With the rise of app-based, on-demand transportation network companies (TNCs) like Uber and Lyft, public transportation agencies began piloting their own on-demand services to test whether the technology could attract new riders or improve service for existing riders. These pilots multiplied in the Covid-19 pandemic, when fixed-route ridership plummeted, as a way to serve remaining transit-dependent riders and essential workers with fewer vehicles. Questions about these publicly funded mobility on demand (MOD) and microtransit services focus on their environmental, efficiency, and transportation equity impacts: do they reduce vehicle miles traveled (VMT) by attracting drivers away from personal vehicles and TNCs, or increase it by inducing trips and shifting trips from non-motorized / higher-occupancy fixed-route modes? Can they serve the same number of passengers per hour as low-ridership bus routes, at lower cost? Do they disproportionately benefit low-income, transit-dependent populations or well-resourced groups that could afford Uber/Lyft?

This abstract describes research objectives and overarching methodology, identifies methodological and empirical contributions, highlights key results from these contributions, and concludes with policy implications and a summary.

Research Objectives: To evaluate the environmental and transportation equity impacts of publicly subsidized on-demand services, I sought to understand: 1) what types of public MOD services currently exist, including benefits and challenges of each type from an environmental, transportation equity, and operational lens; 2) who uses and benefits from public MOD services, how this shifted in the Covid-19 pandemic, and what modes public MOD riders used before; and 3) how public MOD riders perceive public MOD relative to drive alone, TNCs, and fixed-route transit, what service and demographic characteristics impact their likelihood of using each mode, and how this changed in the Covid-19 pandemic.

Methodology: To accomplish these objectives, I completed the following: 1) developed a conceptual framework for understanding variation in the design of public MOD services (Chapter 2); 2) surveyed public microtransit riders and non-riders at 14 public transportation agencies across the U.S. and analyzed the results for demographic variation in frequency of public microtransit use before and during the Covid-19 pandemic, as well as induced trips, replaced modes, and satisfaction with the service (Chapter 3); and 3) designed a stated preference experiment to include in the survey to test the impact of public microtransit service characteristics and Covid-19 conditions on mode choice; estimated a mode choice model from the stated preference results, including a flexible variance/covariance matrix with individual variation in shared error terms between modes; and used the mode choice model to forecast mode shares for a range of demographic groups and service designs (Chapter 4). After examining publicly subsidized MOD services more generally for the typology, I focused on van-based public microtransit services for the stated preference survey and the mode-choice model due to the increasing popularity of that service type.

Methodological Contributions: The dissertation included three main methodological contributions:

1) Typology of partnerships between public transportation agencies and MOD companies (Chapter 2): As of 2019, there was no agreed-upon conceptual framework for understanding the myriad of pilot projects between public transportation agencies and companies offering on-demand, app-based transportation, in particular one that included van-based microtransit services as well as companies offering ride-matching and routing algorithms for public agencies to deploy on public vehicles. I addressed this gap by conducting expert interviews and a comprehensive review of existing partnerships to develop a typology of partnerships, including the benefits and disadvantages of each, trends in the implementation of each type, and potential environmental and transportation equity implications. My typology development was informed by 34 structured interviews with experts at four MOD companies, 21 public transit agencies, two transportation authorities, and three cities (from October 2017 through April 2018).

2) Mode choice preferences among public microtransit riders / non-riders (Chapter 4): I developed one of the first mode choice models to incorporate public microtransit using stated-preference data, and the first to use an enriched sample of experienced public microtransit passengers. I designed and administered a stated preference survey to 1519 public microtransit riders and 320 non-riders at 14 public transportation agencies across the U.S. from April through December 2021. I used the results to estimate a mode choice model that includes public microtransit, drive alone, fixed-route bus, ride-alone Uber/Lyft, pooled Uber/Lyft, and two first-mile/last-mile (FMLM) modes: public microtransit to fixed-route bus and Uber/Lyft to fixed-route bus. Using the mode choice model, I forecasted how public microtransit mode share varies with changes in service variables like cost, frequency of delay, and walk, wait, and in-vehicle times. I also forecasted changes in mode shares for systems without public microtransit, with FMLM microtransit, and with door-to-door microtransit that replaces fixed-route bus service.

3) Flexible variance/covariance matrix from shared error components (Chapter 4): In developing the mode choice model, I applied an innovative method to estimate a flexible and individual-specific covariance matrix of the error terms that is specific to the public microtransit context. Because the covariance matrix determines substitution between public microtransit and the other modes when applying the model to forecast impacts or estimate replaced modes from public microtransit services, it is important to estimate parameters to allow for flexible substitution between modes. For example, the correlation between public microtransit and fixed-route bus and the correlation between public microtransit and TNCs may both be non-zero, may not equal each other, and may also vary across individuals. The commonly used logit and nested logit models cannot capture these effects. Cross-nested logit can capture both correlation values (and more), but these values cannot vary by individual or capture correlation across responses from a single individual. While probit is fully flexible and can capture the full covariance matrix as well as correlation across responses, it is computationally burdensome and the estimated parameters are not readily interpretable as correlation between the modes. Therefore, I employed shared error components (mixed logit) to identify a flexible covariance matrix that was supported by the data, allowed clear interpretation of correlation between modes, and captured correlation across responses from the same individual. This method also enabled me to test individual variation in perceived correlation based on individual characteristics such as experience with public microtransit and access to a personal vehicle by interacting the error components with these characteristics.

Empirical Contributions: The dissertation offered two main empirical contributions:

1) Comprehensive database of partnerships between public transportation agencies and MOD companies (Chapter 2): I compiled a database of projects in the U.S. through April 2019 in which public agencies and private companies collaborated to provide MOD services that meet the FTA definition of public transportation: “regular, continuing shared-ride surface transportation services that are open to the general public or open to a segment of the general public defined by age, disability, or low income” (FTA 2016). The database included 62 partnerships with information on participating public and private entities, start/end dates, fares, eligibility restrictions, and any publicly available evaluation metrics from the project. The UC Berkeley Transportation Sustainability Research Center adopted the database, which contributed to a key research element in a major research project for Ford Motor Company.

2) Multi-agency survey of public microtransit riders (Chapters 3 and 4): Due to privacy concerns, microtransit software companies do not collect demographic data about those who register to receive rides through their apps, leaving public transportation agencies without a direct way to measure equity impacts of the public microtransit services. My stated preference survey, described above, also collected demographic data and mode use patterns (from 1519 public microtransit riders across 14 U.S. public transportation agencies from April through December 2021). I identified demographic differences in frequency of public microtransit use, induced trips, replaced modes, perceptions of the service, change in use during the Covid-19 pandemic, and anticipated change in use after the pandemic (Chapter 3). Demographic dimensions included income, age, race, gender, ability, vehicle access, and two categories of particular salience for app-based, on-demand modes: smartphone ownership and banked/unbanked status.

Typology Results (Chapter 2): I identified four main types of partnerships: 1) first-mile/last-mile (FMLM), 2) low-density, 3) off-peak, and 4) paratransit. I also identified two additional dimensions by which to characterize these partnerships to understand environmental and equity impacts: 1) asset contribution and 2) vehicle type. For asset contribution, public partners either: a) contract with the private partner to provide the entire service or b) provide the vehicles and drivers and contract with the private partner for the algorithm and app. I found that the public agencies have more control over the service design and trip data when they operate the service with their own vehicles and drivers, which impacts their ability to measure and address any disparities in service or competition with fixed-route service. Vehicle type included van, taxi, or drivers’ personal vehicles, and affects potential occupancy and VMT impacts of the service. I found that initial pilot projects were dominated by agency-subsidized TNC services, likely due to ease of implementation, but that agency-operated services using publicly owned vans were becoming more common. Most early partnerships offered “low-density” door-to-door service in a specified zone during the same hours as fixed-route transit, while the next most frequent use case was FMLM, in which subsidized trips had to start or end at a public transit stop.

Demographic Survey Results (Chapter 3): I found that underserved groups report higher rates of induced trips and are more likely to switch to public microtransit from non-motorized/higher-occupancy modes, while more privileged groups are more likely to switch from lower-occupancy modes like drive alone and Uber/Lyft. More privileged groups are also more likely to use public microtransit as a FMLM connection to fixed-route bus/rail and for work commutes, and less privileged groups are more likely to use it for errands, healthcare appointments, and social/recreational trips. Though low-income groups, riders without access to a personal vehicle, and women report greater satisfaction and quality of life improvements from public microtransit than nonmarginalized groups, riders without smartphones and/or bank accounts report lower satisfaction and no statistically significant quality of life improvements. Riders from less privileged groups also perceive more frequent delays past the public microtransit vehicles’ estimated time of arrival (ETA). Regarding Covid-19 impacts, low-income riders, those without vehicle access, and essential workers maintained public microtransit use during the pandemic but decreased their use of fixed-route bus/rail, while other groups decreased their use of fixed-route bus/rail by even more and also decreased their use of public microtransit.

Mode Choice and Shared Error Results (Chapter 4): The forecasts suggested that the combined mode share of fixed-route bus and microtransit is lowest in systems that replace fixed-route bus service with public microtransit and highest in systems with supplemental, door-to-door public microtransit, while the reverse is true for personal vehicle mode share. The model results also showed that respondents without smartphones were more likely than other users to use fixed-route bus, while forecasts found that smartphone ownership and access to a personal vehicle have a larger impact than income on mode splits. Regarding Covid-19 impacts, riders perceive fixed-route bus service more negatively than public microtransit as local in-patient rates increase, while trip cost has less of an impact on mode choice during the Covid-19 pandemic. Regarding the shared error components, perceived correlation between modes varies with vehicle access and familiarity with the public microtransit service. Respondents who sometimes or never have access to a personal vehicle view door-to-door public microtransit as most correlated with drive alone, while vehicle owners perceive less correlation between the two modes. Experienced microtransit riders, meanwhile, view public microtransit as more correlated with Uber/Lyft than with fixed-route bus, while those who have never used public microtransit perceive it as much more correlated with fixed-route bus than Uber/Lyft.

Policy Implications: I identified several policy implications based on my results:

• Public transit agencies should exercise caution when replacing fixed-route service with public microtransit, as forecasting suggests steep declines in overall public transit ridership in this scenario (Chapter 4).• Replacing traditional paratransit and dial-a-ride services with app-based public microtransit offers equity benefits through improved experiences for transit-dependent populations, provided the agency ensures continued call-in and cash options for passengers without smartphones and/or bank accounts (Chapters 2, 3). • More work needs to be done to improve public microtransit service for riders without smartphones or bank accounts. These groups do not report the same benefits from public microtransit services and are the only underserved groups to anticipate reducing their use of public microtransit by as much or more than fixed-route transit after the Covid-19 pandemic. Investing in improved fixed-route service may be a more effective way of serving these riders (Chapters 3, 4). • For new public microtransit services that do not replace fixed-route bus or a pre-existing paratransit or dial-a-ride service, environmental and equity goals may conflict. Public transportation agencies may need to choose between: 1) advancing environmental goals by targeting higher-income riders switching from lower-occupancy modes like drive alone or 2) improving transportation equity by targeting lower-income riders, who are more likely to switch from non-motorized or higher-occupancy modes (Chapter 3).

Summary: This dissertation analyzed the environmental and transportation equity impacts of public MOD and microtransit services through: a comprehensive review and typology of such services in the U.S., demographic analysis of the largest multi-agency survey of public microtransit passengers so far, and a mode choice model with shared error components to understand heterogeneity in preferences among riders and non-riders before and during the Covid-19 pandemic. I found that the environmental and equity goals of public microtransit services may conflict, as riders from underserved groups report higher rates of induced trips and switch disproportionately away from non-motorized / higher-occupancy modes. While public microtransit appears to address barriers that riders with disabilities faced from TNCs, results indicate continued disparities for riders without smartphones and/or bank accounts. Future research could identify displaced riders when fixed-route bus services are replaced with public microtransit services, perhaps through a longitudinal survey, to more fully capture equity impacts. Next steps also include using the mode choice model I estimated with agent-based simulation to compare local environmental and equity impacts of investing in public microtransit versus improved fixed-route bus service.