders tend to live in urban areas; are most
likely to be younger, have an advanced degree, and
a higher income.
Malalgoda &
Lim (2019)
• 50 U.S. transit agencies (2007-
2017)
Rail transit effectiveness
• TNCs availability increased rail transit ridership in
2015
• TNCs are neither complement nor substitutes for
bus transit

12
Data and Methodology
Model Variables
The 2017 National Household Travel Survey (NHTS), which was administered between April 2016 and
April 2017, collected data from 129,969 households (Federal Highway Administration 2018). NHTS 2017
data were organized in four files: persons, households, vehicles, and trips.
In this paper, we analyzed answers to the following two questions:
1. “In the past 30 days, about how many days have you used public transportation such as buses,
subways, streetcars, or commuter trains?”
2. “In the past 30 days, how many times have you purchased a ride with a smartphone rideshare
app (e.g., Uber, Lyft, Sidecar)?”
To select our sample, we extracted respondents who stated that they have access to both transit and
TNCs if their motor vehicles are unavailable. The first question above targeted people 16 years old or
older, who hold a driver’s license, and whose household could access at least one motor vehicle. This
gave us 30,580 observations.
Since travel decisions routinely involve other household members, mode choices of household members
may not be independent, in which case it makes sense to select the household as the basic unit of
analysis. However, all the mode choice studies we found during our literature review were conducted at
the individual level (Buehler and Hamre, 2015; Alemi et al., 2017; Alemi et al., 2018a-b; Sikder, 2019).
Therefore, we conducted both a household-level analysis and an individual-level analysis, and we
contrasted the results of both.

13
Dependent variables
We built our dependent variables by combining data from the two questions mentioned above. For our
household-level analysis, we created four mutually exclusive groups to obtain our dependent variable
based on whether any household member older than 16 took public transportation or used a TNC during
the 30 days ending on their survey day:
• Group 1: at least one household member took public transportation, but none rode with a TNC.
• Group 2: at least one household member rode with a TNC, but none took transit.
• Group 3: some household members took transit, and some rode with a TNC; and
• Group 4: no household member over 16 took transit or rode with a TNC.
For our individual-level analysis, we used the same four groups to define our dependent variable except
that we considered only the mode choice of each respondent.
Explanatory variables
We selected our explanatory variables based on our literature review and the variables available in the
2017 NHTS dataset. From the person file, we retrieved information about age, race, Hispanic status,
educational attainment, the existence of a medical condition that could impair travel, working status
(from home, full-time, or part-time), and whether a respondent was born in the U.S. After aggregating
this information by household, we combined it with data from the household file: household income,
lifecycle variables, the number of household drivers and vehicles, and homeownership.
Many studies (McDonald, 2015; Circella et al., 2017; Alemi et al., 2018a-b) have considered generations
instead of age for exploring how different formative experiences interact with people’s life-cycle and
aging to shape travel behavior. We relied on definitions from the Pew Research Center (2018) to create
our generation variables (birth years are in parentheses): Generation Z (1997 to 2001), who therefore

14
were between 16 and 20 years old in 2017; Generation Y (Millennials) (1981 to 1996); Generation X
(1965 to 1980); Baby Boomers (1946 to 1964); and the Silent Generation (born before 1946). For our
household models, a generation variable equals one if at least one household member belongs to that
category and 0 otherwise; there is no baseline. For our individual-level models, generation variables are
binary and capture the generation of the respondent; the Baby Boomer category serves as a baseline.
The literature also suggests that household educational attainment plays a pivotal role in daily mode
choice (Buehler and Hamre, 2015; McDonald, 2015; Alemi et al., 2017; Circella et al., 2017; Clark, 2017).
To capture the level of education of a household, we created five binary variables that reflect the
highest level of education of household adults based on the categories available in the 2017 NHTS.
Race and Hispanic status may matter for selecting a transportation mode (Buehler and Hamre, 2015;
Clark, 2017). For our household analysis, a binary household race variable equals one if all household
members identify as belonging to that race and zero otherwise. The “mixed” category captures the
remaining households. Hispanic status was defined similarly. We also created binary variables for
whether a respondent was born in the U.S., medical condition, and working status.
In addition, our models include common household variables such as the number of workers, household
size, household structure, annual household income, and vehicle ownership, which have all been found
to matter for explaining travel preferences (Buehler and Hamre, 2015; McDonald, 2015; Clark, 2017;
Alemi et al., 2018a-b). To capture household structure, we retained five variables from the 2017 NHTS.
To represent annual household income, we collapsed the eleven categories from the 2017 NHTS into
five binary categories (see Tables A1 and A2). Homeownership is captured by a binary variable, and
household size by a count variable.
As the decision to take transit or a TNC may depend on whether a household has more drivers than

15
vehicles, we created a binary variable that equals one if a household has more drivers than vehicles and
0 otherwise.
Finally, we added five binary variables that reflect the frequency of smartphone use (daily, weekly,
monthly, yearly, and never) since TNCs rely crucially on smartphone apps.
It is well-known that land use is correlated with mode choice (Buehler and Hamre, 2015; Alemi et al.,
2017; Alemi et al., 2018a-b). Unfortunately, the 2017 NHTS does not provide the location of residences
or places of work, but it includes some common land-use variables. We used population density (1,000
persons/sq. mile) of the home census tract of households in our sample.
To understand how the inclusion of TNCs may have impacted the patronage of different forms of transit,
we created three binary variables to capture the availability of bus, light rail, and heavy rail services for
the households located in a core-based statistical area (CBSA). A CBSA is a smaller geographic unit than
Metropolitan Statistical Area (MSA), with at least 10,000 people and an urban center. The 2017 NHTS
reports information about 53 CBSAs. For each, we gathered information about the availability of bus,
light rail, and heavy rail transit from the APTA, which publishes quarterly reports on ridership by transit
type for primary cities under the jurisdiction of transit organizations in the U.S. We then added this
information to our dataset in the form of binary variables.
For our individual-level models, generation, race, Hispanic status, educational attainment, medical
condition, working status (from home, full-time, or part-time), and the binary variable for “not born in
the U.S.” characterize individual respondents in our sample. Other variables (e.g., household income)
are defined the same as in our household models.
Summary statistics for our model variables, which document their variability, are shown in Tables A1
and A2 in the appendix.

16
Econometric Framework
Nesting structures
For simplicity, we first estimated a Multinomial Logit (MNL) model, which is often a starting point for
modeling mode choice. We tested for the Independence of Irrelevant Alternatives (IIA; see Train, 2009),
but it does not hold here (see the beginning of the results section).
A nested logit (NL) model relaxes the IIA requirement for modes in different nests (Train, 2009). To
select our preferred nested logit model, we experimented with several structures organized around two
nests (public mobility and private mobility), where Group 1 (PT but not TNCs) is in the public mobility
(PT) nest, and Group 4 (Neither PT nor TNCs) is in the private mobility nest, while Groups 2 (TNCs but
not PT) and 3 (Both PT and TNCs) are allocated to either nest in different models. We selected our
preferred NL structure based on AIC and BIC (lower is better) and the requirement of consistent log-sum
parameters (values between 0 and 1) (Koppelman & Bhat, 2006).
A Cross-Nested Logit (CNL) model further relaxes the NL requirement that an alternative belongs to only
one nest. Although uncommon, several mode choice studies have estimated CNL models (Vovsha, 1997;
Ermagun and Levinson, 2017; Hasnine et al., 2018). Here, we estimated two CNL models with Group 2
(“TNCs but no PT”) either in the public mobility nest with Group 1 (“PT but not TNCs”), or in the private
mobility nest with Group 4 (“Neither PT nor TNCs”), and Group 3 (“both PT and TNCs”) in both nests. We
selected our preferred CNL model based on AIC and BIC (again, smaller is better) and suitable log-sum
parameter values.
The CNL model
In a CNL model, an alternative can belong to more than one nest (Train, 2009). The extent to which
alternative j belongs to nest k is given by the allocation parameter αjk≥0.so allocation parameters sum to

17
one over nests for a given alternative, i.e., ∑k αjk = 1 and αjk=0 indicates that alternative j does not belong
in nest k. (Train, 2009).
A second type of parameter plays an important role here: log-sum parameters. Denoted by λk ≥ 0, the
log-sum parameter for nest k reflects the degree of independence among alternatives within nest k,
where a larger value indicates greater independence and less correlation. Log-sum parameter values
between 0 and 1 guarantee consistency with utility maximization, although that property may still hold
for a range of alternatives when log-sum values are above one (Train, 2009).
A choice model is defined by the expression of the probability for alternative “i” available to decision-
maker “n” (here, individual/household “n”). For the CNL, this expression is given by (Train, 2009):
()
()
1
1/ 1/
1/
1
() ( )
()
k
nj
ni k k
k
l
nj l
l
V
V
ik jk
k jB
ni KV
jl
l jB
ee
P
e
λ
λλ
λ
λ
αα
α
−
∈
= ∈
=∑∑
∑∑
, (1)
where Vni is the representative utility of alternative “i” for decisionmaker “n.” For all our models,
∀ ∈ {1,2,3},  =0 +�  ∗
−1
=1
(2)
and Vn4=0 for identification since only differences in utility matter (and can be estimated). In the above,
the
β
iks are unknown coefficients and the xkns are explanatory variables characterizing decisionmaker n
(socio-economic and land use characteristics).
If each alternative enters only one nest, the αjk parameters are either 0 or 1, and the CNL model
simplifies to a NL model. If the log-sum parameters λk of a NL model all equal 1, then the NL model
reduces to a MNL model.
We estimated unknown model parameters via maximum likelihood.

18
Results
Model selection
We estimated our models with Stata 15 and BisonBiogeme (Version 2.6a) (Bierlaire, 2020). A check for
multicollinearity showed that it is not an issue here since the maximum VIF value is <7.1.
Our final MNL, NL, and CNL structures are shown in Figure 1. We first estimated MNL models (Panel A).
To test the Independence of Irrelevant Alternatives (IIA), we ran Hausman tests, but they returned
negative values, which is at odds with their asymptotic χ2 distribution (Vijverberg, 2011). We then ran
Suest tests, which rejected the IIA for both the household and the individual MNL models.
Among the NL structures we explored, only the one shown on Panel B of Figure 1 gave consistent log-
sum parameters (values between 0 to 1; Koppelman & Bhat, 2006). The NL structure with Groups 2 and
3 in the private mobility nest had inconsistent log-sum parameters (values greater than 1). The NL
structure with Groups 1 and 2 in the public mobility nest and Groups 3 and 4 in the private mobility nest
did not converge.
We retained the CNL structure of Figure 1 for its significant log-sum and allocation parameter values and
lower AIC and BIC values. This CNL structure has lower values of AIC and BIC than our preferred NL and
MNL models for both individual-level and household-level analyses (see Table 2). Likelihood ratio (LR)
tests also supported our selected CNL structure over the corresponding Nested Logit structure (Panel C
of Figure 1) with LR test values at the individual and household levels of 39,767.4 (p-values < 0.01) and
29,210.7 (p-values < 0.01), respectively.
To save space, parameter estimates for our preferred MNL and NL models are not included here. In the
rest of this section, we focus on our CNL results.

19
Panel A: Multinomial Logit (MNL) structure
Panel B: Nested Logit (NL) structure
Panel C: Cross Nested Logit (CNL) structure
Figure 1 Structure of preferred MNL, NL, and CNL models
Modes used in the past 30 days
Group 1:
Public
transportation
but not TNCs
Group 4:
Neither public
transportation
nor TNCs
Group 2:
TNCs but not
public
transportation
Group 3:
Both public
transportation
and TNCs
Modes used in the past 30 days
Public Mobility
Private Mobility
Group 2:
TNCs but not
public
transportation
Group 3:
Both public
transportation
and TNCs
Group 1:
Public
transportation
but not TNCs
Group 4:
Neither public
transportation
nor TNCs
Modes used in the past 30 days
Public Mobility
Group 1:
Public
transportation but
not TNCs
Private Mobility
Group 3:
Both public
transportation
and TNCs
Group 4:
Neither public
transportation
nor TNCs
Group 2:
TNCs but not
public
transportation

20
Table 2 Selected Measures of Fit
Preferred model
Multinomial logit
Nested logit
Cross-nested logit
Individual-level models (N= 30,580)
Null Log-Likelihood
-27701.7
-27701.7
-27701.7
Final Log-Likelihood
-22532.5
-22,531.9
-22509.2
McFadden ρ2
0.187
0.187
0.187
AIC
45,311.0
45,311.9
45,274.4
BIC
46,335.3
46,516.4
45,592.5
Household-level models (N=23,947)
Null Log-Likelihood
-22,932.8
-22,932.8
-22,932.8
Final Log-Likelihood
-18,616.5
-18,615.7
-18,592.2
McFadden ρ2
0.188
0.188
0.189
AIC
37,484.9
37,485.5
37,446.3
BIC
38,503.4
38,688.2
36,610.6
CNL results
Results for our preferred CNL household-level and individual-level models are shown in Table 3. Except
for the generation variables, which are defined differently in these two models (see text below or the
footnotes below Table 3) and therefore cannot be compared directly, our t-tests showed that only a few
coefficients differ significantly (at 5%) between these two models (exceptions are “less than high school”
for Groups 1 and 3, “Work for home” for Group 2, and the log of population density for Group 2), which
may reflect that almost half of our respondents who took transit did so less than once a week, and
nearly 60% of those who took TNCs also did so less than once a week.
Let us start with the allocation parameters for the overlapping group (Group 3: both PT and TNCs in the
past 30 days). For our individual model, their values are 0.655*** and 0.345***, so 65.5% of the
overlapping group utility comes from the public mobility nest and 34.5% from the private mobility nest.
These values differ slightly for the household model, with values of 70.8% and 29.2%, respectively.
Furthermore, the log sum parameters for both models are within the required range and significant,
which suggests that the nesting structures for both analyses are valid.
In the following two sections, we discuss our results for each model.

21
TABLE 3: Results for preferred CNL models
Individual model (N=30,580)
Household model (N=23,947)
Group 1:
PT but not
TNCs
(N1=3052)
Group 2:
TNCs but
not PT
(N2=3004)
Group 3:
Both PT and
TNCs
(N3=2597)
Group 1:
PT but not
TNCs
(N1=2574)
Group 2:
TNCs but not
PT
(N2=2485)
Group 3:
Both PT and
TNCs
(N3=2342)
Socio-demographic and economic attributes
Generation♣
(1 if generation of the respondent)
(1 if has at least o household member)
Generation Z
0.776***
0.594***
0.984***
0.414***
0.188***
0.507***
Generation Y (Millennials)
0.391***
0.539***
0.582***
0.455***
0.242***
0.561***
Generation X
0.091**
0.250***
0.185***
0.089
0.011
0.101
Baby Boomers
Baseline
0.050
-0.180***
-0.020
Silent Generation
-0.242***
-0.143***
-0.265***
-0.238
-0.294***
-0.316***
Hispanic status (Hispanic =1)
-0.137**
0.032
-0.108*
-0.130
0.049*
-0.103
Ethnicity (baseline = Caucasian)
African American
-0.117*
-0.102***
-0.145**
-0.103
-0.059*
-0.134*
Asian
-0.086
-0.087**
-0.109*
-0.116
-0.064*
-0.129*
Mixed
0.174***
0.056*
0.163***
0.072
0.052*
0.057
Educational attainment (baseline = some college or associate degree)
Less than high school
-0.206
-0.283***
-0.266
0.216
-0.133
0.266
High school
-0.251***
-0.187***
-0.287***
-0.121
-0.160***
-0.178*
Undergraduate degree
0.460***
0.131***
0.459***
0.459***
0.106***
0.450***
Graduate or professional degree
0.675***
0.125***
0.677***
0.696***
0.114***
0.701***
Annual household income (baseline= $75,000 to $124,999)
<$35,000
-0.063
-0.214***
-0.105*
-0.114*
-0.183***
-0.153**
$35,000 to $74,999
-0.215***
-0.155***
-0.232***
-0.239***
-0.132***
-0.260***
$125,000 to $199,999
0.312***
0.174***
0.351***
0.300***
0.142***
0.339***
>=$200,000
0.557***
0.431***
0.652***
0.549***
0.342***
0.647***
Household structure (baseline = 2+ adults with children)
One adult, no children
0.454***
0.230***
0.465***
0.320***
0.170***
0.320***
2+ adults, no children
0.159***
0.138***
0.176***
0.136*
0.115***
0.147**
One adult, some children
0.235**
0.059
0.277***
0.134
0.041
0.167

22
One retired adult, no children
0.308***
0.135*
0.340***
0.140
0.077
0.152
Two+ retired adults, no children
-0.099
0.012
-0.060
-0.157*
0.033
-0.118
Household size
-0.105***
-0.063***
-0.120***
-0.089***
-0.046***
-0.108***
Household owns home
-0.208***
-0.142***
-0.263***
-0.192***
-0.102***
-0.248***
Household workers (baseline = household with no workers)
Household with one worker
-0.018
0.033
-0.019
-0.020
0.030
-0.019
Household with two workers
0.093
0.005
0.079
0.106
0.014
0.103
Household with three or more workers
0.146
0.023
0.160
0.210*
0.070
0.246**
Works from home (Yes=1)
-0.039
0.198***
0.031
-0.016
0.131***
0.037
Works full time
0.034
0.052*
0.063
-0.005
0.024
0.025
Works part-time
-0.018
-0.020
-0.030
0.011
0.001
-0.0002
Household has fewer vehicles than drivers (Yes=1)
0.873***
0.120***
0.862***
0.903***
0.104***
0.896***
Has medical condition (Yes=1)
-0.220***
-0.160***
-0.223***
-0.19**
-0.089**
-0.208***
Not born in the U.S. (Yes=1)
0.019
-0.019
0.025
0.076
-0.020
0.069
Smartphone use (baseline = daily)
Weekly
0.087
-0.278***
0.012
0.055
-0.204***
-0.028
Monthly
-0.004
-0.246***
-0.114
-0.024
-0.216***
-0.161
Yearly
0.055
-1.010***
-0.114
0.035
-0.756***
-0.120
Never
-0.299***
-0.621***
-0.526***
-0.316***
-0.486***
-0.528***
Land use
Ln of population density (1,000/mi2)
0.141***
0.087***
0.151***
0.140***
0.065***
0.152***
Availability of transit services
Household in a CBSA with bus service
0.149**
0.120***
0.165***
0.138**
0.088***
0.161**
Household in a CBSA with light rail service
0.247***
0.086***
0.291***
0.267***
0.085***
0.302***
Household in a CBSA with heavy rail service
0.988***
0.025
0.965***
1.010***
0.012
0.996***
Constant
-2.660***
-1.130***
-2.370***
-2.540***
-0.696***
-2.240***
Log-sum parameters
Public mobility nest
0.114***
0.111***
Private mobility nest
0.342***
0.270***
Allocation Parameters
Public mobility nest
0.655***
0.708***
Private mobility nest
0.345***
0.292***

23
1. ***, **, and * indicate p-values < 0.01, < 0.05, and < 0.1, respectively.
2. For the individual-level model, generation, educational attainment, Hispanic status, worked from home, worked full-/part-time, medical
condition, race, and US birth status pertain to each respondent.
3. For the household-level model, generation, educational attainment, Hispanic status, worked from home, worked full/part-time, medical
condition, and immigration status indicate that at least one household member has these attributes. For race, all household members are of
that race.
4. The base for both models is Group 4, where N4 = 21,927 for the individual-level model and N4 = 16,546 for the household-level model.
♣: generation variables are defined differently in the two models above. For the household model, a generation variable equals one if at least
one household member belongs to that generation and 0 otherwise. Since the generation variables, in this case, are not mutually exclusive (a
household can have members in different generations), there is no need for a baseline. For the individual model, generation variables are
binary and capture the generation of the respondent; the Baby Boomer category serves as a baseline.

24
Individual-level results
From Table 3, we see that compared to Baby Boomers, Generation Z respondents are more likely to use
only transit (0.776***), only TNCs (0.594***), or both (0.984***) rather than drive only. Results are
similar for millennials and Gen X respondents, with smaller coefficients for older respondents.
Conversely, members of the Silent generation are less likely to use either only transit (-0.242***), only
TNCs (-0.143***), or both (-0.265***) than to drive. These results confirm findings from McDonald
(2015) and Blumenberg et al. (2016), who reported that Millennials (along with Generation Z) tend to
drive less, own fewer vehicles, and rely more on other modes. These differences can be explained by
their preferences, economic status, and life cycle stage (McDonald, 2015; Blumenberg et al., 2016). In
contrast, Baby Boomers and Silent Generation members are less likely to use TNCs. A possible
explanation is that Uber and Lyft vehicles are typically not equipped to easily serve senior citizens or
people with mobility impairments. Another reason could be the digital divide, as older generations are
not as comfortable as younger generations using communication technology to hail rides (Rahman et al.,
2016; Jamal and Newbold, 2020).
Hispanic status and race play a (limited) role here. Hispanics appear less likely to take transit (-0.137**)
or both PT and TNCs (-0.108*) than to drive only (recall that all respondents in our sample have access
to a motor vehicle). Compared to Caucasians, both African Americans (-0.102***) and Asians (-0.087**)
are less likely to use TNCs than to drive only, possibly due to racial bias. Indeed, recent studies have
shown that African Americans face higher cancellation rates from TNC drivers, suggesting racial
discrimination (Ge et al., 2016). These two groups are also less likely to use both (-0.145** for African
Americans and -0.109* for Asians). Conversely, members of mixed-race households are more likely than
Caucasians to take transit only (0.174***), TNCs (0.056*), or both (0.163***).

25
Education also matters. Individuals with less than a high school education do not differ from the
baseline (individuals with some college or an associate degree), except for using TNCs only (-0.283***).
However, individuals with a high school degree are less likely to use either transit (-0.251***), or TNCs (-
0.187***), or both (-0.287***) than to drive, compared to the baseline. Conversely, individuals with
either undergraduate or graduate degrees are more likely to use either PT only (0.460*** and 0.675***
respectively), TNCs only (0.131*** and 0.125*** respectively), or both (0.459*** and 0.677***
respectively) than to drive, possibly because they live in more affluent areas that offer both services.
These results are consistent with findings in Clark (2017), who reported that people with advanced
degrees prefer rail transit because it is more comfortable, environment friendly, and congestion-free.
Results for household income reinforce those for education. Compared to the baseline (individuals with
an annual household income ranging from $75,000 to $124,999), people in the lower-income groups are
less likely to take only PT than to drive (<$35,000: -0.214***). To put this result in perspective, recall
that everyone in our sample has access to a motor vehicle. Members of the lower-income groups are
also less likely to use either only TNCs (<$35,000: -0.214***; $35,000 to $74,999: -0.155***) or both
public transportation and TNCs (<$35,000: -0.105*; $35,000 to $74,999: -0.232***). The opposite holds
in the two higher income brackets, with higher coefficient values for the highest income group. The
explanation for this result is the same as for educational attainment (Clark, 2017).
As expected, family structure influences mode choice. We see that individuals with or without children
are more likely to depend less on their cars and more on either public transportation only (0.454*** for
one adult only, 0.159*** for two or more, 0.235** for one adult with some children), TNCs only
(0.230*** for one adult, 0.138*** for two or more), or both (0.465*** for one adult, 0.176*** for two
or more, and 0.277*** for one adult with some children). When families get larger and have children,
they often have more constrained schedules, so they rely more on their motor vehicles to fulfill their
daily travel needs (Buehler and Hamre, 2015). We also found that retired adults are more likely to use

26
either public transportation (0.308***), TNCs (0.135*), or both (0.340***) than to drive, compared to
the baseline (2+ adults with children). This is again likely due to the driving restrictions of older adults.
As household size increases, people rely increasingly less on modes other than their cars (-0.105***, -
0.063***, and -0.120*** for PT only, TNCs only, and both, respectively). Households who own their
home are also less likely to use PT or TNCs (-0.208***, -0.142***, and -0.263*** respectively), likely
because homeowners tend to live in suburban areas where access to PT and TNCs is less convenient.
The number of workers in the household does not matter here. Moreover, individuals who worked from
home are more likely to take only TNCs (0.198***) than just drive only, possibly because they may not
have a driver’s license. Indeed, coefficients of the variable that indicates if a household has fewer
vehicles than drivers are all positive and highly significant (PT only: 0.873***; TNCs only: 0.120***; Both:
0.862***).
As expected, adults with a physical impairment that limits their mobility are less likely to depend on
transit, TNCs, or both than on their own vehicles (-0.220***, -0.160***, -0.223*** for Groups 1, 2, or 3
respectively). Where people were born does not influence their mode choices in this model. Moreover,
those who do not use smartphones daily are less likely to use TNCs than those who do.
Land use also plays a role here. As expected, individuals who reside in denser areas tend to use more
varied modes (0.141***, 0.087*** and 0.151*** for Group 1, 2, and 3 respectively); the large positive
and significant coefficient of Group 3 (individuals who used both public transportation and TNCs)
reflects the overlap between PT and TNC users. Indeed, Uber and Lyft are primarily present in denser
urban environments that typically also harbor well-developed public transportation networks. The
availability of transit services in a CBSA area tells a similar story. Individuals who reside in a CBSA with
bus, light rail, or heavy rail services use a wider variety of modes (coefficients for all three categories are

27
positive and significant) than those who live in a CBSA without these services (Alemi et al., 2017; Alemi
et al., 2018b).
Household-level results
In this sub-section, we highlight differences between our preferred CNL models at the individual level
(discussed above) and at the household level. Let us start with generation variables, keeping in mind
that a generation variable equals one if at least one household member belongs to that generation and
0 otherwise. First, we note that the Gen X coefficient is not significant for any household group here.
Second, a negative value (-0.180***) indicates that households with Baby Boomers are less likely to use
TNCs, which is even more the case (-0.294***) for households with members from the Silent generation.
The latter are also less likely to use both PT and TNCs (-0.316***). These findings align with our
individual-level results.
Interestingly, Hispanic households are slightly more likely to use TNCs (0.049*) than non-Hispanics,
which was not the case in our preferred individual CNL model. Results are mostly unchanged for race
variables, although they are typically smaller than for the individual-level model: African Americans and
Asian households are less likely to belong to Group 2 (TNCs but not PT) or 3 (both PT and TNCs) than
Caucasian households.
Both models have significant coefficient values for the top two education and income categories
(advanced degree holders and high-income groups), with similar magnitudes and significance. For the
bottom two categories, however, the household-level coefficients don’t differ statistically from the
baseline except for households with a high school education, who are less likely to take TNCs (-
0.160***). Another difference is that households in the lower-income tier (<$35,000) are slightly less
likely to use public transportation (-0.114*), which may seem surprising until we remember that all
households in our sample have access to motor vehicles. That coefficient was not significant for the

28
individual-level model. The other household income variables have the same sign, similar magnitudes,
and similar significance levels.
For household structure variables, two categories (one adult with some children, and one retired adult
without children) lose their significant differences with the baseline (2+ adults with children), but
households with 2+ adults and no children are now less likely to belong to Group 1 (PT but not TNCs).
Other household structure results are similar between the two models, and the same applies to
household size and homeownership.
As for the individual model, the number of workers in the household does not matter except that
households with three or more workers are more likely to use transit (0.210*) or both transit and TNCs
(0.246**) than drive only, possibly because households with more adults are more likely to have at least
one adult without a driver’s license.
For the remaining variables, coefficient magnitude, sign, and significance are similar to those for the
individual-level model.
Discussion and Conclusions
In this paper, we contrasted individuals/households who use public transit (PT), TNCs, and both by
analyzing mode use data collected in the 2017 NHTS. We defined four mutually exclusive categories of
individuals/households and estimated Cross Nested Logit models. To the best of our knowledge, this is
the first nationwide study to contrast public transit and TNC users to understand the potential impact of
TNCs on transit. A second contribution of our study is our comparison between individual and
household-level models to account for intra-household dependencies of mode choice, which we found
to have little impact here because many 2017 NHTS respondents only used TNCs and transit sparingly.

29
To stem the ridership decrease, transit agencies across the U.S. have been forming partnerships with
Uber and Lyft to compensate for abandoned lines, address first and last-mile gaps, and offer service to
night workers. For example, in 2016, San Clemente (in south Orange County, California) implemented a
subsidized Lyft service to recapture some of the riders lost to the closure of two bus routes (191 and
193) (Swegles, 2016). The goal was to provide on-demand service with special considerations for
shopping trips for riders 60 and over. While the COVID-19 pandemic is partly responsible for the failure
of this and similar initiatives in Southern California (Pho, 2020), our results suggest that they were
unlikely to succeed because Baby Boomers and Silent Generation individuals/households, as well as
individuals with physical challenges or households with members with impaired mobility, are less likely
to use TNCs, especially if they live in lower-density areas. Indeed, vehicles in use by TNCs typically are
not equipped to accommodate customers with impaired mobility. Moreover, many older adults avoid
such services because they are not comfortable using smartphones and because of discriminatory
practices towards older citizens by some TNC drivers (Williams, 2021).
Our results show that transit and TNCs target individuals/households who share common socio-
economic characteristics and live in similar (higher density) areas. These groups are more likely to be
Millennials and belong to Gen Z, with higher incomes, advanced degrees, no children, and fewer vehicles
than driver’s license holders. They reside in denser areas and CBSAs served by PT and now TNCs.
Compared to PT, TNCs provide much more convenient and typically much faster point-to-point service,
which this group of individuals/households is likely to be able to afford, so increasing the exposure of
these individuals/households to TNCs may hasten their exodus to TNCs.
Instead of outsourcing to TNCs, transit agencies should consider exploring partnerships with micro-
mobility operators to extend the reach of transit and take care of the first- and last-mile problem.
Multimodal connectivity with bike-sharing and micro-mobility has been adopted in countries around the
world, but the U.S. is lagging (Mohiuddin, 2021), even though these mobility options could potentially

30
replace cars for up to 30% of trips under five miles, which make up more than half of all trips in the U.S.
(Abduljabbar et al., 2021). Recent studies have shown that well-educated, younger adults, childless
households, upper-income households, and urban dwellers with multiple mode options favor micro-
mobility (Shaheen and Cohen, 2019), so partnerships with transit where micro-mobility stations are
conveniently located by PT stops, and seamless payment options (such as apps integrating transit and
micro-mobility) may help transit recover as it emerges from the pandemic. Embracing this approach may
also enhance public health and help achieve GHG reduction goals.
Many low-income individuals/households (often belonging to disadvantaged groups) also reside in core
urban areas and CBSAs served by transit. However, we found that these individuals/households are less
likely than higher-income groups to take both TNCs and PT. The lower use of TNCs by less affluent
individuals/households is unsurprising since it is typically not the cheapest transportation option.
Extensive partnerships between transit and micro-mobility providers could prove attractive to them if
pricing is right, and if micro-mobility stations are in secure areas in minority neighborhoods. We note
that African American and Asian individuals/households are also less likely to use TNCs (all else being
equal), which suggests racial discrimination as uncovered in other studies (e.g., see Ge et al., 2016).
Our results also show that lower-income individuals/households are less likely to use PT, which seems at
odds with the disproportionate use of bus transit by lower-income individuals/households. The reason
for this apparent discrepancy is that all individuals/households in our sample have access to motor
vehicles because the NHTS question analyzed in this paper was restricted to motorized respondents, so
none of the individuals/households in our datasets are fully captive as defined in the literature. Their
disaffection for PT reflects that transit lacks the convenience and the reach of private vehicles, as recent
laws have made it easier for undocumented immigrants to obtain a driver’s license while some bus lines
were discontinued, some bus frequencies were reduced, and investments shifted from bus transit to
commuter rail.

31
One limitation of this study is that we do not have information about the type of public transportation
that was taken by NHTS respondents in the 30 days prior to their survey day. This prevents us from
distinguishing between bus and heavy rail/metro users, which is potentially problematic because the
literature shows that TNCs impacts bus transit differently than heavy rail/metro systems (Rayle et al.,
2016; Clewlow and Mishra, 2017; Feigon & Murphy, 2018; Hall et al., 2018; Malalgoda & Lim, 2019). A
second limitation is the restriction of our dataset to individuals/households who have access to motor
vehicles, as explained above. This feature explains the apparent contradiction between our finding that
low-income people are reluctant to take transit and the literature, which reports that low-income
people are prime users of bus transit in the U.S. (Myers, 1997; Garrett & Taylor, 1999; Polzin et al., 2000;
Kim et al., 2007; Krizek and El-Geneidy, 2007; Taylor and Morris, 2015). A third limitation is the absence
in the 2017 NHTS of detailed location data and mode-specific cost and travel time, which would have
helped us better understand mode choice. A fourth limitation is that Groups 1-3 contain infrequent and
frequent users of a given mode. However, accounting for the frequency of use of transit or TNCs would
likely have required a more complex model that would have been more difficult to interpret.
Future work could explore whether TNCs are complements or substitutes for different types of public
transportation (e.g., heavy rail or light rail versus buses). It would also be of interest to compare the
travel behavior of individuals/households before and after the emergence of TNCs (using, for example,
matching methods such as propensity score matching), and analyze the potential opportunities and
obstacles for transit to partner with micro-mobility providers.
References
Abduljabbar, R.L., Liyanage, S., Dia, H. (2021). The role of micro-mobility in shaping sustainable cities: A
systematic literature review. Transp. Res. Part D Transp. Environ. 92, 102734.
https://doi.org/10.1016/j.trd.2021.102734

32
Alemi, F., Circella, G., Handy, S., Mokhtarian, P. (2017). What influences travelers to use Uber? Exploring
the factors affecting the adoption of on-demand ride services in California. In: TRB 96th annual
meeting. pp. 17-05630.
Alemi, F., Circella, G., Mokhtarian, P., Handy, S. (2018b). On-demand Ride Services in California:
Investigating the Factors Affecting the Frequency of Use of Uber/Lyft. In: Transportation
Research Board 97th Annual Meeting.
Alemi, F., Circella, G., Sperling, D. (2018a). Adoption of Uber and Lyft, Factors Limiting and/or
Encouraging Their Use and Impacts on Other Travel Modes among Millennials and Gen Xers in
California. In: Transportation Research Board 97th Annual Meeting.
Bierlaire, M. (2020). A short introduction to PandasBiogeme. Technical report TRANSP-OR 200605.
Transport and Mobility Laboratory, ENAC, EPFL.
Blumenberg, E., Ralph, K., Smart, M., Taylor, B.D. (2016). Who knows about kids these days? Analyzing
the determinants of youth and adult mobility in the U.S. between 1990 and 2009. Transp. Res.
Part A Policy Pract. 93, 39–54. https://doi.org/10.1016/j.tra.2016.08.010
Brown, A.E., Blumenberg, E., Taylor, B.D., Ralph, K., Voulgaris, C.T. (2016). A taste for transit? Analyzing
public transit use trends among youth. J. Public Transp. 19, 49–67.
https://doi.org/10.5038/2375-0901.19.1.4
Buehler, R., Hamre, A. (2015). The multimodal majority? Driving, walking, cycling, and public
transportation use among American adults. Transportation (Amst). 42, 1081–1101.
https://doi.org/10.1007/s11116-014-9556-z
Chen, Z. (2015) Impact of ride-sourcing services on travel habits and transportation planning. 75.
Circella, G., Berliner, R., Lee, Y., Handy, S.L., Alemi, F., Tiedeman, K., Fulton, L., Mokhtarian, P.L. (2017).
The Multimodal Behavior of Millennials: Exploring Differences in Travel Choices between Young
Adults and Gen Xers in California. Transp. Res. Board 96th Annu. Meet.

33
Clark, H.M. (2017). Who rides public transportation? Washington D.C.: American Public Transportation
Association.
Clewlow, R.R., Mishra, G.S. (2017). Shared mobility: Current adoption, use, and potential impacts on
travel behavior. In: Transportation Research Board 96th Annual Meeting. pp. 9–10.
Clewlow, R. R., & Mishra, G. S. (2017). Disruptive Transportation: The Adoption, Utilization, and Impacts
of Ride-Hailing in the United States. In Institute of Transportation Studies ◦ University of
California, Davis: Vol. UCD-ITS-RR. https://escholarship.org/uc/item/82w2z91j
Erhardt, G.D., Roy, S., Cooper, D., Sana, B., Chen, M., Castiglione, J. (2019). Do transportation network
companies decrease or increase congestion? Sci. Adv. 5.
https://doi.org/10.1126/sciadv.aau2670
Ermagun, A., Levinson, D. (2017). Public transit, active travel, and the journey to school: a cross-nested
logit analysis. Transp. A Transp. Sci. 13, 24–37. https://doi.org/10.1080/23249935.2016.1207723
Feigon, S., & Murphy, C. (2018). Broadening Understanding of the Interplay Between Public Transit,
Shared Mobility, and Personal Automobiles. In Pre-publication draft of TCRP Research Report
195. The National Academies Press. https://doi.org/10.17226/24996
Foth, N., Manaugh, K., El-Geneidy, A.M. (2013). Towards equitable transit: Examining transit accessibility
and social need in Toronto, Canada, 1996-2006. J. Transp. Geogr. 29, 1–10.
https://doi.org/10.1016/j.jtrangeo.2012.12.008
Garrett, M., Taylor, B. (1999). Reconsidering social equity in public transit. Berkeley Plan. J. 13, 6–27.
https://doi.org/10.5070/bp313113028
Ge, Y., Knittel, C., MacKenzie, D., Zoepf, S. (2016). Racial and Gender Discrimination in Transportation
Network Companies. Natl. Bur. Econ. Res. https://doi.org/10.3386/w22776

34
Grahn, R., Harper, C.D., Hendrickson, C., Qian, Z., Matthews, H.S. (2019). Socioeconomic and usage
characteristics of transportation network company (TNC) riders. Transportation (Amst). 47,
3047–3067. https://doi.org/10.1007/s11116-019-09989-3
Hall, J.D., Palsson, C., Price, J. (2018). Is Uber a substitute or complement for public transit? J. Urban
Econ. 108, 36–50. https://doi.org/10.1016/j.jue.2018.09.003
Hampshire, R.C., Simek, C., Fabusuyi, T., Di, X., Chen, X. (2019). Measuring the Impact of an
Unanticipated Disruption of Uber/Lyft in Austin, TX Robert. Public Adm.
Harold M. Kohm (2000). Factors affecting Urban Transit Ridership. Bridging the Gaps. Canadian
Transportation Research Forum, Proceedings of the 35th Annual Conference.
Hasnine, M.S., Lin, T.Y., Weiss, A., Habib, K.N. (2018). Determinants of travel mode choices of post-
secondary students in a large metropolitan area: The case of the city of Toronto. J. Transp.
Geogr. 70, 161–171. https://doi.org/10.1016/j.jtrangeo.2018.06.003
Henderson, P. (2017) Some Uber and Lyft riders are giving up their own cars: Reuters/Ipsos poll,
https://www.reuters.com/article/us-autos-rideservices-poll/some-uber-and-lyft-riders-are-
giving-up-their-own-cars-reuters-ipsos-poll-idUSKBN18L1DA.
Hunt, J. (2019). Canadian transit ridership continues to trend upwards,
https://www.globenewswire.com/news-release/2019/11/13/1946490/0/en/Canadian-transit-
ridership-continues-to-trend-upwards.html
Jamal, S., Newbold, K.B. (2020) Factors Associated with Travel Behavior of Millennials and Older Adults:
A Scoping Review. Sustainability, 12, 8236. https://doi.org/10.3390/su12198236
Kim, S., Ulfarsson, G.F., Todd Hennessy, J. (2007). Analysis of light rail rider travel behavior: Impacts of
individual, built environment, and crime characteristics on transit access. Transp. Res. Part A
Policy Pract. 41, 511–522. https://doi.org/10.1016/j.tra.2006.11.001

35
Kooti, F., Djuric, N., Grbovic, M., Radosavljevic, V., Aiello, L.M., Lerman, K. (2017). Analyzing Uber’s ride-
sharing economy. In: 26th International World Wide Web Conference 2017, WWW 2017
Companion. pp. 574–582. Perth, Australia.
Koppelman, F. S., & Bhat, C. (2006). A Self Instructing Course in Mode Choice Modeling : Multinomial
and Nested Logit Models by with technical support from Table of Contents. In Elements (Vol. 28,
Issue 3).
Krizek, K., El-Geneidy, A. (2007). Segmenting Preferences and Habits of Transit Users and Non-Users. J.
Public Transp. 10, 71–94. https://doi.org/10.5038/2375-0901.10.3.5
LaChapelle, U. (2009). Transit Dependence and Choice Riders in the NHTS 2009: Associations with Walk,
Bicycle and Transit Trips. 1–16.
Leistner, D.L., Steiner, R.L. (2017). Uber for seniors? Exploring transportation options for the future.
Transp. Res. Rec. 2660, 22–29. https://doi.org/10.3141/2660-04
Malalgoda, N., Lim, S.H. (2019) Do transportation network companies reduce public transit use in the
U.S.? Transp. Res. Part A Policy Pract. 130, 351–372. https://doi.org/10.1016/j.tra.2019.09.051
McDonald, N.C. (2015). Are millennials really the “go-Nowhere” Generation? J. Am. Plan. Assoc. 81, 90–
103. https://doi.org/10.1080/01944363.2015.1057196
Mohiuddin, H. (2021) Planning for the first and last mile: A review of practices at selected transit
agencies in the United States. Sustain. 13, 1–19. https://doi.org/10.3390/su13042222
Myers, D. (1997). Changes over time in transportation mode for journey to work: Effects of aging and
immigration. Decenn. census data Transp. Plan. Case Stud. 2, 84–99.
Neff, J., Pham, L. (2007). A Profile of Public Transportation Passenger Demographics and Travel
Characteristics Reported in On-Board Surveys. Washington, DC.

36
Pew Research Center. (2018). Defining generations: Where Millennials end and post- Millennials begin.
http://www.pewresearch.org/fact-tank/2018/0 3/01/defining-generations-where-millennials-
end-and-post-millennials-begin
Pho, B. (2020). Orange County’s Outsourcing of Public Transit to Lyft Nearly Left Residents Stranded,
https://voiceofoc.org/2020/08/orange-countys-outsourcing-of-public-transit-to-lyft-nearly-left-
residents-stranded/.
Polzin, S.E., Chu, X., Rey, J.R. (2000). Density and captivity in public transit success: Observations from
the 1995 nationwide personal transportation study. Transp. Res. Rec. 10–18.
https://doi.org/10.3141/1735-02
Rahman, M., Strawderman, L., Adams-Price, C., Turner, J.J. (2016) Transportation alternative
preferences of the aging population. Travel Behav. Soc. 4, 22–28.
https://doi.org/10.1016/j.tbs.2015.12.003
Rayle, L., Dai, D., Chan, N., Cervero, R., Shaheen, S. (2016). Just a better taxi? A survey-based comparison
of taxis, transit, and ridesourcing services in San Francisco. Transp. Policy. 45, 168–178.
https://doi.org/10.1016/j.tranpol.2015.10.004
Rayle, L., Shaheen, S., Chan, N., Dai, D., Cervero, R. (2014). App-Based, On-Demand Ride Services:
Comparing Taxi and Ridesourcing Trips and User Characteristics in San Francisco. Berkeley.
Sadowsky, N., Nelson, E. (2017) The Impact of Ride-Hailing Services on Public Transportation Use: A
Discontinuity Regression Analysis. Econ. Dep. Work. Pap. Ser. 28.
Schaller, B. (2018). The New Automobility: Lyft, Uber, and the Future of American Cities. Brooklyn NY
11215.
Shaheen, S., Chan, N., Bansal, A., Cohen, A. (2015). Shared Mobility. Definitions, Industry Developments,
and Early Understanding. UC Berkeley.

37
Shaheen, S., Cohen, A. (2019). Shared Micromobility Policy Toolkit: Docked and Dockless Bike and
Scooter Sharing.
Sikder, S. (2019) Who Uses Ride-Hailing Services in the United States? Transp. Res. Rec. 2673, 40–54.
https://doi.org/10.1177/0361198119859302
Sperling, D. (2018). The Three Transportation Revolutions. Inst. Transp. Stud. Univ. California, Davis.
Swegles, F. (2016). San Clemente partners with Lyft to fill gaps after 2 OCTA bus routes end,
https://www.ocregister.com/2016/10/05/san-clemente-partners-with-lyft-to-fill-gaps-after-2-
octa-bus-routes-end/.
Taylor, B.D., Morris, E.A. (2015). Public transportation objectives and rider demographics: are transit’s
priorities poor public policy? Transportation (Amst). 42, 347–367.
https://doi.org/10.1007/s11116-014-9547-0
The Federal Highway Administration (2018). 2017 NHTS Data User Guide.
Train, K.E. (2009). Discrete choice methods with simulation, 2nd edition.
Vijverberg, W. P. (2011). Testing for IIA with the Hausman-McFadden test (No. 5826). IZA Discussion
Papers.
Vovsha, P. (1997). Application of cross-nested logit model to mode choice in Tel Aviv, Israel,
metropolitan area. Transp. Res. Rec. 6–15. https://doi.org/10.3141/1607-02
Williams, D. (2021). After blind woman was denied rides more than a dozen times,
https://www.cnn.com/2021/04/02/business/uber-blind-passenger-arbitration-trnd/index.html.

38
Appendix
Table A1: Descriptive statistics for individual-level analysis (N = 30,580)
Variables
Group 1:
PT but not
TNCs
Group 2:
TNCs but not
PT
Group 3:
Both PT and
TNCs
Group 4:
Neither PT nor
TNCs
Overall Sample
(N1 = 3,052)
(N2 = 3,004)
(N3 = 2,597)
(N4 = 21,927)
(N = 30,580)
Socio-demographic and economic attributes
Generation
Generation Z
1.87%
1.83%
2.19%
1.82%
1.86%
Generation Y (Millennials)
19.92%
47.40%
48.40%
18.28%
23.87%
Generation X
26.21%
28.53%
28.38%
23.30%
24.54%
Baby Boomers
43.28%
20.24%
18.83%
43.84%
39.34%
Silent Generation
8.72%
2.00%
2.19%
12.75%
10.39%
Hispanic Status (Hispanic =1)
5.73%
8.95%
7.86%
6.50%
6.78%
Ethnicity
Caucasian
81.65%
83.06%
81.32%
85.61%
84.60%
African American
5.96%
3.89%
4.00%
5.50%
5.26%
Asian
6.55%
6.19%
7.93%
4.06%
4.85%
Mixed
5.83%
6.86%
6.74%
4.83%
5.29%
Educational attainment
Less than high school
1.21%
0.60%
0.73%
1.60%
1.39%
High school degree
5.21%
2.80%
2.23%
11.48%
9.22%
Some college or associated degree
16.97%
16.38%
11.74%
28.72%
24.90%
Undergraduate degree
33.09%
42.41%
38.24%
29.86%
32.13%
Graduate or professional degree
43.51%
37.82%
47.05%
28.34%
32.37%
Annual household income
<$35,000
13.79%
9.02%
9.09%
18.11%
16.02%
$35,000-$74,999
21.00%
20.31%
16.63%
30.82%
27.60%
$75,000 to $124,999
27.88%
27.53%
24.10%
28.85%
28.22%
$125,000 to $199,999
21.66%
22.14%
24.99%
14.81%
17.08%
>=$200,000
15.66%
21.01%
25.18%
7.38%
11.06%
Household structure

39
One adult, no children
16.68%
23.77%
21.02%
13.01%
15.11%
2+ adults, no children
31.49%
41.41%
43.43%
25.06%
28.87%
One adult, some children
2.33%
2.46%
2.81%
2.80%
2.72%
2+ adults with children
23.69%
22.90%
22.53%
23.38%
23.29%
One retired adult, no children
7.37%
1.80%
1.81%
8.94%
7.48%
2+ retired adults, no children
18.45%
7.66%
8.39%
26.80%
22.52%
Household owns home
74.48%
61.68%
56.10%
78.09%
74.25%
Household workers
Household with no workers
20.54%
7.19%
7.97%
27.80%
23.36%
Household with one worker
35.32%
40.81%
36.73%
34.73%
35.56%
Household with two workers
37.91%
46.30%
48.63%
31.81%
35.27%
Household with three or more workers
6.23%
5.69%
6.66%
5.67%
5.81%
Works from home (Yes=1)
9.17%
15.75%
13.52%
8.26%
9.54%
Works full time
51.77%
70.47%
71.01%
44.29%
49.88%
Works part-time
12.78%
10.95%
9.74%
12.58%
12.20%
Household has fewer vehicles than drivers (Yes=1)
18.58%
9.35%
22.53%
8.63%
10.87%
Has medical condition (Yes=1)
4.29%
1.53%
1.89%
6.56%
5.44%
Not born in the US (Yes=1)
11.89%
10.79%
13.90%
8.26%
9.35%
Smartphone use
Daily
81.23%
97.04%
96.96%
76.28%
80.57%
Weekly
5.54%
1.50%
1.85%
5.39%
4.72%
Monthly
2.33%
0.57%
0.39%
2.28%
1.96%
Yea r l y
1.18%
0.03%
0.15%
1.20%
0.99%
Never
9.73%
0.87%
0.65%
14.85%
11.76%
Land use
Availability of transit services
Household in a CBSA with bus service
67.86%
68.94%
81.94%
43.24%
51.51%
Household in a CBSA with light rail service
59.99%
59.45%
75.47%
34.89%
43.25%
Household in a CBSA with heavy rail service
41.38%
23.50%
52.14%
13.78%
20.75%
Notes:
1. All explanatory variables in our models are binary, except for “Number of household members (Mean: 2.32; S.D: 1.16; Min: 1; Max: 11)” and
“Ln of population density (measured in 1,000/mi2)” (Mean: 1.11; S.D: 1.30; Min: -3.00; Max: 3.40), which are count and continuous variables,
respectively. These two variables are not shown in Table A1.
2. CBSA stands for Core Based Statistical Area.

40
Table A2: Descriptive statistics for household-level analysis (N = 23,947)
Variables
Group 1:
PT but not
TNCs
Group 2:
TNCs but not
PT
Group 3:
Both PT and
TNCs
Group 4:
Neither PT nor
TNCs
Overall Sample
(N1 = 2,574)
(N2 = 2,485)
(N3 = 2,342)
(N4 = 16,546)
(N = 23,947)
Socio-demographic and economic attributes
Generation
Generation Z
2.53%
2.70%
3.33%
1.89%
2.18%
Generation Y (Millenials)
21.87%
48.09%
50.26%
18.56%
25.08%
Generation X
28.40%
32.60%
32.41%
24.80%
26.74%
Baby Boomers
47.16%
23.34%
23.23%
47.09%
42.30%
Silent Generation
10.06%
2.37%
2.73%
15.33%
12.19%
Hispanic Status (Hispanic =1)
5.59%
8.93%
7.47%
6.21%
6.55%
Ethnicity
Caucasian
82.05%
82.90%
82.11%
85.74%
84.70%
African American
6.33%
4.35%
4.06%
5.89%
5.60%
Asian
6.14%
6.04%
7.81%
3.67%
4.59%
Mixed
5.48%
6.72%
6.02%
4.69%
5.12%
Educational attainment
Less than high school
0.62%
0.20%
0.38%
0.93%
0.77%
High school degree
4.74%
1.81%
1.45%
9.89%
7.68%
Some college or associated degree
15.38%
14.21%
9.82%
27.61%
23.17%
Undergraduate degree
32.40%
40.48%
34.97%
29.96%
31.81%
Graduate or professional degree
46.85%
43.30%
53.37%
31.60%
36.58%
Annual household income
<$35,000
15.35%
9.86%
9.44%
21.19%
18.24%
$35,000-$74,999
22.96%
22.05%
17.46%
32.63%
29.01%
$75,000 to $124,999
28.09%
27.81%
24.81%
27.09%
27.05%
$125,000 to $199,999
20.05%
21.29%
24.34%
12.93%
15.68%
>=$200,000
13.56%
18.99%
23.95%
6.16%
10.03%
Household structure
One adult, no children
19.77%
28.73%
23.31%
17.24%
19.30%
2+ adults, no children
28.67%
36.78%
40.31%
21.91%
25.98%
One adult, some children
2.64%
2.90%
3.03%
3.38%
3.22%

41
2+ adults with children
22.26%
21.69%
22.93%
20.00%
20.71%
One retired adult, no children
8.74%
2.17%
2.01%
11.85%
9.55%
2+ retired adults, no children
17.91%
7.73%
8.41%
25.61%
21.25%
Household owns home
73.74%
60.93%
56.75%
76.26%
72.49%
Household workers
Household with no workers
22.22%
7.89%
8.20%
30.80%
25.29%
Household with one worker
38.07%
45.39%
39.07%
37.58%
38.59%
Household with two workers
34.69%
41.73%
46.33%
27.81%
31.80%
Household with three or more workers
5.01%
4.99%
6.40%
3.81%
4.32%
Works from home (Yes=1)
12.04%
18.23%
17.21%
9.80%
11.64%
Works Full time
56.02%
75.37%
76.56%
48.21%
54.64%
Works Part-time
16.55%
14.69%
13.83%
14.71%
14.82%
Household has fewer vehicles than drivers (Yes=1)
16.16%
7.69%
20.20%
7.14%
9.45%
Has medical condition (Yes=1)
5.67%
2.25%
2.52%
8.21%
6.76%
Not born in the U.S. (Yes=1)
14.02%
12.19%
16.48%
9.02%
10.62%
Smartphone use
Daily
80.07%
96.74%
96.88%
74.24%
79.41%
Weekly
5.71%
1.69%
1.75%
5.58%
4.82%
Monthly
2.41%
0.56%
0.43%
2.36%
1.99%
Yea r l y
1.20%
0.04%
0.17%
1.21%
0.99%
Never
10.61%
0.97%
0.77%
16.60%
12.79%
Land use
Availability of transit services
Household in a CBSA with bus service
66.39%
68.09%
81.04%
42.20%
51.28%
Household in a CBSA with light rail service
58.66%
58.75%
74.30%
33.86%
43.06%
Household in a CBSA with heavy rail service
39.28%
21.93%
50.51%
13.13%
20.51%
Notes:
1. All explanatory variables in our models are binary, except for “Number of household members (Mean: 2.19; S.D: 1.15; Min:1; Max:11)” and
“Ln of population density (measured in 1,000/mi2)” (Mean: 1.11; S.D: 1.31; Min: -3.00; Max: 3.40), which are count and continuous variables,
respectively. These two variables are not shown in Table A2.
2. CBSA stands for Core Based Statistical Area.

42
About the Authors
Farzana Khatun (fkhatun@uci.edu) is a Ph.D. candidate in Transportation Science at the University of
California, Irvine. Her research interests include public transit, active transportation, TNCs, land use and
transportation, and applications of GIS to transportation planning and travel behavior.
Jean-Daniel Saphores (saphores@uci.edu) is a Professor in the Department of Civil and Environmental
Engineering at the University of California, Irvine, and a member of the Institute of Transportation
Studies. His research interests include understanding travel behavior, and the nexus between
transportation, energy, the environment, and health.