Many studies on preferences for AVs have focused on factors associated with private AV ownership, such as consumers’ perceived comfort with riding in an AV and their willingness to pay for features associated with different levels of automation (Daziano et al., 2017; Moody et al., 2020; Nair & Bhat, 2021; Shin et al., 2015). These studies often employ Likert scales (typically ranging from 1 to 5) or other similar rating systems to assess attitudes towards different automated modes. While these studies provide insights into general consumer perceptions of AVs, they lack the ability to gauge potential substitution patterns between automated and non-automated transportation modes.
Table 1: Summary of modes investigated in prior AV conjoint studies.
|
|
Non-Automated
|
Automated
|
Study
|
Location
|
Private Car
|
Ride-hailing
|
Shared Ride-hailing
|
Transit
|
Other (walking, biking)
|
Private Car
|
Ride-hailing
|
Shared Ride-hailing
|
Transit
|
Krueger et al. 2016*
|
Australia
|
X
|
X
|
X
|
X
|
X
|
|
X
|
X
|
|
Yap et al. 2016
|
Netherlands
|
X
|
|
|
X
|
X
|
|
X
|
X
|
|
Steck et al. 2018
|
Germany
|
|
|
|
X
|
X
|
X
|
X
|
X
|
|
Ashkrof et al. 2019
|
Netherlands
|
X
|
|
|
X
|
|
|
X
|
|
|
Winter et al. 2020
|
Netherlands
|
X
|
X
|
|
X
|
|
|
X
|
|
|
Etzioni et al. 2020
|
Cyprus, UK, Slovenia, Montenegro, Hungary, Iceland
|
X
|
|
|
|
|
X
|
X
|
X
|
|
Daziano et al. 2017
|
U.S.
|
X
|
|
|
|
|
X
|
|
|
|
Haboucha et al. 2017
|
U.S., Canada & Israel
|
X
|
|
|
|
|
X
|
X
|
|
|
Lavieri & Bhat 2019
|
U.S.
|
|
|
|
|
|
|
X
|
X
|
|
Gurumurthy & Kockelman 2020
|
U.S.
|
|
|
|
|
|
X
|
X
|
|
|
Zhong et al. 2020
|
U.S.
|
X
|
|
|
|
|
X
|
X
|
X
|
|
This study
|
Washington, D.C.
|
|
X
|
X
|
X
|
|
|
X
|
X
|
X
|
*Non-automated modes were collected as a self-reported reference trip
|
To address this, some researchers have used choice-based conjoint (CBC) surveys. In CBC surveys, respondents choose from a set of options with varying attributes, and researchers estimate discrete choice models to infer the relative importance of each attribute and the relative desirability of each option. Conjoint surveys offer unique advantages, including the ability to explore hypothetical products, present multiple choice sets to the same respondent, fix all attributes of a given option, and avoid multicollinearities (Louviere et al., 2000). Rather than gauge preferences for different modes in isolation, conjoint surveys allow researchers to simulate the menu of transportation options available to an individual, typified by the experience of looking up directions via GoogleMaps or via a transportation planning app. Table 1 presents a selection of recent CBC studies investigating public preferences for different automated and non-automated transportation modes. The majority of these prior studies compare automated modes to conventional, non-automated private cars.
The general consensus across most of these studies is that conventional, non-automated vehicles continue to dominate preferences. For example, in Krueger et al.’s (2016) conjoint study of 435 residents of major metropolitan areas in Australia, respondents chose an automated mode in only 28% of the choice situations. Haboucha et al. (2017) surveyed 721 commuters in Israel, the United States, and Canada and found that 44% of respondents preferred conventional vehicles over private or shared AVs. This preference was even more pronounced among the North American respondents, with 54% preferring conventional vehicles. Yap et al. (2016) asked Dutch travelers about their interest in AVs as a transportation option for filling the last mile trip between a train station and a traveler’s final destination, and even in this limited context respondents mostly selected the individual vehicle alternative over all other transportation options. Etzioni et al. (2020) surveyed 1,669 individuals across six EU countries and similarly found strong preferences for conventional vehicles, with respondents selecting conventional vehicles in 70% of the choices. Respondents in Zhong et al.’s (2020) survey of U.S. residents in small and medium metropolitan areas in the U.S. preferred their current private vehicles over private AVs and AV ride-hailing options. These studies signal that individuals are not likely to relinquish their personal vehicles in favor of AVs in the near future.
The strong preferences for conventional vehicles, however, may mask other potential substitution effects that could occur with the introduction of AVs. Many of the aforementioned studies restricted their survey sample to individuals who have a driver’s license, with some also requiring that respondents drive a personal vehicle frequently (Ashkrof et al., 2019; Haboucha et al., 2017; Zhong et al., 2020). In doing so, these studies fail to capture the preferences of individuals who do not currently rely on personal vehicles as their primary mode of transportation, including individuals with disabilities or those who do not own a car. Such individuals are typically the primary users of public transit. Moreover, even individuals who typically use their private vehicles might use transit for specific types of trips (e.g., traveling within the city after commuting from the suburbs, trips where parking is expected to be difficult, etc.). Substitutions of these trips with AVs, in conjunction with changing transportation patterns for frequent transit users, would likely have a greater impact on transit ridership.
The existing literature on AV substitution with transit is limited and inconclusive. Some studies find a preference for AVs over transit modes, such as in Steck et al.’s (2018) survey of 173 Germans. In the study, respondents could select from among privately owned AVs, automated ride-hailing (both shared and non-shared), walking, biking, and public transit. Overall, respondents found the private AV option most attractive, followed by AV ride-hailing, and finally transit. Ashkrof et al. (2019) explored preferences for conventional cars, AV ride-hailing, and transit among a sample of 663 Dutch respondents. Individuals similarly preferred AVs over transit, especially when the choice question was framed in terms of a long-distance trip. Yet other studies suggest more limited competition of AVs with transit. In Yap et al.’s (2016) study of AVs as a potential egress mode for train trips, first-class train passengers valued AVs more than transit modes, but second-class train passengers actually preferred transit over AVs. Winter et al. (2020) identified different classes of users amongst a sample of 796 Dutch survey respondents and found that respondents who currently commute by public transport actually show the lowest preference for automated modes, affirming Krueger et al. (2016)’s finding that current transit users were not more likely to switch to an automated mode.
There are some mode features that are particularly relevant when considering an AV future. Ride-sharing (i.e., riding with a stranger who is traveling in a similar direction) is already available in some cities via services like UberPool and Lyft Shared[1] (Lyft, 2022b; Uber, 2022b). Sharing rides decreases the cost for both riders, and these cost savings could become even more substantial if the services are automated. Further, sustainability advocates emphasize that fleets of shared AVs are critical for ensuring a sustainable AV future (Creger et al., 2019). Despite enthusiasm from environmental advocates, the public seems less interested in a future of shared rides. As of 2017, pooled rides comprised just 20% of all Uber rides and 40% of all Lyft rides (Shaheen & Cohen, 2019), and current literature suggests that these preferences may persist in an AV future. In Lavieri and Bhat’s (2019) conjoint study on automated ride-hailing with sharing and non-sharing options, respondents chose to ride alone in 48.3% of choice occasions with work trips and 54% of choice occasions for leisure trips. Over the past few years, greater exposure to ride-sharing services as well as the COVID-19 pandemic may have altered individuals’ attitudes towards sharing. Thus, sharing as a feature of automated ride-hailing services warrants further investigation.
A second mode feature—the presence of an AV attendant—is associated with additional services that a driver might fulfill beyond operating the vehicle. Though AVs would be operated by computer systems, an attendant could help individuals enter and exit the vehicle—a potential barrier to AV use for elderly individuals and individuals with disabilities—and provide a social monitoring function. This monitoring function might affect who feels comfortable using shared AV services. The consensus from many stated preference surveys and choice studies on AVs is that women appear less likely to use AVs than men (Becker & Axhausen, 2017; Gkartzonikas & Gkritza, 2019), and some hypothesize that this hesitation towards AVs may stem in part from personal security concerns (Khoeini, 2021; Nair & Bhat, 2021; Polydoropoulou et al., 2021). Dong et al. (2019)’s survey of University of Pennsylvania employees found that only 13% of respondents would agree to ride an automated bus without an employee onboard. Similarly, in their multi-country survey on potential AV use, Kyriakidis et al. (2020) asked respondents about their willingness to use an AV when a human operator was and was not onboard. The study found that people were more willing to travel in an AV and to allow their children to travel in an AV with an operator present. These findings suggest that operator presence might be an important feature that might impact whether individuals would prefer AVs over traditional modes. While some AV companies are already operating their vehicles with attendants onboard in small pilots (Fort Worth Business Press, 2021), companies will eventually need to decide whether the attendant feature is worth the additional operating cost in large-scale deployments.
Conjoint studies have enabled an avenue of research to explore potential substitution patterns between various transportation modes in an AV future. This area of research, however, is still quite immature, with many studies conducted only within the past six years. Few studies have considered the impacts of AVs on current transit use, and no conjoint studies have examined impacts of AVs on transit in a U.S. context. Furthermore, there is a lack of understanding about key features associated with AV use, such as ride-sharing and the presence of an AV attendant. We address these gaps by fielding a U.S.-based conjoint study on preferences for automated (ride-hailing, shared ride-hailing, bus) and non-automated (ride-hailing, shared ride-hailing, bus, rail) modes.
1 Though ride-hailing companies canceled these services during the COVID-19 pandemic, some are now starting to reintroduce them.