Reliability Engineering and System Safety 185 (2019) 341–347
Contents lists available at ScienceDirect
Reliability Engineering and System Safety journal homepage: www.elsevier.com/locate/ress
The effect of population age on the acceptable safety of self-driving vehicles ⁎
T
Liu Peng , Zhang Yawen, He Zhen College of Management and Economics, Tianjin University, No. 92 Weijin Road, Nankai District, Tianjin 300072, PR China
A R T I C LE I N FO
A B S T R A C T
Keywords: Self-driving vehicles Acceptable safety Older population Expressed-preference approach Public acceptance
Keeping hands off the steering wheel poses a great safety challenge for self-driving vehicles (SDVs). Determining how safe SDVs should be before being allowed on public roads is a pressing question for all stakeholders, including the public exposed to the risk of SDVs in the future. The elderly may be major beneficiaries in the future age of SDVs. This research aims to understand age differences regarding the acceptable safety of SDVs and in affective, cognitive, and behavioral responses to SDVs. This study showed that the older participants implicitly required SDVs to be safer; their level for acceptable safety of SDVs was about twice as high as that for the younger participants in terms of fatality risk. Other age differences were identified, including that the older participants held a less positive attitude toward and acceptance of SDVs than the younger participants. We discuss the implications of the results for theory and practice.
1. Introduction Automated vehicle (AV) technology has the potential to significantly reduce traffic collisions, traffic congestions, and air pollution, and to increase fuel efficiency, space utilization, human mobility, and productivity; simultaneously, AV also poses risks and challenges related to safety, security, legal liability, and regulation [1–4]. According to the Society of Automotive Engineers (SAE)’s definition for vehicle automation [5], vehicles with conditional automation (Level 3), high automation (Level 4), and full automation (Level 5) can function in “selfdriving” (i.e., “automated driving” or “driverless”) mode. SAE Level 5 AVs are also referred to as self-driving vehicles (SDVs) that do not require any human intervention. Keeping hands off the steering wheel for majority of the time or all of the time poses a great safety challenge for AVs. Safety is the top priority for AVs, and therefore we should pay close attention to their safety performance and the socially acceptable safety level that they need to reach before they are allowed on public roads. 1.1. Current safety performance of AVs Safety assessments were conducted to compare the safety of AVs, which are undergoing testing on public roads, with that of human drivers; the reported results have been mixed. Two studies [6,7] compared the crash rate of Google's AVs and the national crash rate in the USA, concluding that Google's AVs would be safer than conventional vehicles. However, other safety assessments [8–10] reached the
⁎
opposite conclusion when they included data from other companies. Schoettle and Sivak [8] observed that AVs from three companies (i.e., Google, Delphi, and Audi) tested in California might have a higher accident rate than conventional vehicles, but the overall severity of crash-related injuries involving AVs was noted to be lower than for conventional vehicles. Similarly, relying on AVs’ crash reports in California, Favarò et al. [9] concluded that “conventional vehicles drive one order of magnitude more miles compared to AVs before encountering an accident, with a mean mileage before a crash for conventional vehicles of about 500,000 miles, compared to 42,017 miles for AVs” (p. 18) and Banerjee et al. [10] concluded that “autonomous vehicles are 15‒4000× worse than human drivers for accidents per cumulative mile driven” (p. 586). Note that several factors, such as insufficient miles traveled by AVs, AVs driven in limited conditions, and disengagements of AVs, can affect the conclusions derived in these comparative studies. Nevertheless, it is clear that the current so-called self-driving cars (actually, Level 3 AVs), which are being tested on public roads, are unsafe and not reliable enough to drive themselves unsupervised. Demonstrating the safety and reliability of AVs is a challenge for policymakers and the transportation industry, which could take ages given the size of the existing AV fleets and the low probability of traffic crashes [11]. 1.2. Acceptable safety of AVs A more pressing question for stakeholders and the public would be: How safe is safe enough for AVs? [12,13]. Acceptable safety (or its opposite, acceptable risk) level would be the most widely sought factor in
Corresponding author. E-mail address:
[email protected] (P. Liu).
https://doi.org/10.1016/j.ress.2019.01.003 Received 2 July 2018; Received in revised form 18 December 2018; Accepted 2 January 2019 Available online 03 January 2019 0951-8320/ © 2019 Elsevier Ltd. All rights reserved.
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
the management and regulation of technologies [14]. There are no universal numbers expressing what levels of safety/risk should be regarded as acceptable. So far, no law or regulation policy has specified the required safety of SDVs and other levels of AVs [15], which is an obstacle to the proliferation of this technology. Certain lawmakers and regulators have been reported to be considering allowing AVs to be deployed on roads as they are deemed equally safe in comparison with human drivers [13] or twice as safe as human drivers [16]. A policy study [12] claimed that less stringent safety policies (e.g., allowing AVs to be slightly safer than the average human driver) should be considered to save more human lives. However, the public and consumers might hesitate to accept such a utilitarian standpoint. A recent study showed that SDVs are implicitly required to be 4‒5 times as safe as conventional human-driven vehicles [17]. For any emerging technology, arriving at a consensus decision over its acceptable risk is challenging and is dependent of the public, government, and industry. The acceptable risk decision should serve the interests of different parties. Hence, for example, the industry alone cannot formulate risk acceptance criteria [18]. Public preferences and values should be considered while solving the acceptable risk problem [14,19,20]. For AVs, a type of future mass consumer product, public input and preferences are particularly important. If AVs cannot reach the expected safety level or reduce the traffic risks to the expected level, the public and consumers will insist on using conventional vehicles or manually drive the vehicles themselves. Acceptable risk should not be based on the judgments of risk probability alone [21,22]. It has a conditional, context-dependent nature [19]. For instance, as we assess acceptable risk from the perspective of the public, their values, interests, and even demographic characteristics would affect their decisions on acceptable risk.
Becker and Axhausen [40] for a review). Higher age has been associated with more negative perceptions of AVs. For example, older participants perceived AVs as less useful and more difficult to use, compared to younger participants [24]. They hold a more negative attitude toward AVs [41] and expressed more concerns about AVs [42]. Accordingly, they were less interested in using AVs [24,42], and less willing to use one when AVs are eventually available [24] and less willing to pay for this technology [1,43]. Therefore, one may conclude that older people are less open to the introduction of AVs [40]. However, certain studies (see two cross-national surveys [44,45]) also observed non-significant effects of age on perceptions, attitudes, and acceptance of AVs. Even, certain studies [46,47] reported a positive correlation between age and AV acceptance. Rödel et al. [46] reported a stronger intention to use AVs with increasing age. They explained that older participants prefer a higher level of automation because it facilitates driving while providing comfort and safety simultaneously. Nordhoff et al. [47] invited participants to experience an automated shuttle and found a positive correlation between age and intention to use automated shuttles. In sum, previous studies reported mixed results about the age effect on perceptions, attitudes, and acceptance of SDVs and other AVs. However, none of these studies centered on the age effect, and thus they are hard-pressed to provide a complete understanding of the age effect. Our second objective is to understand age differences in affective responses (trust, negative affect), cognitive responses (perceived benefit and risk), and attitudinal and behavioral responses (attitude and acceptance) to SDVs. These responses can be linked to the acceptable risk of SDVs (e.g., acceptance) or be determinants of risk acceptance (e.g., trust [48] and affect [49,50]).
1.3. Research objectives
2. Methodology
With the aging of the population, the number of older drivers increases rapidly. For instance, in the USA, there were more than 40 million licensed drivers aged 65 and older in 2015, a 50% increase from 1999 [23]. As people age, their mobility and driving ability decline. Due to this, the older population are expected to benefit the most from SDVs in terms of mobility [17,24]. To enable the older population to be the major beneficiaries in the age of SDVs, we should be clear in our minds about their acceptable risk of and responses to SDVs.
We apply an expressed-preference approach [17] to determine the age difference in the acceptable risk of SDVs and use psychometric scales to capture age differences in affective, cognitive, and behavioral responses to SDVs. As the determination of acceptable risk is the core of this study, we will first introduce the expressed-preference approach [17] and then other methodological issues.
1.3.1. Age difference in acceptable risk Age difference in acceptable risk is ignored in the risk perception literature, according to our observation. Driver safety research suggests that younger drivers usually have a greater propensity for sensation seeking, engage in more risky behavior while driving [25], and perceive less risk in hazardous conditions [26] compared to older drivers. Personality research [27,28] reveals that propensity for risk taking and sensation seeking tends to decline across the life span. Accordingly, as age increases, risk acceptance decreases [29]. According to these hints, we assume that older participants (vs. younger participants) entertain a lower level of acceptable risk of SDVs. In other words, older participants are assumed to have a higher requirement over the safety of SDVs. Our first objective is to understand the potential age difference in the acceptable risk of SDVs.
This approach [17] assumes the proportion of people accepting a risk event (e.g., injury and fatality), which is called risk acceptance rate, as a function of the frequency of this event. Fewer people will accept a risk with a higher likelihood of occurrence. If we are able to build such a quantitative function between risk acceptance rate and risk frequency, we can predict acceptable risk frequency given specific risk acceptance rates. As shown in Fig. 1, given the same risk acceptance rate, the predicted acceptable risk frequency in A is lower than in B, indicating a lower level of acceptable risk in A. We can calculate the ratio of the predicted acceptable risk frequencies between A and B, and then determine their quantitative difference in acceptable risk. In Fig. 1, A and B could be two demographic groups (e.g., older vs. younger) or two technologies (e.g., SDVs vs. conventional vehicles). Following Liu et al. [17], in Fig. 2, we described the four major steps of the expressed-preference approach to determine the age difference in acceptable risk. Older and younger participants were faced with a series of traffic risk scenarios (injury and fatality) differing in risk frequencies; they were asked whether they accepted these risk scenarios, and then their responses were collected to calculate the risk acceptance rate for each risk scenario (Step 1). We constructed the logarithmic regression model [51] between risk acceptance rate and risk frequency, and then examined whether age was a significant predictor of risk acceptance rate in the regression model (Step 2). Given several specific risk acceptance rates, the associated acceptable risk frequencies were
2.1. Expressed-preference approach for measuring acceptable risk
1.3.2. Age difference in affective, cognitive, and behavioral responses There is unprecedented interest in understanding how the public respond to AVs. Beliefs about AV technology (e.g., perceived usefulness) [24,30–35], general perceptions (e.g., perceived benefit pertaining to the technology) [36–38], knowledge about the technology [38], and affect evoked by the technology [37,39] have certain influences on AV acceptance. In addition, the influences of certain demographic factors are noted; for example, men were observed to have a higher intention to use AVs than women [39] (the reader is directed to 342
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
Table 1 Traffic risk descriptions (translated from Chinese). Traffic risk description Injury risk Per 100 persons per year, 1 person was injured in SDV crashes Per 1000 persons per year, 1 person was injured in SDV crashes Per 5000 persons per year, 1 person was injured in SDV crashes Per 10,000 persons per year, 1 person was injured in SDV crashes Per 50,000 persons per year, 1 person was injured in SDV crashes Per 100,000 persons per year, 1 person was injured in SDV crashes Per 500,000 persons per year, 1 person was injured in SDV crashes Per 1 million persons per year, 1 person was injured in SDV crashes Fatality risk Per 1000 persons per year, 1 person died in SDV crashes Per 5000 persons per year, 1 person died in SDV crashes Per 10,000 persons per year, 1 person died in SDV crashes Per 50,000 persons per year, 1 person died in SDV crashes Per 100,000 persons per year, 1 person died in SDV crashes Per 500,000 persons per year, 1 person died in SDV crashes Per 1 million persons per year, 1 person died in SDV crashes Per 10 million persons per year, 1 person died in SDV crashes
Fig. 1. Acceptable risk in A is lower than that in B. A and B can be two technologies or two demographic groups. Risk acceptance rate is operationalized as the proportion of people accepting a risk event with given specific risk severity and risk frequency.
Risk frequency
1E − 2 1E − 3 2E − 4 1E − 4 2E − 5 1E − 5 2E − 6 1E − 6
1E − 3 2E − 4 1E − 4 2E − 5 1E − 5 2E − 6 1E − 6 1E − 7
acceptance (very low = 1, very high = 5) of SDVs. The following items were rated with five levels (very low = 1, very high = 5). For perceived benefit, the participants were asked to rate the extent to which SDVs are beneficial to their family and themselves and to society; for perceived general risk, the participants were asked to rate the extent to which SDVs will be threatening to their family and themselves and to society. These four items of perceived benefit and risk were adapted from [55]. Then the participants were asked to indicate how they feel (i.e., worry and fear) if they are asked to ride in an SDV. These two items were adapted from [56]. Finally, participants were asked to indicate their level of overall trust in SDVs. As suggested by previous studies [48,49], these affective responses (worry, fear, and trust) could in part explain age difference in acceptable risk of SDVs. Part III described traffic scenarios with varying risk frequencies and severities (injury and fatality). Risk frequency can be expressed as one injury or fatality per a certain number of societal populations. Eight frequencies were designed for each level of risk severities (see Table 1), in line with Liu et al. [17]. Part III began with an assumption that the future daily transportation tools are SDVs, and then instructed participants to indicate their acceptance of these traffic risk scenarios on a four-point scale: “never accept,” “hard to accept,” “easy to accept,” and “fully accept.” For each risk scenario, a participant was considered to accept it if he/she chose “fully accept,” or “easy to accept.” Its risk acceptance rate was calculated as the percentage of participants accepting the risk [17,51]. In Part IV, participants reported their demographic information (e.g., sex and age) and also whether they had heard of SDVs prior to the survey (see Table 2).
Fig. 2. The expressed-preference approach for measuring the acceptable risk of SDVs, adapted from Liu et al. [17].
inversely predicted (Step 3). Finally, we quantified the acceptable risk entertained by older participants against that for younger participants (Step 4). Section 2.2 describes how to use the psychometric technique to collect the data of risk acceptance rate for each risk scenario as required in Step 1. Results of Steps 2‒4 are reported in Section 3.1. 2.2. Questionnaire design The questionnaire comprised four parts. Part I briefed participants about the basic statistics of road traffic injuries and deaths across the world [52] and in China [52,53] to facilitate participants’ understanding of the real traffic risk (see Appendix for the benchmark information on the road traffic risks we used), similar to Liu et al. [17]. Participants were provided a textual description about SDVs from [44]. The current study differs from Liu et al.’s study [17] in that it provides participants with benefit and risk information about SDVs [1–3,36] (see Appendix). People's acceptable risk for a technology is dependent on risk-benefit tradeoffs [19,54]. Thus, the provision of benefit and risk information may facilitate deliberated decisions by the participants about their acceptable level of traffic risks of SDVs. Part II collected the participants’ attitudinal and behavioral responses (i.e., attitude and acceptance), cognitive responses (i.e., perceived benefit and risk), and affective responses (i.e., worry, fear, and trust) to SDVs by applying five-point scales. The first two items asked for a rating of global attitude (very negative = 1, very positive = 5) and
2.3. Participants This survey was conducted between February and May 2018. A convenience sample of the residents of Tianjin, China was invited to complete the questionnaire through direct intercept while in recreational areas. In total, 304 younger participants and 300 older participants completed the survey. Among them, data of 28 younger participants and eight older participants were removed from the final analysis due to their abnormal responses (e.g., all responses were identical in Part III). The demographic information of the final participants is 343
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
Table 2 Demographic information of younger (n = 276) and older (n = 292) participants. Variables Have heard of SDVs Yes No Sex Female Male Education Middle school and below High school Junior college Undergraduate Graduate Driving activity 0 time per week 1–3 times per week 4–7 times per week >1 time per day Occupation Company employee Civil servant Public-sector employee Self-employed Retired Student Others Monthly income (CNY) 1,000–3,000 3,000–5,000 5,000–7,000 7,000–10,000 10,000–20,000 >20,000 Holder of driver license Yes No
Younger
Older
95.7% 4.3%
72.3% 27.7%
44.6% 55.4%
44.5% 55.5%
0.4% 3.3% 5.1% 77.9% 13.4%
7.2% 29.8% 25.3% 26.0% 11.6%
76.8% 13.8% 4.7% 4.7%
57.5% 11.0% 13.0% 18.5%
19.9% 0.0% 3.3% 2.2% 0.0% 73.2% 1.4%
10.3% 2.7% 19.5% 11.6% 55.1% 0.0% 0.7%
64.1% 13.8% 4.7% 9.1% 6.5% 1.8%
47.6% 16.1% 11.6% 16.4% 8.2% 0.0%
59.8% 40.2%
55.5% 44.5%
Table 3 Regression of risk acceptance rate on In(risk frequency) and age. Predictors
(Intercept) In(risk frequency) Age (younger = 0, older = 1) F(2, 13) p Adjusted R2
Injury risk β SE
p
Fatality risk β SE
p
−0.2963 −0.0914 0.0141
<0.001 <0.001 0.635
−0.6732 −0.0921 −0.0687
<0.001 <0.001 0.058
170.0 < 0.001 0.956
0.0528 0.0050 0.0290
0.0696 0.0058 0.0331
127.2 <0.001 0.944
Note: β, standardized coefficients; SE, standard error.
shows the significant logarithmic relationships (ps < 0.001) between the risk acceptance rate and frequencies of injury risk (left) and fatality risk (right). All logarithmic regression models had a R2 value greater than 0.900. As expected, the risk acceptance rate decreased with an increase in risk frequency. The interaction effect of age and risk frequency on risk acceptance rate was not significant for both risk severities (ps > 0.05). Next, we tested whether the regression models were different between the two age groups. Table 3 shows the significant, negative influence of risk frequency on the risk acceptance rate. When risk severity is injury, the effect of age on the risk acceptance rate was not significant (p = 0.635), suggesting that the acceptable level of injury risk in SDV crashes was no different between these two age groups. When risk severity includes fatality, age had a marginal, negative influence on the risk acceptance rate (β = − 0.0687, p = 0.058), suggesting that older (vs. younger) participants had a marginally lower acceptable level of fatality risk in SDVs given the same risk frequency. Finally, we determined the quantitative difference between the two age groups regarding acceptable level of fatality risk in SDVs. We inversely predicted the acceptable risk frequency given certain risk acceptance rates (from 0.3 to 0.7 with an interval of 0.1) and then computed the ratio of the predicted acceptable risk frequencies between the two groups. As shown in Table 4, the acceptable level of fatality risk in the older group was estimated to be 0.42 times that of the younger group. In other words, the acceptable safety level in the older group was 2.39 (1/0.42) times that of the younger group.
Note: CNY, Chinese Yuan (1 CNY = $0.145).
presented in Table 2. The mean age of younger participants was 23.0 (standard deviation, SD = 3.5; 18‒39 years old); the mean age of older participants was 58.8 (SD = 6.6; 50‒78). A higher percentage of participants in the younger group had heard of SDVs before the survey than that in the older group (χ2(1, N=568) = 56.70, p < 0.001).
3.2. Age difference in affective, cognitive, and behavioral responses 3. Results
Analysis of variance revealed that these two age groups did not differ in three items (perceived benefit to their family and themselves, worry, and trust) (see Fig. 4). The younger participants reported a higher positive attitude and acceptance of SDVs than the older participants. They perceived a higher benefit of SDVs to society, whereas they also perceived higher risk of SDVs to their family and themselves
3.1. Age difference in acceptable risk Following previous studies [17,51], we determined the logarithmic relationship between the risk acceptance rate and risk frequency. Fig. 3
Fig. 3. The relationship between the risk acceptance rate and frequencies of injury risk (left) and fatality risk (right). 344
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
In addition to the personality account, one may argue that the age differences in perceptions and responses of a technology can account for age differences in acceptable risk of the technology. For instance, lower trust in a technology [48] or higher negative affect evoked by the technology [49] may cause this technology to be less accepted. In this study, the two age groups did not differ in trust in SDVs and perceived worry, whereas the older group perceived higher fear than the younger group (ΔMean = 0.18, p = 0.01) when they were asked to ride in an SDV. Thus, higher perceived fear could partly explain why older participants had a lower acceptable level of fatality risk related to SDVs. Future research can move beyond cross-sectional survey design to experimental design to theorize this post hoc explanation.
Table 4 Predicted acceptable risk frequencies of younger and older participants and their ratios (Risk severity: fatality). Risk acceptance rate
30% 40% 50% 60% 70% Mean SD
Predicted risk frequency Younger
Older
2.5E − 5 8.8E − 6 3.1E − 6 1.1E − 6 3.7E − 7
1.2E − 5 4.0E − 6 1.3E − 6 4.2E − 7 1.4E − 7
Ratio of predicted risk frequencies (Older/Younger)
0.48 0.45 0.42 0.39 0.37 0.42 0.05
Note: Risk acceptance rate = the percentage of participants accepting the risk.
4.2. Age difference in responses to SDVs
and to society, than older participants. When asked to ride in an SDV, older participants felt a higher level of fear than younger participants.
Older adults are positioned to benefit from the enhanced mobility provided by SDVs [17,24]. Their attitude and acceptance of SDVs will be a decisive factor in determining whether they are willing to use SDVs and then obtain the benefit in mobility. Certain SDV studies reported the non-significant correlation between age and acceptance [44,45,59], and even the positive correlation between them [46,47]. Contrastingly, our older participants (vs. younger participants) were less positive toward SDVs and expressed lower acceptance of SDVs. This finding is consistent with most studies on age differences in acceptance of AVs, which have already reported that older people are less likely to accept AVs [24,42] and less willing to pay for AVs [1,43] than younger people. Significant age differences were reported in three out of four benefit and risk items. The older participants perceived less benefit of SDVs to society and also less risk of SDVs to themselves and their family and to society. This finding seems to be counterintuitive, at first glance. As perceived benefit usually has an inverse relationship with perceived risk [60], age differences in perceived benefit and perceived risk are supposed to be inverse. We offer a possible account for this seeming paradox. Although the younger and older participants were provided with the same information about the potential benefits and risks of SDVs before they responded to SDVs, the younger participants might have a higher likelihood of having received this information and/or other benefit and risk information from other sources (e.g., internet, newspaper) (Note: a higher percentage of younger participants had heard of SDVs before the survey), and thus they perceived both greater benefit and risk than the older participants.
4. Discussion and conclusions 4.1. Age difference in acceptable risk The older and younger participants did not have different levels of acceptable injury risk related to SDVs. However, the older participants seemed to be more averse to fatality risk. Their acceptable fatality risk was estimated to be 0.42 times than that for the younger participants. In other words, the acceptable safety level in the older group was about twice that of the younger group in terms of fatality risk. To the best of our knowledge, this is the first time such an age difference has been witnessed in acceptable risk of technologies. Risk perception research associates risk acceptability with qualitative characteristics of risks (e.g., controllability, dread, catastrophic potential) [57], but seldom associates risk acceptability with the characteristics of population exposed to risks or potentially affected by risks. Thus, our finding is a contribution to the risk perception literature. Previous studies [29,58] on risk acceptance indirectly supported our finding. Turner and McClure [29] found that older participants (vs. younger participants) had a lower degree of acceptance for risky activities such as riding a bicycle without a helmet. Huang et al. [58] reported that older participants were more reluctant to accept the risk of the chemical industry than younger participants. Age difference in acceptable risk of SDVs can be explained by the age differences in risk taking and sensation seeking. Personality research [27,28] indicates that propensity for risk taking and sensation seeking decreases with age. For instance, older drivers (vs. younger drivers) usually have lower propensity for risk taking and are less likely to engage in risky behaviors while driving [25].
4.3. Practical implications Previous research [17] indicates that the acceptable safety level of SDVs is higher than that of conventional vehicles. Because current SDVs are riskier than human drivers [9,10], it is highly necessary to improve
Fig. 4. Mean value of the nine items in the two age groups. Error bars are 95% confidence intervals. *p < 0.05, 345
⁎⁎
p < 0.01,
⁎⁎⁎
p < 0.001.
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
There is no consistent statistical data on road traffic risk in China. China's Ministry of Public Security reported that 60,000 deaths are caused by road traffic crashes annually, that is, four to five road traffic deaths per 100,000 population per year; according to statistical analysis by China's Department of Health, 160,000 die in road traffic crashes annually, that is, 13 road traffic deaths per 100,000 population per year; the WHO estimated that 260,000 die annually in road traffic crashes in China, that is, 19 road traffic deaths per 100,000 population per year. The description for the graphic scenario of SDVs used in Part I, from [17], reads as below: The picture shows one possible application scenario of self-driving. Selfdriving enables the driver (that is, passengers) to perform more non-driving activities, such as reading a book, watching a film, surfing the Internet, playing on their phones, dealing with their working affairs, sleeping, and so on. The driver and front-seat passenger are able to swivel their seats and engage in face-to-face communication and conversation. The introductory text of benefit information of SDVs, from [36], is as below: Self-driving cars might have the following benefits: reducing traffic crashes, traffic congestions, vehicle emission, and air pollution, improving fuel efficiency, reducing travel cost, and the mobility for older population and people who are currently unable to drive. The introductory text of risk information of SDVs, from [36], is as below: Self-driving cars might have the following risks: equipment or system failures; unclear legal liability of drivers or owners; computer systems being hacked; privacy disclosure of travelers; high purchase cost; drivers cannot control vehicles in crashes.
system safety of SDVs. We recognized the age difference in the acceptable safety, that is, the acceptable safety level of SDVs for the older participants is about twice that for the younger participants in terms of fatality risk. The implication of this finding for safety management is that SDVs for older people must be designed to be safer and more reliable (e.g., adding more safety-related functions) or assessed and regulated through stricter safety standards to allow older people to be the major beneficiaries, as intended, in the age of SDVs. Furthermore, our results may provide insights for public communication and marketing campaigns for SDVs. Younger people (vs. older people) are more likely to adopt SDVs and can accept more risks related to SDVs in general. Thus, younger people would be the first adopters of SDVs and the first targets in marketing campaigns. From the technical perspective, increasing the safety and reliability of SDVs will be challenging and time-consuming, since accumulating sufficient real driving experience and miles traveled is necessary to improve their safety and reliability. Parallel non-technical activities are necessary. For instance, to increase the number of adopters, policy or marketing campaigns should be tailored to reduce age differences in behavioral intentions to use AVs. Policymakers could promote end-user training tailored for older drivers to decrease their feelings of fear and increase their perceived safety in riding in SDVs. 4.4. Limitations and future research Several limitations should be noted. First, the sample was relatively small and from only one country, and thus it was not representative. Generalizing our findings to the general population should be treated with caution. Second, the relatively small sample size leads to difficulty in analyzing the effects of other demographic factors (e.g., sex, education, income) and their potential interaction effects on acceptable risk decisions. Third, the participants did not have practical experience of the self-driving mode of AVs, which is a common issue for current social science studies related to SDVs. Thus, the responses of the participants might be biased. Fourth, people's limitations in numeracy (e.g., they might be biased in the interpretation of very small numbers) could affect the relationship between the risk acceptance rate and risk frequency [17]. Fifth, the majority of the younger participants (76.8%) and more than half of the older participants (57.5%) reported that they had no regular driving activity, which could affect their perceptions of SDVs and their acceptable risk of SDVs. Certain avenues of research are suggested here. First, it is worthwhile clarifying the contribution of the trait of sensation seeking to acceptable risk. To the best of our knowledge, few studies have been conducted to explain the acceptable risk decision from a perspective of personality. Second, future studies can examine the effect of direct experience of AVs on acceptable risk; for instance, effort can be put into understanding the age difference in acceptable risk between the younger and older participants after they gain experience of the selfdriving mode in an AV. Third, replicating our survey in other populations, countries, and cultures is necessary to provide effective safetyrelated communication, management, and policy making for SDVs.
References [1] Bansal P, Kockelman KM, Singh A. Assessing public opinions of and interest in new vehicle technologies: an Austin perspective. Transp Res Part C Emerg Technol 2016;67:1–14. [2] Anderson JM, Kalra N, Stanley KD, Sorensen P, Samaras C, Oluwatola OA. Autonomous vehicle technology: a guide for policymakers. Santa Monica, CA: RAND Corporation; 2016. [3] Fagnant DJ, Kockelman K. Preparing a nation for autonomous vehicles: opportunities, barriers and policy recommendations. Transp Res Part A Policy Pract 2015;77:167–81. [4] NHTSA. Federal automated vehicles policy: accelerating the next revolution in roadway safety. Washington, D.C.: National Highway Traffic Safety Administration, U.S. Department of Transportation; 2016. [5] SAE. Taxonomy and definitions for terms related to on-road motor vehicle automated driving systems. Washington, D.C.: SAE International; 2014. [6] Blanco M, Atwood J, Russell S, Trimble T, McClafferty J, Perez M. Automated vehicle crash rate comparison using naturalistic data. Virginia: Virginia Tech Transport Institute; 2016. [7] Teoh ER, Kidd DG. Rage against the machine? Google's self-driving cars versus human drivers. J Safety Res 2017;63:57–60. [8] Schoettle B, Sivak M. A preliminary analysis of real-world crashes involving selfdriving vehicles. UMTRI-2015-34. Ann Arbor, MI: Transportation Research Institute, University of Michigan; 2015. [9] Favarò FM, Nader N, Eurich SO, Tripp M, Varadaraju N. Examining accident reports involving autonomous vehicles in California. PLoS One 2017;12(9):e0184952. [10] Banerjee SS, Jha S, Cyriac J, Kalbarczyk ZT, Iyer RK. Hands off the wheel in autonomous vehicles? A systems perspective on over a million miles of field data. 48th annual IEEE/IFIP international conference on dependable systems and networks (DSN). 2018. [11] Kalra N, Paddock SM. Driving to safety: how many miles of driving would it take to demonstrate autonomous vehicle reliability? Transp Res Part A Policy Pract 2016;94:182–93. [12] Kalra N, Groves DG. The enemy of good: estimating the cost of waiting for nearly perfect automated vehicles. Santa Monica, CA: RAND Corporation; 2017. [13] Mervis J. Not so fast. Science 2017;6369:1370–4. [14] Fischhoff B. Acceptable risk: a conceptual proposal. Risk Health Saf Environ 1994;1:1–28. [15] Taeihagh A. and Lim H.S.M., Governing autonomous vehicles: emerging responses for safety, liability, privacy, cybersecurity, and industry risks. Transport Rev 39 (1), 2019, 103–128. [16] Demers J. Self-driving cars will kill people and we need to accept that. https:// thenextweb.com/contributors/2018/06/02/self-driving-cars-will-kill-people-hereswhy-you-need-to-get-over-it/; 2018, Accessed date: 26 September 2018. [17] Liu P., Yang R., Xu Z., How safe is safe enough for self-driving vehicles? Risk Anal in press; 10.1111/risa.13116. [18] Abrahamsen EB, Aven T. Why risk acceptance criteria need to be defined by the
Acknowledgments This study was supported by the National Natural Science Foundation of China (no. 71601139, 71661147003, and U1733117) and the Seed Foundation of Tianjin University (no. 2018XRG-0026). Appendix The benchmark information of the real traffic risk in Part I in the questionnaire, from [17], reads as below: According to a statistical analysis by the World Health Organization (WHO), 1.25 million persons die in traffic crashes in the world every year, that is, 17 deaths per 100,000 per year. 346
Reliability Engineering and System Safety 185 (2019) 341–347
P. Liu et al.
authorities and not the industry? Reliab Eng Syst Saf 2012;105:47–50. [19] Fischhoff B, Lichtenstein S, Slovic P, Derby SL, Keeney R. Acceptable risk. Cambridge: Cambridge University Press; 1981. [20] Pidgeon N, Hood C, Jones D, Turner B, Gibson R. Risk perception. In: Royal Society Study Groupeditor. Risk: analysis, perception and management. London: Royal Society; 1992. p. 89–134. [21] Aven T. An emerging new risk analysis science: Foundations and implications. Risk Anal 2018;38(5):876–88. [22] Aven T, Ylönen M. A risk interpretation of sociotechnical safety perspectives. Reliab Eng Syst Saf 2018;175:13–8. [23] Centers for Disease Control and Prevention. Older adult drivers. https://www.cdc. gov/motorvehiclesafety/older_adult_drivers/index.html; 2017, Accessed date: 6 May 2018. [24] Lee C, Ward C, Raue M, D'Ambrosio L, Coughlin JF. Age differences in acceptance of self-driving cars: a survey of perceptions and attitudes. In: Zhou J, Salvendy G, editors. Human aspects of it for the aged population. Aging, design and user experience. London: Springer; 2017. p. 3–13. [25] Jonah BA. Accident risk and risk-taking behaviour among young drivers. Accid Anal Prev 1986;18(4):255–71. [26] Deery HA. Hazard and risk perception among young novice drivers. J Safety Res 1999;30(4):225–36. [27] Mata R, Josef AK, Hertwig R. Propensity for risk taking across the life span and around the globe. Psychol Sci 2016;27(2):231–43. [28] Canale N, Vieno A, Lenzi M, Griffiths MD, Perkins DD, Santinello M. Cross-national differences in risk preference and individual deprivation: a large-scale empirical study. Pers Indiv Differ 2018;126:52–60. [29] Turner C, McClure R. Age and gender differences in risk-taking behaviour as an explanation for high incidence of motor vehicle crashes as a driver in young males. Inj Control Saf Promot 2003;10(3):123–30. [30] Nordhoff S, van Arem B, Merat N, Madigan R, Ruhrort L, Knie A, et al. User acceptance of driverless shuttles running in an open and mixed traffic environment. 12th ITS European Congress. 2017. [31] Madigan R, Louw T, Wilbrink M, Schieben A, Merat N. What influences the decision to use automated public transport? Using UTAUT to understand public acceptance of automated road transport systems. Transp Res Part F Traffic Psychol Behav 2017;50(Suppl C):55–64. [32] Moták L, Neuville E, Chambres P, Marmoiton F, Monéger F, Coutarel F, et al. Antecedent variables of intentions to use an autonomous shuttle: moving beyond TAM and TPB? Eur Rev Appl Psychol 2017;67(5):269–78. [33] Choi JK, Ji YG. Investigating the importance of trust on adopting an autonomous vehicle. Int J Hum Comput Interact 2015;31(10):692–702. [34] Buckley L, Kaye S-A, Pradhan AK. Psychosocial factors associated with intended use of automated vehicles: a simulated driving study. Accid Anal Prev 2018;115:202–8. [35] Xu Z, Zhang K, Min H, Wang Z, Zhao X, Liu P. What drives people to accept automated vehicles? Findings from a field experiment. Transp Res Part C Emerg Technol 2018;95:320–34. [36] Liu P., Yang R., Xu Z., Public acceptance of fully automated driving: effects of social trust and risk/benefit perceptions. Risk Anal in press; 10.1111/risa.13143. [37] Hohenberger C, Spörrle M, Welpe IM. Not fearless, but self-enhanced: the effects of anxiety on the willingness to use autonomous cars depend on individual levels of self-enhancement. Technol Forecast Soc Chang 2017;116:40–52. [38] Ward C, Raue M, Lee C, Ambrosio LD, Coughlin JF. Acceptance of automated driving across generations: the role of risk and benefit perception, knowledge, and trust. 19th international conference on human-computer interaction. 2017. [39] Hohenberger C, Spörrle M, Welpe IM. How and why do men and women differ in
[40] [41] [42]
[43]
[44]
[45]
[46]
[47]
[48] [49] [50] [51]
[52] [53]
[54]
[55] [56] [57] [58]
[59]
[60]
347
their willingness to use automated cars? The influence of emotions across different age groups. Transp Res Part A Policy Pract 2016;94:374–85. Becker F, Axhausen KW. Literature review on surveys investigating the acceptance of automated vehicles. Transportation 2017;44(6):1293–306. Hulse LM, Xie H, Galea ER. Perceptions of autonomous vehicles: relationships with road users, risk, gender and age. Saf Sci 2018;102:1–13. Schoettle B, Sivak M. A survey of public opinion about autonomous and self-driving vehicles in the U.S., the U.K., and Australia. UMTRI-2014-21. Ann Arbor, MI: Transportation Research Institute, University of Michigan; 2014. Abraham H, Lee C, Brady S, Fitzgerald C, Mehler B, Reimer B, et al. Autonomous vehicles and alternatives to driving: trust, preferences, and effects of age. Proceedings of the transportation research board 96th annual meeting. 2017. Kyriakidis M, Happee R, de Winter JCF. Public opinion on automated driving: results of an international questionnaire among 5000 respondents. Transp Res Part F Traffic Psychol Behav 2015;32:127–40. Nordhoff S, de Winter J, Kyriakidis M, van Arem B, Happee1 R. Acceptance of driverless vehicles: results from a large cross-national questionnaire study. J Adv Transport 2018;2018:5382192. Rödel C, Stadler S, Meschtscherjakov A, Tscheligi M. Towards autonomous cars: the effect of autonomy levels on acceptance and user experience. Proceedings of the 6th international conference on automotive user interfaces and interactive vehicular applications. 2014. Nordhoff S, de Winte J, Madigan R, Merat N, van Arem B, Happee R. User acceptance of automated shuttles in Berlin-Schöneberg: a questionnaire study. Transp Res Part F Traffic Psychol Behav 2018;58:843–54. Siegrist M, Sütterlin B. Importance of perceived naturalness for acceptance of food additives and cultured meat. Appetite 2017;113:320–6. Siegrist M, Sütterlin B. Human and nature-caused hazards: the affect heuristic causes biased decisions. Risk Anal 2014;34(8):1482–94. Rhodes N, Pivik K. Age and gender differences in risky driving: the roles of positive affect and risk perception. Accid Anal Prev 2011;43(3):923–31. Huang L, Zhou Y, Han Y, Hammitt JK, Bi J, Liu Y. Effect of the Fukushima nuclear accident on the risk perception of residents near a nuclear power plant in China. Proc Natl Acad Sci USA 2013;110(49):19742–7. WHO. Global status report on road safety 2015. Geneva, Switzerland: World Health Organization; 2015. Huang Y, Tian D, Gao L, Li L, Deng X, Mamady K, et al. Neglected increases in rural road traffic mortality in China: findings based on health data from 2005 to 2010. BMC Public Health 2013;13(1):1111. Fox-Glassman KT, Weber EU. What makes risk acceptable? Revisiting the 1978 psychological dimensions of perceptions of technological risks. J Math Psychol 2016;75:157–69. Clothier RA, Greer DA, Greer DG, Mehta AM. Risk perception and the public acceptance of drones. Risk Anal 2015;35(6):1167–83. Terpstra T. Emotions, trust, and perceived risk: affective and cognitive routes to flood preparedness behavior. Risk Anal 2011;31(10):1658–75. Slovic P. Perception of risk. Science 1987;236(4799):280–5. Huang L, Ban J, Sun K, Han Y, Yuan Z, Bi J. The influence of public perception on risk acceptance of the chemical industry and the assistance for risk communication. Saf Sci 2013;51(1):232–40. Payre W, Cestac J, Delhomme P. Intention to use a fully automated car: attitudes and a priori acceptability. Transp Res Part F Traffic Psychol Behav 2014;27(Part B):252–63. Alhakami AS, Slovic P. A psychological study of the inverse relationship between perceived risk and perceived benefit. Risk Anal 1994;14(6):1085–96.