Integrating Shared Mobility Services with Public Transit in Areas
of Low Demand
, Krishna M. Gurumurthya, Cesar N. Yahiaa, Kara M.
Kockelmana, Randy B. Machemehla
aThe University of Texas at Austin, Department of Civil, Architectural and Environmental Engineering,
301 E. Dean Keeton St. Stop C1761, Austin, TX 78712, United States
The increasing demand for shared mobility services has led to multiple Public-Private Part-
nerships (PPPs), where public transit agencies and mobility providers work together to in-
crease transit service coverage. However, several partnerships have failed due to the lack of
demand and budget constraints. Furthermore, the limited data-sharing agreements restrict
the analysis and dissemination of outcomes. This paper analyzes the integration of shared-
mobility services into the public transit system using data from pilot programs in Austin,
Texas. The use of Transportation Network Companies (TNCs) and microtransit service
is evaluated as ﬁrst-mile-last-mile (FMLM) solutions. The associated TNC, microtransit,
and bus transit demand are analyzed using ridership data, and the trip characteristics are
discussed and compared. Then, after controlling for confounding variables through match-
ing, a diﬀerence-in-diﬀerences (DID) model is used to infer the eﬀect of microtransit on bus
transit demand. Speciﬁcally, by evaluating bus ridership before and after implementing the
microtransit program, the DID model estimates the change in bus demand attributed to
microtransit. The results suggest that the TNC services, focusing only on FMLM, were not
as popular as microtransit services oﬀering FMLM and intrazonal trips. Even though the
microtransit trips were longer and, in some cases, the driver response times were higher than
for TNC trips. Also, the DID model demonstrates that the microtransit service did not
substantially impact the bus transit ridership, suggesting that the trips were mainly used
for intrazonal door-to-door trips and not for FMLM trips. A further discussion highlights
the need for performance metrics beyond demand and costs to capture the beneﬁts of these
partnerships for improving mobility, safety, and customer experience.
Keywords: public-private partnerships, public transportation, ridesourcing, microtransit,
on-demand transit, diﬀerence in diﬀerences.
Email addresses: email@example.com (Natalia Zuniga-Garcia), firstname.lastname@example.org (Krishna
M. Gurumurthy), email@example.com (Cesar N. Yahia), firstname.lastname@example.org (Kara M.
Kockelman), email@example.com (Randy B. Machemehl)
Preprint submitted to Transport Policy December 8, 2020
Public transit agencies are evaluating the incorporation of on-demand shared mobility op-
tions to increase current transit service coverage. While traditional ﬁxed-routes face chal-
lenges to provide eﬃcient and high-quality services in areas with low or sparse demand, al-
ternative on-demand shared modes can supplement ﬁxed-routes and provide ﬂexible service
(Mishra et al.,2020). In several cities, public transit authorities have started implementing
Public-Private Partnerships (PPPs) to provide services in conjunction with Transportation
Network Companies (TNCs) and microtransit companies; examples include Boston, San
Francisco, and Austin (Blodgett et al.,2017;Feigon et al.,2018).
Compared to TNCs, microtransit requires higher operating costs and signiﬁcant subsidies.
However, the use of TNCs raises accessibility and equity concerns. In particular, TNCs are
not subject to regulations by the Americans with Disabilities Act (ADA), and the TNC
service can not be used by riders that do not have a smartphone (Lader and Klein,2018).
While incorporating these options seems to provide more public transportation solutions
and alleviate transit ﬁrst-mile-last-mile (FMLM) access and egress concerns, TNCs and
microtransit partnership pilot programs have failed due to the lack of ridership and budget
constraints (Westervelt et al.,2018). Therefore, more research is needed to understand
the eﬀective integration of shared mobility with the public transit system. This research
compares TNCs and microtransit services using information from two pilot programs in
Ridesourcing or TNCs, such as Uber, Lyft, and Didi, provide pre-arranged or on-demand
transportation services for compensation (Shaheen et al.,2016). These services are expanding
at a rapid pace and increasing their market share (Yahia et al.,2020). After its establishment
in 2009, Uber completed 140 million annual trips during 2014, and this number increased to
a total of 5.22 billion trips worldwide in 2018 (Smith,2019).
On the other hand, microtransit services are deﬁned as privately owned and operated
transit systems (using large vans or minibusses) that can oﬀer ﬁxed and ﬂexible route services
(Shaheen et al.,2016); examples include Via, Chariot, and Bridj. While on-demand transit
services have existed for decades, emerging microtransit services (enabled by technology
similar to the mobile smartphone applications pioneered by TNCs) have gained popularity
in recent years (Lucken et al.,2020;Pike and Kazemian,2020).
The increasing demand for shared mobility options led to the introduction of multiple
pilot programs. Public transit agencies partner with mobility providers to integrate these
services into the transit system (Westervelt et al.,2018). Lucken et al. (2020) survey of
62 PPPs in the U.S. showed that they typically oﬀer four service model types: FMLM (32
percent), low-density (39 percent), oﬀ-peak (8 percent), and paratransit (21 percent). The
FMLM model includes trips with start or end at a ﬁxed-route or ﬁxed-stop location. Low-
density models serve trips with starts or ends outside a ﬁxed-route service area and can also
be used for FMLM trips. Oﬀ-peak services are oﬀered to ﬁll the ﬁxed-route temporal gaps,
and paratransit trips serve a speciﬁc population (e.g., senior citizens). Although several
pilot programs have tested these diﬀerent service models, there are limited analyses of the
outcomes, mainly due to the lack of openly-available data. Some partnerships between public
transit agencies and private companies restrict the data exchange agreements; this limits the
types of analysis that the agencies or external sources can provide (Pike and Kazemian,2019;
Terry and Bachmann,2020).
1.2. Objective and Contributions
This research analyzes the integration of TNCs and microtransit with local ﬁxed-route bus
service in areas with low demand in Austin, Texas. Data from two pilot programs are used,
including PPPs oﬀering the FMLM and low-density service models in the same geographic
location during diﬀerent periods. The study area corresponds to four zones where land use
and road networks make it diﬃcult to provide cost-eﬀective bus service. Previous ﬁxed-route
buses in the area were replaced before introducing the pilot programs due to low ridership.
These four areas present diﬀerent socioeconomic attributes, allowing for further comparison
of the inﬂuence of various characteristics. The PPPs analyzed are described as follows:
1. The ﬁrst pilot program is a partnership with RideAustin, an Austin-based TNC. It
lasted approximately six months and consisted of free TNC trips to bus transit users
from the designated areas. The service was designed to solve FMLM accessibility, and
it was used for trips where either the origin or the destinations occur at a ﬁxed set of
2. The second pilot program, currently operating, consists of a geofenced microtransit
service in a partnership with Via called “Pick-Up.” In this case, the service has the
same cost as a bus ride, but it includes intrazonal door-to-door trips rather than only
FMLM trips (i.e., the trips are not restricted to a ﬁxed set of stops).
The analytical framework is divided into two main tasks. The ﬁrst task includes an analysis
of the bus, TNC, and microtransit demand and compares their primary service characteristics
in each of the four zones. The second task develops a causal analysis to estimate the eﬀect of
microtransit service on public transit. In particular, the impact of microtransit is evaluated
by measuring the change in monthly bus ridership attributed to the implementation of the
microtransit program. To this end, a diﬀerence-in-diﬀerences (DID) study is implemented
using bus ridership and microtranist demand information.
The contributions of this work include (i) a comparison of the characteristics of two
shared-mobility services oﬀered in the same area during two diﬀerent periods, (ii) an anal-
ysis of the eﬀects of these services on bus demand across four zones with heterogeneous
socioeconomic characteristics, and (iii) a causal study implementation to measure the inte-
gration of microtransit into the transit system. The main results show that the microtransit
service did not result in signiﬁcant changes in the bus ridership, which suggests that it was
mainly used for intrazonal trips and not for FMLM.
Subsequent sections of the paper are organized as follows: Section 2provides a literature
review of the principal aspects of TNCs and microtransit and their integration with public
transit systems; Section 3describes the dataset; Section 4presents the methodology de-
scribing the DID method; Section 5includes the empirical analysis and provides results and
discussion; and ﬁnally, Section 6contains conclusions and ﬁnal remarks.
2. Literature Review
Public transit provides an eﬃcient means of travel that can serve captive riders who do not
have access to alternative modes. Travelers, however, are unwilling to walk longer than about
a quarter-mile (El-Geneidy et al.,2014) to reach their transit stops or their destinations from
a stop. This imposes a major hurdle and makes transit an unattractive mode for many origin
and destination pairs.
Several studies have looked into transit in the era of ridesourcing and microtransit, both
from a competition that erodes transit ridership, as well as an FMLM solution that may
improve transit use. There are mixed opinions on whether these services help or erode
existing transit services. Shaheen and Chan (2016) studied shared mobility as a potential
to overcome the FMLM problem. Their comparison of several shared modes, along with
the more recent microtransit option, shows that there is indeed a gap to ﬁll in terms of
improving public transit use by solving the FMLM problem. Throughout their study, each
shared mode (e.g., shared vehicles, bikes, or scooters) had the potential to improve transit
use, but this was not always the case as they may emerge as competing modes. Stiglic et al.
(2018) corroborates this from a carpooling perspective, where travelers may share rides to
get to and from a rail line. Hall et al. (2018) conducted a statistical analysis of Uber trip
data, and their conclusions also aligned with Shaheen and Chan (2016). The three-year
dataset (2013-16) showed declining transit ridership after Uber’s entry into the market in
large regions and regions with small transit agencies. Even though smaller cities saw an
increase in transit ridership, the ridesourcing service seemed to decrease transit use overall.
Bian and Liu (2019) used a static optimization approach for a train station in New Jersey
to check the feasibility of a personalized pooled service for ﬁrst-mile access while taking into
account preferred arrival times, disutilities from sharing with strangers, and inconvenience
from early arrival. Overall, the results seemed to suggest that such a service should improve
revenues and mode shares for transit. Similarly, Huang et al. (0) used a simulation-based
method to analyze the use of shared automated vehicles (SAVs) for FMLM service to rail
transit connections in Austin, Texas. The study found that SAV helped to increase the
percentage of users shifting from driving alone to transit, but the total travel time served
as the most signiﬁcant determinant in trip-makers’ mode share. Using a simulation-based
approach as well, Gurumurthy et al. (2020) showed that SAVs are not as useful to bus transit
on a regional scale. Their results suggest that providing low-cost fares may deter the use of
transit and lead to a rise in the SAV mode.
A stated-preference survey analysis by Yan et al. (2019) revealed that transit ridership
might increase when used with shared mobility services, where this boost in ridership results
from reduced wait and in-vehicle travel times. However, transfer penalties were not con-
sidered, so the university-focused result may be biased. Reck and Axhausen (2020), on the
other hand, incorporated transfer penalties and showed that even small penalties of about
5-10 minutes deprives nearly 90% of all travelers the time-cost savings that such ridesourcing
services may provide.
In terms of PPPs, existing case studies agree that merging innovative shared modes with
transit may be beneﬁcial, but care must be taken to ensure the proper distribution of these
beneﬁts. Feigon and Murphy (2016) reported on the rise in shared modes, the corresponding
beneﬁts to the system, and the need for transit agencies to evolve from ﬁxed-route mass
movers to mobility agencies. Travelers in regions with high transit and ridesourcing use
were found to spend less on transportation, on average, and were less likely to buy a new
car. Although shared mobility options were not preferred for commute trips, transit use,
on average, was found to be higher. Blodgett et al. (2017), Westervelt et al. (2018) and
Lucken et al. (2020) conducted interviews of several U.S. transit agencies that have PPP’s
with ridesourcing companies; their studies highlight how quickly such PPP contracts are
emerging throughout the country. In general, PPPs were found to be either agency-owned
or agency-subsidized partnerships.
Most agencies are continuously changing their PPPs to ensure the provision of ADA
compliant and equitable service (Lucken et al.,2020). As private companies, TNC’s are
often concerned with maximizing proﬁt at the expense of equitable service. TNCs also require
smartphones and do not cater to travelers without a credit or debit card. In contrast, the
use of microtransit allows a transit agency to oﬀer ADA compliant vehicles and are better
placed to serve all travelers. Irrespective of equity-related disadvantages, many agencies are
seeing an increase in transit use after implementation of PPP programs (Blodgett et al.,
2017;Westervelt et al.,2018;Lucken et al.,2020). Moreover, without PPPs, the inequity in
obtaining aﬀordable and accessible travel may worsen– Jin et al. (2019) report low pickups
by Uber in areas with a larger percentage of minority groups.
While both ridesourcing and microtransit have advantages and disadvantages from a
qualitative point of view, further analysis is needed to quantify the impact of PPPs on transit
ridership. This paper’s contribution stems from the quantitative analysis of ridesourcing and
microtransit programs; speciﬁcally, the proposed study investigates the impact of shared
mobility partnerships on the ﬁxed-route transit service in Austin, TX.
3. Data Description
This study uses data from several sources. The public transit demand is obtained from
the Automatic Passenger Count (APC) information for Austin, publicly available through
the Texas Open Data Portal.1It provides public transit vehicle information such as bus
boarding and alighting counts, arrival and departure times, vehicle location (latitude and
longitude coordinates), among others. This data is combined with stop locations obtained
from the General Transit Feed Speciﬁcation (GTFS) data. The vehicle location from APC
was matched with the stop location identiﬁcation number from GTFS. The processing and
cleaning stage included the removal of double counts per stop, counts located more than
50 meters away from the corresponding stop, and extreme values of boarding and alighting
The information from the TNC trips and the microtransit service was provided by
Austin’s public transit agency, Capital Metropolitan Transportation Authority (CapMetro).
In June 2018, CapMetro made signiﬁcant changes in the bus network (CapMetro,2018).
These changes are referred to as CapRemap. After the renovations, CapMetro deﬁned sev-
eral “Mobility Innovation Zones,” where they implemented alternative mobility options by
using various pilot projects. This change implied the removal of some of the ﬁxed-route
1This website can be accessed at https://data.texas.gov/
services and the introduction of a TNC (and eventually a microtransit) service. Figure 1
shows the location of the four areas analyzed, represented as Traﬃc Analysis zones (TAZs),
and the bus network before and after the bus services changes.
(a) Bus routes before CapRemap
(b) Bus routes after CapRemap
Figure 1: Description of mobility innovation zone TAZs
The ﬁrst pilot program implemented by CapMetro was a partnership with RideAustin,
a non-proﬁt TNC based in Austin, Texas. The program started with the Exposition area in
June 2018 (right after CapRemap happened) and ended in October 2018. The other zones
initiated it in October 2018 and ended in April 2019, approximately. The TNC program
consisted of free TNC trips oﬀered to public transit users from the designated areas. The
service was implemented exclusively to solve FMLM accessibility, and it involved trips with
origins or destinations set to a speciﬁc stop.
The second pilot program, currently in operation, consists of a geofenced microtransit
service in a partnership with Via. In this case, the service has the same cost as a bus
ride (USD 1.5), but it includes intrazonal door-to-door travel and also allows for FMLM
trips. The service started in August 2019 in the same areas where the TNC program was
functioning four months before.
In addition to TNC, microtransit, and public transit trip information, socio-demographic
information was obtained, including race and ethnicity, age distribution, and household
information to characterize the study area. This information is obtained using TAZ-level
data from the Capital Area Metropolitan Planning Organization (CAMPO) website and
using the American Community Survey (ACS) 2016. The ACS information is aggregated at
Block Groups (BG) level; therefore, an additional spatial process was required to summarize
at TAZ-level. The process consisted of intersecting TAZ and BG areas to estimate the
proportion of BG per TAZ. The TAZ summary included the average BG values weighted by
the BG area and population density. Table 2provides a summary of description variables
for each study area. The four areas show diﬀerent characteristics. For example, Exposition
has the highest average annual household income of $78k, while for the East Austin zone,
the average household income is $ 31k, a diﬀerence of 40 percent.
Table 1: Description of the mobility innovation zones
Variables Exposition East Northeast Walnut Creek
Transit supply [Sept. 2018]
Stop density (stops/km2) 4.73 11.04 7.03 6.05
Bus frequency weekday peak hour (bus/hour) 10.37 7.18 11.03 8.49
Bus frequency weekend peak hour (bus/hour) 6.09 5.84 8.86 5.65
Socio-demographic information 
Population density (residents/km2) 1,247 1,346 1,730 1,754
Employment density (employees/km2) 778 1,122 466 1,779
Retail employment density (employees/km2) 108 188 131 293
Race or ethnicity 
Proportion of White population 0.85 0.71 0.70 0.65
Proportion of Black/African American pop. 0.02 0.12 0.22 0.12
Proportion of Asian population 0.10 0.13 0.09 0.06
Household information 
Average household size 2.62 2.61 2.85 1.91
Median household income (USD) 78,288 31,453 43,740 45,207
Sample size (number of TAZs) 9 17 10 14
This study aims to evaluate the impact of incorporating shared mobility services on public
transit ridership. For this, a quasi-experimental diﬀerence-in-diﬀerences (DID) analysis was
developed to investigate the causal relationship between mobility services and bus transit.
The proposed DID approach calculates the eﬀect of a treatment on an outcome; this eﬀect
is determined by comparing the average change of the outcome variable within a treatment
group to the corresponding change in a control group. In this case, the treatments are the
implementation of shared-mobility programs, and the outcome corresponds to the bus transit
ridership. The treatment groups are the areas where the service was implemented. The con-
trol groups are areas with similar socio-economic, demographic, and mobility characteristics,
but where the shared services were not implemented.
The causal analysis requires that the only change in the study area’s conditions is due
to the treatment application. This restriction comes from the stable unit treatment value
assumption (SUTVA). SUTVA requires that a particular unit’s response depends only on
the assigned treatment, not the treatments of other events or other areas around it. In
this case, the TNC service was implemented right after CapRemap when CapMetro made
several public transportation network changes. Therefore, it is not possible to isolate the
eﬀect of TNC trips from the bus transit demand since it may also capture the impact of
the CapRemap changes. The microtransit service, or Pick-Up program, started operating
in August 2019, more than a year after CapRemap began, which allows for a more isolated
analysis. For this reason, the DID analysis is focused only on microtransit, and TNC services
will be analyzed qualitatively in the results section.
To evaluate the eﬀect of the microtransit services on the bus transit demand, ﬁrst, an
intervention date is deﬁned as the date the microtransit started operating, corresponding to
August 2019. Then, the pre-intervention period is described from August 2018 to August
2019. Although this period includes months close to the CapRemap date, its eﬀect would not
signiﬁcantly aﬀect the results since there is a large sample (a year). The CapRemap eﬀect
would eventually stabilize during this period. Also, although this period includes months in
which the TNC service was functioning, the low demand for the service implies that it would
not have a signiﬁcant impact either. The post-intervention period is deﬁned from August
2019 until December 2019. Figure 2shows a timeline with the most critical events in the
study area and the period deﬁnitions.
Figure 2: Timeline and period deﬁnition
Furthermore, the analysis is extended to include areas surrounding the service zones to
account for trips made as FMLM with routes and stops outside the limits of the service area.
The following sections further discuss the DID method and explain the methodology used
to ﬁnd the control zones and the buﬀer-area analysis.
4.1. Diﬀerence-in-diﬀerences (DID) method
DID is a quasi-experimental research study design often used to study causal relationships
in settings where randomized trials are infeasible. This study aims to compare the potential
outcome that results from the implementation of the microtransit program (i.e., treatment)
with the alternative potential outcome describing the case when the program is not im-
plemented (control). The outcome variable considered is bus transit ridership. Since both
potential outcomes can not be observed at the same time and in the same area, the potential
outcomes with treatment are factual (can be observed), but the potential outcomes with no
treatment are counterfactual (cannot be observed). For this, data from a control group is
used to impute untreated outcomes in the treated group. This control group corresponds
to areas with similar characteristics but where the Pick-Up (microtransit) service was not
The DID setup uses the Pick-Up service zone as the unit of analysis. This Pick-Up zone
is composed of a set of TAZs that represent the treatment group. To deﬁne the control
group, the treatment group TAZs are matched to reference TAZs not in the Pick-Up service
zone. Thus, there are two sets of TAZs composing the treated group and the control group.
These two groups g= 1,2 (where group 1 is treatment and group 2 is control) are observed
in multiple time periods summarized in monthly bus transit demand from August 2018 until
December 2019. These months are assigned to the pre-intervention period and the post-
intervention period dummy variable based on the description shown in Figure 2. In the
ﬁrst period, both groups are exposed to the control conditions. In the second period, the
treatment is applied to the treated group but not to the control group. The standard DID
estimate equation is described as follows:
Ygt =β0+β1Tg+β2Pt+β3Dgt +gt (1)
Where, Yis the monthly bus transit demand, gis the group number (g= 1,2) observed
in the two time periods t(t= 1,2). The dummy variable Tgidentiﬁes the time invariant
observations in group g, i.e., its value is 1 for group 1 (treatment) and 0 for group 2 (control).
While the dummy variable Ptindicates observations from period tand does not vary across
the groups (i.e, its value is 1 for pre-intervention period and 0 for post-intervention period).
The treatment variable Dg t is the product of these two dummy variables. The β0coeﬃcient
corresponds to the baseline average, β1is the time trend in the control group, β2is the
diﬀerence between the two groups before the intervention, and β3is the diﬀerence in changes
over time, also known as the average eﬀect of treatment on the treated (ATT) – the eﬀect
of treatment on the treated group in the post-treatment period, as described in Equation 3.
Figure 3shows a graphical representation of the coeﬃcients.
AT T ≡E[Y1(2) −Y0(2)|A= 1] (3)
Where, Ya(t) is the potential outcome given treatment aat time t. Here, t= 2 represents the
post-treatment period, a= 1 represents the treatment, a= 0 represents the no treatment,
and A = 1 refers to the treated group.
The identiﬁcation of causal eﬀects from the DID model depends on several assumptions.
In addition to the SUTVA described previously, the model assumes a “parallel trend” es-
tablishing that both the treated and control units would evolve along a parallel path in
the absence of treatment. A plot of the outcomes for the treated and control group in the
pre-treatment period is used and discussed in the results section as evidence supporting this
Figure 3: Diﬀerence-in-diﬀerences
4.2. Matching control groups
The control groups are found using a matching technique. Rather than using the entire
sample population to estimate the DID eﬀect, units in the control group are selected based
on their “closeness” to units in the treated group. Matching is attractive in DID analysis
because it improves the groups’ comparability and possibly their outcome trends.
The matching is done at the TAZ-level by estimating the Mahalanobis distance between
pairs of TAZs; the Mahalanobis distance is a metric of closeness used to ﬁnd how similar
two sets of covariates are. The covariates used include population density, retail employment
density, median income, the proportion of White population, stop density, and bus frequency
during weekdays and weekends. The distance corresponds to the square root of the squared
diﬀerence of the set of covariates scaled using its covariance, as denoted in Equation 4.
For each TAZ in the treatment group (Pick-Up service zone), the Mahalanobis distance
is computed with every possible TAZ in the control group. Then, the treatment TAZ is
matched to the closest control TAZ (minimal Mahalanobis distance). A TAZ in the control
group is matched to at most one TAZ in the treatment group. The set of eligible TAZs for
the control groups correspond to those located in the CapMetro Austin service area, where
the TAZs located within the Pick-Up service zones are excluded.
D(Xi, Xj) = q(Xi−Xj)TS−1(Xi−Xj) (4)
In Equation 4,D(Xi, Xj) is the Mahalanobis distance between TAZ pairs. Xiis the vector
of covariates for TAZ iin the treated group (Pick-Up service zone). Xjis the vector of
covariates for TAZ jfrom the set of TAZs in the control group. Sis the covariance matrix
of Xused to scale the variables.
Table 2shows the description of the control groups found using this method. This table
can be compared to Table 3, corresponding to the treatment groups. Although the matching
technique is not perfect, the pair of groups share similarities that make the DID analysis
Table 2: Description of the control group TAZs
Variables Exposition East Northeast Walnut
Transit supply 
Stop density (stops/km2) 5.27 6.95 15.53 4.52
Bus frequency weekday peak hour (bus/hour) 9.49 6.72 11.99 9.19
Bus frequency weekend peak hour (bus/hour) 8.88 4.88 7.30 6.75
Socio-demographic information 
Population density (residents/km2) 1,563 1,561 2,023 1,545
Employment density (employees/km2) 679 1,654 1,879 1,430
Retail employment density (employees/km2) 348 323 528 354
Race or ethnicity 
Proportion of White population 0.85 0.73 0.70 0.70
Proportion of Black/African American pop. 0.06 0.09 0.10 0.13
Proportion of Asian population 0.10 0.07 0.05 0.06
Household information 
Average household size 2.47 2.37 2.64 2.20
Median household income (USD) 68,711 43,629 38,640 43,395
Sample size (number of TAZs) 9 17 10 14
4.3. Buﬀer areas
Changes that aﬀect the transit demand in the Pick-Up service zones may inﬂuence the
proximate areas causing a spillover eﬀect. For example, users may take the microtransit
service to access bus stops outside the service area. The spillover eﬀect would violate the
DID SUTVA assumption since the transit demand changes are not isolated to the Pick-Up
service areas but may include the surrounding areas. To limit this eﬀect, the matching and
DID process is repeated with a diﬀerent unit of analysis; speciﬁcally, the modiﬁed treatment
group consists of the Pick-Up service zones in addition to surrounding buﬀers. In other
words, the treatment group would include an augmented area that adds adjacent TAZs to
the Pick-Up service areas. Figure 4shows the location of the buﬀer zones around each service
Figure 4: TAZs buﬀer illustration: the DID analysis is also implemented for augmented service areas that
include a surrounding buﬀer
5. Results and Discussion
This section presents the results and main ﬁndings. First, public transit, TNC, and micro-
transit demand and trip characteristics are described. The second part presents the DID
results used to evaluate the integration of microtransit into the ﬁxed-route services.
5.1. Demand analysis
The historical monthly demand for the bus service in each mobility innovation zone is ob-
tained using information from the passenger counts. This demand is compared with the
monthly demand reported for the TNC and microtransit service. Figure 5presents the
comparison of the estimates.
Figure 5: Monthly demand for bus transit, TNC, and microtransit by zone
The information in Figure 5shows a marked diﬀerence in the demand for TNC FMLM
trips and microtransit. On average, there were 37 TNC trips per month and 597 micro-
transit trips per month, a diﬀerence of approximately 1,500 percent. Furthermore, TNC
demand seems to decrease with time while the microtransit demand increased. However, it
is important to highlight that these two modes functioned on diﬀerent service model types.
TNC trips, focused exclusively on FMLM trips, were not as popular as the microtransit
service that allowed for FMLM trips and intrazonal trips. This comparison of demand is
particularly useful because it allows diﬀerentiating the ridership across zones even though
the service model is diﬀerent. The Exposition area showed the lowest demand for TNC and
microtransit, while the Northeast area showed the highest TNC demand. Walnut Creek
showed the highest microtransit ridership, followed by the Northeast and East zones.
The study areas present diﬀerent characteristics. The household income ranges between
$31k to $78k – where the city’s mean is $55k (2015), and the average household size varies
considerably around Austin’s mean of 2.3 (2015). Transit stop density is low in Exposition,
at ﬁve stops per sq. km, whereas East Austin has more than twice as many stops per sq. km.
Population composition by race indicates a prevalence of white population on Exposition (85
percent) compared to Walnut Creek (65 percent). The Northeast area presents the highest
population density and household size. East Austin and Walnut Creek show the highest
employment density, and Walnut Creek’s retail employment density is almost twice the
Research on shared mobility modes showed that areas with high income and a high pro-
portion of Caucasians do not tend to generate high demand for ridesourcing trips (Lavieri
et al.,2018) or micromobility services (Zuniga-Garcia et al.,2020a;Zuniga-Garcia and
Machemehl,2020) and users are more prone to travel by car only (Giuliano,2003;Smart,
2015). This can be related to the low demand for both TNC and microtransit in the Exposi-
tion zone. In contrast, Walnut Creek shows the highest employment and retail employment
density, which can generate more intrazonal trips and, thus, more microtransit demand.
Figure 6shows a comparison of the TNC and microtransit trips in terms of trip distance,
duration, and driver reach time across the study areas. The plots describe the distribution
of the trip’s characteristics summarized in box plots. The TNC trip distance average ranges
between 0.6 miles (Exposition) and 1.8 miles (East Austin). In contrast, the microtransit
distance of these two zones is greater for Exposition, with an average of 1.0 miles, and
lower for East Austin, with 1.1 miles. Northeast and Walnut Creek areas show similar trip
distances, both across modes and across sites, with an average of 1.2 miles approximately.
On average, the trip duration for microtransit services is higher than for TNC trips.
The driver reaching time is the time it took the driver to pick up the user after the
service was requested and is a good indicator of the level of service of the mode (Zuniga-
Garcia et al.,2020b;Tec et al.,2019). The average TNC driver reaching time was higher
than the microtransit services for East and Northeast Austin zones, with diﬀerences of 2.9
and 1.9 minutes, respectively. However, the Walnut Creek area showed a lower TNC time
with a diﬀerence of 3.5 minutes. The information on this variable for the Exposition area was
not available. Overall, these results indicated that microtransit services oﬀered a better level
of service than TNCs at East and Northeast Austin. At the same time, TNCs served better
at Walnut Creek compared to microtransit service. However, Walnut Creek showed a higher
microtransit demand, which indicates that the level of service diﬀerence is not inﬂuential on
The results described indicate that, among the two shared mobility modes servicing the
same area, microtransit trips seem to present the highest user demand. Their trips tend to
last longer, and in some cases, the level of service diﬀerence did not inﬂuence user demand.
Likely, the FMLM trips within these areas are not predominant, explaining the diﬀerence in
demand for TNCs services compared to microtransit.
(a) Trip distance
(b) Trip duration
(c) Driver reach time
Figure 6: Description of TNC and microtransit trips
5.2. Diﬀerence-in-diﬀerences results
The quasi-experiment consists of a DID analysis of the integration of microtransit into the
ﬁxed-route services. A DID model was used to understand the eﬀect of microtransit in
altering bus transit demand. The case study of Austin, Texas, revealed that the mobility
innovation zones envisioned by the public transit agency proved to be useful for travel within
the zones, as seen from the demand for microtransit. However, the results suggest that
microtransit had no signiﬁcant impact on transit use in three of the four mobility zones.
(a) Exposition zone (b) Exposition zone with buﬀer
(c) East ATX zone (d) East ATX zone with buﬀer
(e) Northeast ATX zone (f) Northeast ATX zone with buﬀer
(g) Walnut Creek ATX zone (h) Walnut Creek ATX zone with buﬀer
Figure 7: Average transit demand trends of the mobility innovation zones and the control groups before
and after the microtransit service
Figure 7presents the monthly bus transit demand of the treated groups compared to
the control groups. The dashed lines represent the average value for the demand in the
pre-intervention and post-intervention periods. The left column presents the zone analysis,
while the right column shows the zone analysis when including the buﬀer areas. East Austin
and Walnut Creek buﬀer areas had a signiﬁcantly higher transit demand compared to the
control group. Thus, the control group is not included in the plot to show the monthly
changes in the demand trend. Although the control groups were selected based on similarity
in resident population characteristics and bus transit service characteristics, a signiﬁcant
diﬀerence in transit demand was inherent between the control and treatment zones even
before treatment. This ﬁnding is expected, but the trend in transit use across time for the
control and treatment groups remained parallel from one month to the next, satisfying the
parallel trend assumption of the DID model.
Table 3shows the estimated coeﬃcients for each of the four zones studied, as well as the
estimates of the zones, including the buﬀered area around the mobility innovation zones.
The coeﬃcient β1is an indicator of the time trend in the control groups and the treatment
group’s expected tendency. The results indicate that the bus transit demand increased over
time except for the Exposition zone, where the demand decreased. The mobility innovation
zone was envisioned for neighborhoods deemed transit deserts, so this result is intuitive for
the three regions (with better transit use in the buﬀered area). Exposition is located close
to Austin’s downtown, and residents have a higher than average household income that
likely translates to lower transit reliance. The clear delineation by the MoPac expressway to
the east of this mobility zone also supports the continued lower trend in transit use in the
The diﬀerence in transit demand between the control and treatment zones is captured by
the coeﬃcient β2, and is largely negative across all but one zone – East ATX, which implies
that the control group in this area showed a lower transit ridership than the treatment
group. The magnitude of β2is higher for the buﬀer areas, as expected. In the buﬀer case,
the treatment group included a higher number of TAZs, and therefore, it captured a higher
transit ridership. However, it can be highlighted that the magnitude in East ATX increased
signiﬁcantly when the buﬀer area was included. This result is signiﬁcant because it captures
the transit demand in one of the main terminals of CapMetro, a place that includes several
The primary objective of this paper was to test the diﬀerential impact of bus transit
use with microtransit availability in these mobility zones. Prior studies have been divided
on whether microtransit or TNC availability for FMLM transit access helps improve transit
demand. The DID estimate is captured by the coeﬃcient β3or the ATT (refer to Equation
3), which will directly assess the microtransit treatment eﬀect on transit trips. The DID
analysis found no signiﬁcant diﬀerence in transit use based on the introduction of microtransit
use in Exposition, East Austin, and Walnut Creek neighborhoods, but found a statistically
signiﬁcant decrease in transit use in Northeast Austin.
Table 3: DID Results
Coeﬃcient Estimate p-value Estimate p-value
Zone: Exposition Exposition with buﬀer
β028,641 0.00 * 28,641 0.00 *
β1-1,302 0.35 -1,302 0.37
β2-16,467 0.00 * -14,527 0.00 *
β3495 0.80 604 0.77
Adj. R20.92 0.89
Zone: East ATX East ATX with buﬀer
β029,428 0.00 * 29,428 0.00 *
β13,435 0.01 * 3,435 0.51
β23,200 0.00 * 159,805 0.00 *
β3-738 0.69 50 1.00
Adj. R20.42 0.99
Zone: Northeast ATX Northeast ATX with buﬀer
β051,150 0.00 * 51,150 0.00 *
β14,139 0.01 * 4,139 0.05 *
β2-13,490 0.00 * 23,566 0.00 *
β3-4,227 0.05 * -1,805 0.54
Adj. R20.88 0.92
Zone: Walnut Creek Walnut Creek with buﬀer
β035,984 0.00 * 35,984 0.00 *
β11,071 0.45 1,071 0.73
β2-6,661 0.00 * 79,701 0.00 *
β3-1,000 0.62 -718 0.87
Adj. R20.67 0.98
*Signiﬁcant at 95 percent conﬁdence level
The three mobility zones not showing a diﬀerence in transit use are quite diﬀerent in
household and socio-economic characteristics. The key diﬀerence between these three zones
and Northeast Austin was the ratios of population to employment density, which means
that many short trips may have been possible within these mobility zones. Contrary to
that, Northeast Austin had an imbalance in trip-generations and trip-attractions. This may
intuitively mean that this neighborhood sees a larger share leaving the zone for work, which
would also explain that this zone had the highest number of TNC trips. The high average
household size may mean that a microtransit FMLM trip to a transit route may not have
been more attractive than a personal vehicle available at the origin location.
These results can be inferred from two lenses. From the transit operator’s perspective,
the deployment of microtransit in transit deserts helps provide mobility options for travel
within the demarcated zone. Travelers cannot be expected to choose to transfer from a
microtransit ride to a ﬁxed-route transit line to reach their destinations outside these zones.
Suppose the transit operator’s primary objective is to provide mobility options while saving
costs associated with unused ﬁxed-route transit lines. In that case, mobility innovation zones
in locations that may not be transit-dependent but have suﬃcient zonal trip generations and
attractions can be expected to ﬁnd the microtransit service useful. However, the microtran-
sit service does not help increase ﬁxed-route transit ridership. From a system perspective,
microtransit geofenced in transit deserts may not help improve travelers’ mobility and ac-
cess at the system level. Personal auto ownership may continue to be essential to access
destinations outside of these zones.
6. Summary and Conclusions
This study analyzed the integration of two shared mobility services into the public transit
service using data from pilot programs oﬀering TNC and microtransit services in the same
area during diﬀerent periods. Ridership data was used to compare the demand for transit,
TNC, and microtransit and describe the trips’ principal characteristics, including trip dis-
tance, duration, and driver reaching times. Furthermore, a DID quasi-experimental method
was developed using monthly transit ridership data to analyze the causal relationships be-
tween the microtransit service and bus demand. The main results suggest that TNC services,
focusing only on FMLM, were not as popular as microtransit services oﬀering FMLM and
intrazonal trips. Even though the microtransit trips were longer and, in some cases, the
driver response times were higher than for TNC trips.
This case study reﬂects the challenges that public transit agencies face in areas with low
demand, where ﬁxed-route services are not cost-eﬀective. Although the results indicate that
there is no integration between these shared mobility options and the transit system, the
beneﬁts of the program go beyond measurable demand and costs. There is a need for per-
formance metrics that can capture the beneﬁts of these options for both users and providers,
capable of measuring the improved mobility, increased safety, and enhanced customer expe-
rience that these users experienced.
The results and methods proposed in this study can serve multiple purposes. First,
from the transit agency point of view, this study provides a methodological framework for
analyzing mobility services’ integration into the transit system that can be used in other areas
with similar PPPs services. Second, it evaluates how these partnerships can potentially be
used to help under-served areas and improve transit accessibility from the users’ point of
view. Finally, from a transportation research point of view, this study contributes to the
scarce literature in this area providing empirical evaluations of the eﬀects of incorporating
these services into the transit system.
The authors would like to thank Capital Metropolitan Transportation Authority for provid-
ing data for the demand response services.
Bian, Z., Liu, X., 2019. Mechanism design for ﬁrst-mile ridesharing based on personalized
requirements part i: Theoretical analysis in generalized scenarios. Transportation Research
Part B: Methodological 120, 147–171.
Blodgett, M., Khani, A., Negoescu, D., Benjaafar, S., 2017. Public/Private Partnerships in
transit: Case studies and analysis. Technical Report. Minnesota Council on Transportation
CapMetro, C., 2018. Cap Remap: Austin changed and then we changed. Technical Report.
El-Geneidy, A., Grimsrud, M., Wasﬁ, R., T´etreault, P., Surprenant-Legault, J., 2014. New
evidence on walking distances to transit stops: Identifying redundancies and gaps using
variable service areas. Transportation 41, 193–210.
Feigon, S., Murphy, C., 2016. Transit Cooperative Research Program (TCRP) Research
Report 188: Shared mobility and the transformation of public transit. Project J-11, Task
Feigon, S., Murphy, C., McAdam, T., 2018. Transit Cooperative Research Program (TCRP)
Research Report 196: Private Transit: Existing services and emerging directions. Project
J-11, Task 24.
Giuliano, G., 2003. Travel, location and race/ethnicity. Transportation Research Part A:
Policy and Practice 37, 351–372.
Gurumurthy, K.M., Kockelman, K.M., Zuniga-Garcia, N., 2020. First-Mile-Last-Mile
Collector-Distributor System using Shared Autonomous Mobility. Transportation Re-
search Record 2674, 638–647.
Hall, J.D., Palsson, C., Price, J., 2018. Is uber a substitute or complement for public transit?
Journal of Urban Economics 108, 36–50.
Huang, Y., Kockelman, K.M., Garikapati, V., Zhu, L., Young, S., 0. Use of shared auto-
mated vehicles for ﬁrst-mile last-mile service: Micro-simulation of rail-transit connec-
tions in austin, texas. Transportation Research Record 0, 0361198120962491. URL:
Jin, S.T., Kong, H., Sui, D.Z., 2019. Uber, public transit, and urban transportation equity:
A case study in new york city. The Professional Geographer 71, 315–330.
Lader, C., Klein, N., 2018. Westchester County Bee-Line system ﬁrst and last mile connec-
tions mobility study. Westchester County Department of Public Works and Transporta-
tion, February .
Lavieri, P.S., Dias, F.F., Juri, N.R., Kuhr, J., Bhat, C.R., 2018. A model of ridesourc-
ing demand generation and distribution. Transportation Research Record 2672, 31–40.
URL: https://doi.org/10.1177/0361198118756628, doi:10.1177/0361198118756628,
Lucken, E., Frick, K.T., Shaheen, S.A., 2020. “Three Ps in a MOD:” Role for mobility
on demand (MOD) public-private partnerships in public transit provision. Research in
Transportation Business & Management , 100433.
Mishra, S., Mehran, B., Sahu, P.K., 2020. Assessment of delivery models for semi-ﬂexible
transit operation in low-demand conditions. Transport Policy 99, 275–287.
Pike, S., Kazemian, S., 2019. Partnerships between Ridehailing Companies and Public Tran-
sit Agencies: An Exploration of Inter-agency Learning about Pilot Programs. Technical
Report UC-ITS-2018-17. University of California, Davis.
Pike, S., Kazemian, S., 2020. Inﬂuential Factors in the Formation of Partnerships Between
Ridehail Companies and Public Transportation. Technical Report UC-ITS-2019-08. Uni-
versity of California, Davis.
Reck, D.J., Axhausen, K.W., 2020. Subsidized ridesourcing for the ﬁrst/last mile: How
valuable for whom? European Journal of Transport and Infrastructure Research 20,
Shaheen, S., Chan, N., 2016. Mobility and the sharing economy: Potential to facilitate the
ﬁrst-and last-mile public transit connections. Built Environment 42, 573–588.
Shaheen, S., Cohen, A., Zohdy, I., 2016. Shared mobility: Current practices and guiding
principles. Technical Report FHWA-HOP-16-022. U.S. Department of Transportation.
Smart, M.J., 2015. A nationwide look at the immigrant neighborhood eﬀect on travel mode
choice. Transportation 42, 189–209.
Smith, C., 2019. Amazing uber stats and facts (2019). URL: https://expandedramblings.
Stiglic, M., Agatz, N., Savelsbergh, M., Gradisar, M., 2018. Enhancing urban mobility:
Integrating ride-sharing and public transit. Computers & Operations Research 90, 12–21.
Tec, M., Zuniga-Garcia, N., Machemehl, R.B., Scott, J.G., 2019. Large-scale spatiotemporal
density smoothing with the graph-fused elastic net: Application to ride-sourcing driver
productivity analysis arXiv:1911.08106.
Terry, J., Bachmann, C., 2020. Spatial characteristics of transit-integrated ridesourcing trips
and their competitiveness with transit and walking alternatives. Transportation research
record 2674, 329–340.
Westervelt, M., Huang, E., Schank, J., Borgman, N., Fuhrer, T., Peppard, C., Narula-Woods,
R., 2018. UpRouted: Exploring Microtransit in the United States. Technical Report. Eno
Center for Transportation.
Yahia, C.N., de Veciana, G., Boyles, S.D., Rahal, J.A., Stecklein, M., 2020. Book-ahead &
supply management for ridesourcing platforms. arXiv preprint arXiv:2002.09664 .
Yan, X., Levine, J., Zhao, X., 2019. Integrating ridesourcing services with public tran-
sit: An evaluation of traveler responses combining revealed and stated preference data.
Transportation Research Part C: Emerging Technologies 105, 683–696.
Zuniga-Garcia, N., Machemehl, R.B., 2020. Dockless electric scooters and transit use in an
urban/university environment, in: 99th Annual Meeting of the Transportation Research
Board, Washington, DC.
Zuniga-Garcia, N., Scott, J.G., Tec, M., Machemehl, R.B., 2020a. Evaluation of e-scooters
as transit last-mile solution. Under review for publication.
Zuniga-Garcia, N., Tec, M., Scott, J.G., Ruiz-Juri, N., Machemehl, R.B., 2020b. Evaluation
of ride-sourcing search frictions and driver productivity: A spatial denoising approach.
Transportation Research Part C: Emerging Technologies 110, 346–367.