Public priorities and consumer preferences for selected attributes of automated vehicles

Public opinion and consumer preferences are among the various constraints on the rollout of automated cars, as they will affect the decision-making of both automotive industry actors and public-sector regulators. This study contributes to the growing body of the literature regarding this issue, through a moderate-scale survey (n = 370) that incorporated both prioritization/attitudinal questions (regarding public opinion) and a stated-preference module (to identify consumer preferences). The survey protocol includes a stated-preference approach to investigate consumers’ preferences for the possibility of very high rates of speed in automated cars on long-distance journeys. We found separately identifiable effects for average travel speeds (manifested as journey duration) and maximum travel speed in the stated-preference scenarios. In the ‘prioritization’ component of the survey, respondents ranked having the ‘highest possible level of safety’ as the single most important benefit that they would like automated cars to deliver, ahead of benefits such as being able to performing activities while traveling or having traffic congestion reduced. This result has consequences for the car-following distances that are programmed into the control algorithms of automated cars. Documenting this finding is important, as decisions must be made in the near future by driving-algorithm designers, public-sector regulators, and ultimately the judiciary regarding the guidelines for acceptable automated driving-behavior instructions.


Introduction
The consequences of vehicle automation are potentially far reaching; however, the body of the literature regarding consumer preferences and public opinion is now in its formative stages (Table 1 summarizes these recent studies [1][2][3][4][5][6]) and relatively little is known with reliability.
The objective of this study is to identify novel aspects of consumer preferences and public opinion regarding highly automated cars (levels 2 through 4 under the NHTSA taxonomy [7]). For instance, we investigate consumers' relative prioritization of various prospective benefits of automated cars and their preferences for travel at higher rates of free-flow speed than at present; neither of these issues are addressed in the prior literature. To do this, we designed a survey protocol that included both attitudinal questions and a stated-preference component. The survey was administered to a moderate-sized sample (n = 370), which was generated according to a sampling plan based on age/gender quotas (the sampling strategy is described in detail in Sect. 3). Therefore, the results can be interpreted as being nationally representative along these dimensions; however, larger sample (n [ [ 1000) follow-on surveying will be required to have greater confidence in the findings.
The remainder of this paper is organized as follows. Section 2 describes the survey protocol that we designed and executed in this study. Section 3 then presents and discusses the empirical results, and Sect. 4 summarizes and concludes this paper.

Overview of survey protocol
The survey protocol comprised three distinct modules, which respondents completed in the following sequence: • Demographic information.
The survey protocol was reviewed and approved by the SUNY New Paltz Human Research Ethics Board and pilottested with college staff as respondents prior to the main fieldwork. The survey fieldwork was performed via webbased computer-aided self-administration (CASI). The fieldwork was undertaken by the market-research firm Qualtrics, which maintains market-research panels to The survey questionnaire was identical across the four batches, with the exception of the numerical values in the stated-preference module. The logic for varying the numerical values in the stated-preference module is to increase the variability in the dataset, which increases the robustness of the regression parameters and minimizes the opportunity for collecting stated-preference data that do not provide useful information in model estimation (see Sect. 3).
For the purposes of communicating with respondents within the survey, we elected to consistently use the term 'driverless car' rather than 'automated car.'

Design of stated-preference module
In the stated-preference module, respondents were presented with the task of deciding how to travel to 'see relatives that live in another part of the country' and were presented the following options: • Drive a 'normal' car or take a bus (see two paragraphs below). • A 'semi-driverless' car. 2 • A 'completely driverless' car.
• Take a commercial air flight.
'Visiting relatives' was chosen as the hypothetical activity, because in comparison with other activities that were considered by the research team (e.g., business travel, tourism) it was felt that visiting relatives would likely be a type of long-distance travel that is familiar to a larger share of respondents.
The stated-preference module consisted of 10 replications for each respondent. During each replication, the cost and duration and maximum speed of each of the options in the listing above were varied according to a pre-defined D-efficient design prepared by the research team [8].
The questionnaire branched on the basis of whether or not the respondent holds a driver's license. In the statedpreference module, licensed respondents were presented with the option to drive a 'normal' (i.e., not automated) car if they wish, whereas unlicensed respondents were instead provided the option to take an inter-city bus. Also, respondents who do not hold a driving license were only presented with the option of using a 'fully automated' car (i.e., NHTSA level 4), whereas respondents holding a driving license were also presented in specific scenarios with the option of using a 'partly automated' car (NHTSA level 2/3). The questionnaires were otherwise identical, with the exception of a question about how fast drivers drive on Interstate Highways (which was not asked of unlicensed respondents and is not analyzed in this paper). Table 2 presents descriptive results of the sample regarding socio-demographics and spatial characteristics of their residence. Pluralities reported being married (42%) and living in a suburban location (46%); 70% reported household income of under $75,000/year, approximately two- 1 Personal communication from Qualtrics' S.J. Campbell, 4/14/2016. 2 The text describing the driverless cars is as follows:

Results
Carmakers are now experimenting with driverless cars, which use sensors and computers to 'drive' themselves. It works a bit like an advanced form of Cruise Control, but you tell a driverless car the destination address of your journey and it is able to brake by itself, change lanes, and make turns to get you there. In the different scenarios, you will see two types of ''driverless cars'': 1. A ''semi-driverless'' car can drive itself for nearly all of the journey (you just need to enter the destination address). However, you must keep your hands on the steering wheel and be ready to take control at any time in case something goes wrong to avoid accidents. 2. A ''completely-driverless'' car will drive itself for the entire journey, after you tell it your destination. You are free to do whatever you wish as you travel with no need to ever keep your hands on the wheel, as the car avoids accidents on its own. thirds (68%) of respondents reported living without children in their household, and 90% reported having a full (i.e., excluding learner's permits) driving license.

Prioritization and attitudinal results
After the stated-preference module (see Sect. 3.2), the survey concluded with a set of attitudinal questions. The first three of these questions asked about speed selection when driving 'manually' (not discussed further in this paper), preferences for programming instructions of autonomous cars, and prioritization of their benefits. In order to minimize primary effects for the questions consisting of agreement with statements or prioritization, the order in which these statements were presented was randomized (i.e., different for different respondents). In Tables 3 and 4, color coding is used for ease of interpretation (darkest green for the largest percentages and darkest red for the smallest percentages, excluding 'Unsure' responses).

Prioritization of automated cars' programming instructions
Respondents were presented the following paragraph of text which describes the trade-off between safety and congestion. Respondents then indicated which of the following responses best describes their view: Driverless cars will need to be programmed with instructions of how to follow behind other cars. Following closely can reduce the severity of traffic congestion, but this could increase the risk of rearend crashes. Which of these statements best describes your view of how driverless cars should be programmed: • Driverless cars should be programmed to follow closely behind the car ahead of it in traffic, in order to reduce traffic congestion, even if this increases the possibility of rear-ending the car ahead: 9%. • The person riding in a driverless car should have the choice of whether to leave a large distance behind the car ahead of it, if they wish to reduce the possibility of rear-ending the car ahead, even if this makes traffic congestion worse: 48%. • Driverless cars should be programmed to leave a large distance behind the car ahead of it, in order to reduce the possibility of rear-ending the car ahead, even if this makes traffic congestion worse: 42%.
It can be seen that only a small minority of respondents (9%) indicated that congestion reduction should be prioritized over safety. The large majority felt that either the choice of how to make this trade-off should rest with the occupant of an automated car (48%) or that automated cars should be programmed to prioritize safety over congestion reduction.  Table 3 shows results from a question in which respondents were asked to prioritize among five prospective benefits of automated cars:

Prioritization of benefits from automated cars
• Being able to read, sleep, send text messages, or do other activities inside the car besides driving, while the car does the driving. • Being able to send a driverless car to pick up or drop off packages, groceries, or children, without a human driver inside the vehicle. • Having the highest possible level of safety in a driverless car. • Having traffic congestion reduced, so that traffic moves more smoothly even when there are many cars on the road. • When there are few other cars on the road, being able to travel much faster (higher speed) than drivers are allowed to drive today.
A simple majority of respondents (51%) chose the 'highest possible level of safety' as their top priority, more than three times the proportion that selected any other item as their top priority. The most frequently cited second priority was congestion reduction, and at the opposite end of the scale, the most frequently cited item as respondents' lowest priority is the ability to send an unoccupied driverless car to perform errands. The item cited least frequently as the top priority was being able to do other activities while inside a car (9%); this is consistent with findings from Piao et al. [2] (see Table 2). Table 4 shows results from a set of questions designed by the research team to investigate attitudes thought to affect one's views regarding automated cars. In the interests of space, we comment on a subset of the results presented in Table 4. Respondents indicated strong preferences for preferring to travel by car than public transportation, yet a majority indicated that they agreed that they care about environmental issues. Fully 65% of respondents reported enjoying driving.

Attitudinal results
Majorities expressed willingness to pay more (51%) and accept slower travel (54%) in exchange for greater comfort while traveling. Nearly two out of five respondents (39%), however, indicated that speed is the most important factor in how they travel.   Table 5 contains results from the estimation of two specifications of mode choice models, using the data from the stated-preference survey. Alternative specific constants and alternative specific travel time parameters were estimated, with generic parameters estimated for journey cost and maximum travel speed [9,10]. The multinomial logit results are included for completeness; the mixed logit with panel effects model is preferred as it accounts for the fact that stated-preference responses are not each independent from one another, because each respondent performed a set of ten scenarios. Taking this 'panel effect' into account improved goodness of fit (McFadden's rho-squared) and accounts for one dimension of bias in the parameter estimates, thereby yielding parameter estimates that are closer to the theoretical 'true' values.

Stated-preference results
All parameters for travel time and cost have the expected (negative) sign, indicating that, ceteris paribus, respondents were less likely to select alternatives that were more expensive or took additional time. These diagnostic results provide a measure of confidence that the data appear reasonable and suitable for subsequent analysis. To the authors' knowledge, the result that travelers appear to (positively) value the maximum speed at which they would travel in an automated car during their journey, independently from the duration of their journey, is a novel finding; it will have consequences for how automated cars are programmed to operate on both arterial streets [12] and freeways [13]. This apparent expression of consumer preference is somewhat in tension with, for instance, the suggestion by Anderson et al. [11, p. 30] that automated cars might 'enable lower peak speeds (improving fuel economy) but higher effective speeds (improving travel time). ' A counterintuitive result from the mode choice analysis is that the parameter for travel time for the 'Semi-driverless car' option is smaller than for the 'Completely driverless car.' This is contrary to a priori expectations, as it has been theorized by researchers [13,14] that travelers in fully automated cars may have lower values (disutilities) of travel time, as being disengaged from the driving task would allow the vehicle occupant to focus on other productive or leisurely activities. Further analysis into this issue is needed to identify whether the result reported here is anomalous, or indicative of a misunderstanding on the part of researchers regarding people's preferences for the various technologies of automated cars.

Conclusions
In this study, we present results from a moderate-sample survey of the public's priorities for automated cars and preferences for their specific attributes in the context of a stated-preference survey. The context of the stated-preference survey is long-distance travel (specifically to visit relatives), and a novel result is that travelers appear to value both their 'effective' or average speed and their maximum speed (i.e., their 'cruising speed' under free-flow conditions). Results regarding the public's priorities demonstrate that safety appears to clearly be a higher priority than congestion reduction, a finding which has consequences for the car-following distances that are programmed into the control algorithms of automated cars. Documenting this finding is important, as decisions must be made in the near future by driving-algorithm designers, public-sector regulators, and ultimately the judiciary regarding the guidelines for acceptable automated drivingbehavior instructions.