The effect of safety news and facts on the willingness to adopt the
autonomous car
Eelco de Gans 11420871 MSc. in Business Administration – Entrepreneurship & Innovation Universiteit van Amsterdam Dhr. Dr. G.T. Vinig Final Thesis June 20, 20182
Statement of originality
This document was written by Eelco de Gans, who takes full responsibility for the contents thereof.I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in its creation.
The Faculty of Economics and Business is responsible solely for the supervision of the completion of this work, not its contents.
3
Abstract
Although the autonomous car may sound like a concept for the future, in reality such cars already exist. Various driver-assistance systems have been supplemented and extended to create a system that can drive completely independently in certain situations. Autonomous cars have a number of advantages; they can improve safety and increase time and fuel efficiency. However, despite these advantages self-driving cars are not fully accepted by the general public. Consumer resistance to the autonomous car is largely based on safety issues.The objective of this study was to examine the influence of perceived safety on the willingness to adopt autonomous cars. This research studied the effect of negative safety news and safety facts on both perceived safety and the willingness to adopt autonomous vehicles. This study found that higher levels of perceived safety lead to a higher willingness to adopt. In addition, this study demonstrated that reading safety facts about autonomous cars ensures that consumers have a higher level of perceived safety and thereby a higher willingness to adopt self-driving cars. This finding was consistent with arguments in the literature on autonomous cars, which contends that the more knowledge consumers have about innovations the more easily they accept them. The outcomes are discussed along with the theoretical and managerial implication. Finally, it is presented possible future research directions in the field of willingness to adopt and diffusion of the autonomous car and innovation in general.
Keywords: Innovation, diffusion, resistance, autonomous car, willingness to adopt, safety
4
Table of contents
1. Introduction 6 2. Literature review 9 2.1 Innovation diffusion 9 2.1.1 Introduction 9 2.1.2 Diffusion 9 2.1.3 Innovation 10 2.1.4 Innovation decision process 11 2.1.5 Adopter categories 13 2.1.6 Technology acceptance model 15 2.1.7 Summary 16 2.2 Resistance to innovation 16 2.2.1 Introduction 16 2.2.2 Resistance to change and innovation 17 2.2.3 Resistance to technology 18 2.2.4 Loss-aversion 19 2.3 Autonomous cars 19 2.3.1 Innovation in the automotive industry 19 2.3.2 Autonomous cars 20 2.3.3 Benefits of the autonomous car 22 2.4 Consumer willingness to adopt the autonomous car 25 2.5 Influence of information 27 2.5.1 Introduction 27 2.5.2 Negative news 28 2.5.3 Facts 29 2.6 Conceptual model 315 3. Method 32 3.1 General 32 3.2 Pre-test 34 3.3 Variables 34 3.3.1 Independent variables 34 3.3.2 Perceived safety 35 3.3.3 Willingness to adopt 36 3.3.4 Control variable 37 3.4 Sample 37 3.5 Data analyses 38 4. Results 40 4.1 Statistical analysis 40 4.1.1 Missing Data 40 4.1.2 Recoding 40 4.1.3 Validity 40 4.1.4 Reliability 42 4.1.5 Computing scale means 43 4.2 Correlation matrix 44 4.3 Hypotheses testing 47 4.4 Additional results 49 4.4.1 Familiarity 49 4.4.2 Demographics 50 5. Discussion 53 5.1 Theoretical implications 53 5.2 Managerial implications 55 5.3 Limitations and recommendations for future research 58 6. Conclusion 60 Bibliography 62 Appendices 71 Appendix 1 Questionnaire 71 Appendix 2 News and Facts item of questionnaire 75 Appendix 3 Codebook 77
6
1. Introduction
Although the introduction of the autonomous car, or self-driving car, may sound like a futuristic concept, in fact the autonomous car is already a reality. Without any realization of where the innovation might lead, the journey to the self-driving car began in the 1990s with the invention of cruise control. This initial innovation in assisted driving has slowly been supplemented and extended into a system that can drive completely independently in certain situations.Schumpeter describes innovation as a destructive process that challenges current best practices by overruling existing products or processes with new ones (Schumpeter, 1939). According to König and Neumayr (2017), the autonomous car therefore fits well with the concept of innovation. In addition, the autonomous car has the potential to disrupt current modes of transportation because of its level of radicalness (Hauschildt, 2004). Because of this, it is plausible that the autonomous car will generate resistance (Douma & Palodichuk, 2012). Autonomous cars have a number of advantages – they can improve safety and increase time and fuel efficiency (Beiker, 2012; Douma & Palodichuk, 2012; Silberg et al., 2014). Despite these advantages, research shows that autonomous cars are not fully accepted by the general public.
In the 1990s, Bekiaris et al. (1997) conducted one of the first studies on resistance towards the autonomous car. They concluded that while drivers appreciate driver-assistance systems that have warning functions, research showed a clear rejection of “automatic driving” (Bekiaris et al., 1997). Although more recent studies evince a greater positive attitude towards autonomous cars, the more sophisticated the systems of autonomous driving become, the more skeptically people react to these systems (Begg, 2014; König &
7
Neumayr, 2017). This skepticism is currently making autonomous cars a hot topic. Events about cars equipped with forms of autonomous driving make the news, especially when these events concern safety issues such as crashes or hacking (NOS-a, 2018; Lemkes, 2018).
Despite negative public perceptions substantial investments are being made in autonomous driving technology; car manufacturers and governments see the autonomous car as the solution for safety problems in current traffic. This opinion is based on safety facts that make the autonomous car attractive. Vehicle crashes are the leading cause of death for Americans aged 4-34 and 90% of traffic accidents occur due to human error (Olarte, 2011; Silberg & Wallace, 2012; Silberg et al., 2014). The objective of this study is to examine the effect of negative safety news and safety facts on both the willingness to adopt the autonomous car and the perceived safety of this type of vehicle. According to the literature, perceived safety is determined, among other things, by information the consumer receives about a product. Providing insight into the perceived safety and willingness to adopt the autonomous car is an important feature of this research.
First steps have been made towards understanding the acceptance of autonomous cars and possible resistance to them. Laukkanen et al. (2007) state that “innovations mean change to consumers, and resistance to change is a normal consumer response that has to be overcome before adoption may begin” (Laukkanen et al. 2007, p.420). Understanding possible resistance towards an innovation like the autonomous car is needed before the innovation is introduced. The literature on autonomous cars states that safety is the biggest driver for resistance towards the autonomous car. Therefore, understanding perceived safety and its effect on the willingness to adopt the autonomous car is needed. This makes it possible to reduce resistance before companies are ready for consumers to adopt
8
autonomous cars, which can help make the introduction of autonomous cars more successful. Based on this, the research question explored in this thesis is: RQ: What is the effect of perceived safety on the willingness to adopt the autonomous car, and what is the effect of negative safety news and safety facts on both the willingness to adopt the autonomous car and its perceived safety?
Four different online surveys were designed and respondents were randomly assigned to one of these four, making this a true experiment. The surveys had a 2 (News/No News) x 2 (Facts/No Facts) between subjects factorial research design. The online questionnaire was conducted among Dutch consumers.
The structure of the rest of this thesis is as follows. In Chapter 2 the literature on innovation diffusion and resistance to innovation is discussed. In addition, the literature on autonomous cars is reviewed with special attention paid to consumer reasons for resistance. The influence of different types of information on the decision to adopt or reject autonomous vehicles is also addressed in this section. In the third chapter the method used in this study is outlined and explained. In Chapter 4 the results of the study are presented and the hypotheses are tested. The results are subsequently discussed in the fifth chapter, which describes the implications of this study, its limitations, and suggestions for further research. Finally, the conclusion is presented in the sixth chapter.
9
2. Literature review
This chapter provides a review of the literature related to the topic of this study. The aspects of innovation diffusion, resistance to innovations and the autonomous car, the willingness to adopt the autonomous car, and the influence that information could have on the adoption decision are discussed.
2.1 Innovation diffusion
2.1.1 Introduction
Innovations change the way we live our lives, and in modern times these changes happen more quickly than ever before. Simply innovating for the sake of innovation is not rational – innovations must actually add something and present an advantage over existing alternatives. These benefits must be recognized by the innovation’s target group to ensure that innovations are adopted. Companies highly value the ability to predict in advance what the adoption and diffusion of a new product or service will be. For this reason, different theories and methods have been designed to predict innovation adoption and diffusion. In this section, Rogers’ diffusion theory and the technology acceptance model are discussed.
2.1.2 Diffusion
In 1962, Rogers published his book “Diffusion in Innovation.” On the basis of more than 500 studies Rogers designed a theory for the adoption of innovation called the innovation diffusion theory. The concepts described by Rogers have since served as the leading literature for understanding of the adoption process of innovation and his theory can be applied in different fields.
10
According to Rogers (2003), diffusion is the process by which an innovation is communicated through certain channels among the members of a social system over a period of time. Communication is a process in which participants create and share information with each other to reach a mutual understanding. Diffusion is a special form of communication because of the subject of the communication is a new idea (Rogers, 2003).
2.1.3 Innovation
Different definitions for the term “innovation” are used. Early on, Schumpeter (1939) described innovation as a destructive process that challenges current best practices by overruling existing products or processes with new ones (Schumpeter, 1939). For Rogers (2003), however, innovation is a broader concept. Rogers (2003) describes innovation as “an idea, practice, or project that is perceived as new by an individual or other unit of adoption” (Rogers, 2003, p. 12). For Rogers, the focus is not on the "objective" measurements of the novelty of an innovation, but rather on the perceived novelty by a certain individual. Rogers describes the five attributes of innovations as: • Relative advantage: “the degree to which an innovation is perceived as being better than the idea it supersedes” (p.15); • Compatibility: “the degree to which an innovation is perceived as consistent with the existing values, past experiences and needs of potential adopters” (p.15); • Complexity: “the degree to which an innovation is perceived as relatively difficult to understand and use” (p. 15);
• Trialability: “the degree to which an innovation may be experimented with on a limited basis” (p. 16);
11
• Observability: “the degree to which the results of an innovation are visible to others” (p. 16).
The perceived characteristics of an innovation by the members of a social system determine the rate of adoption. According to Rogers (2003), innovations that are perceived by the social system as having higher relative advantage, compatibility, trialability, and observability while being perceived as less complex will be more easily adopted by that social system. The relative speed with which members of a social system adopt an innovation is called the rate of adoption. Of these attributions, relative advantage is the strongest predictor of the rate of adoption of an innovation. Not every innovation is adopted by users and the speed with which an innovation is adopted varies (Rogers, 2003). Rogers (2003) states that speed is influenced by the type of innovation decision, the communication channels diffusing the innovation at various stages in the innovation decision process, the nature of the social system, and the extent of change agents’ efforts in diffusing the innovation (Rogers, 2003). 2.1.4 Innovation decision process The innovation decision process made by individuals includes different stages, from the first knowledge of an innovation to the final confirmation of the individual’s decision regarding it (Rogers, 2003). This process consists of five steps: (1) knowledge, when the individual is exposed to the innovation's existence and gains an understanding of how it functions; (2) persuasion, in which the individual forms a favorable or unfavorable attitude towards the innovation; (3) decision, where the individual engages in activities that lead to a choice to adopt or reject the innovation; (4) implementation, when the individual puts an innovation into use; and finally (5) confirmation, when the individual seeks reinforcement for an innovation decision already made but may reverse his or her decision if exposed to
12
conflicting messages about the innovation (Rogers, 2003). This study focuses on the influences on the willingness to adopt the autonomous car, meaning that step 3 of the diffusion process is the endpoint of this study. The innovation decision process is exhibited in Figure 1. Figure 1 Innovation decision process According to Rogers (2003), the knowledge stage of the innovation decision process is crucial for all of the subsequent stages. During this first stage the individual receives initial knowledge about the innovation, e.g. what kind of innovation it is and how it should work. The knowledge received can be divided into three groups: (1) awareness-knowledge, which reflects that the individual is aware of the innovation; (2) how-to-knowledge, which describes how the innovation could be used; and (3) principles-knowledge, which refers to how and why the innovation could work.
According to Rogers (2003), every stage of the innovation decision process comes with uncertainty that is reduced by the collection of additional information. A technological innovation needs to present some degree of advantage in order for the potential user to
13
adopt the innovation (Rogers, 2003). These advantages are not always clearly visible for potential users. In such cases, the superiority of the innovation with respect to current practices will be questioned. According to Rogers (2003), this creates uncertainty around the expected usability of the innovation and the consequences that choosing to adopt the innovation will bring. The potential user will only make an effort to understand the innovation better if it is expected to provide a solution to a perceived problem that will outweigh the consequences of its adoption. Gathering information reduces uncertainty about the possible impact of the innovation, allowing the potential user to make an accurate decision about adopting the innovation rather than a quick rejection based on uncertainty (Rogers, 2003). 2.1.5 Adopter categories Individuals come into contact with an innovation at different stages of the innovation diffusion process and decide to adopt innovations at different times (Mahajan et al., 1990). Development of adopter categories is important because it can support targeting prospective adopters for a new product at different points in time (Kotler & Zaltman, 1986), developing marketing strategies for penetrating various adopter categories (Engel, Blackwell, & Miniard, 1986), and predicting the continued acceptance of a new product (Bass, 1969; Mahajan et al., 1990). The most widely accepted method of adopter categorization is the method proposed by Rogers (2003) (Mahajan et al., 1990). This method assumes that the distribution of adopters takes a bell-shaped curve (Rogers, 2003). According to Rogers (2003), there are several characteristics that separate those who learn about an innovation early from those who learn about it later. Those who learn about an innovation early on are characterized by having more years of formal education, higher social status, greater exposure to mass media channels of communication, greater exposure
14
to interpersonal channels of communication, more contact with change agents, higher social participation, and a higher level of cosmopolitanism (Rogers, 2003).
In the beginning, only a few individuals from the social group adopt the innovation. These individuals communicate information about the innovation to their network, which lowers their network’s uncertainty about the product or service. These adopters are called “innovators” and they form the beginning of the diffusion process. If the diffusion process continues, innovators are followed by early adopters, the early majority, the late majority, and finally the laggards (Mahajan et al., 1990; Rogers, 2003). The adoption process’ bell-shaped curve and adopter categories are illustrated in Figure 2.
Figure 2 Technology adoption lifecycle (HomeAway Software, 2017)
15
2.1.6 Technology acceptance model
Many other theories and models have been developed to predict and explain the acceptance and adoption of innovations by users as counterparts to the innovation diffusion theory. One example of such a model is the well-established technology acceptance model (TAM) created by Davis (1989).
As stated by Venkatesh and Davis (2000), numerous empirical studies have found that the TAM consistently explains a substantial proportion of the variance (typically about 40%) in usage intentions and behavior. In addition, the TAM compares positively with alternative models such as the theory of reasoned action and the theory of planned behavior (Venkatesh & Davis, 2000).
According to the TAM, an individual's behavioral intention to use a particular system is determined by two attributes. As with the innovation diffusion theory, perceived usefulness is the first attribute, defined in the TAM as the extent to which a person believes that the use of the system improves performance. The second attribute is the perceived ease of use, defined as the extent to which a person believes that the use of the system requires no effort. The TAM theorizes that the effects of external variables (e.g. system characteristics, development process, training) on a person’s intention to use an innovation are mediated by perceived usefulness and perceived ease of use (Davis, 1989; Venkatesh & Davis, 2000).
16 2.1.7 Summary On the basis of the two theories discussed in this section, it can be concluded that an individual’s intention to use or adopt an innovation depends on multiple factors. The main factors are the perceived usefulness and the perceived ease of use of the innovation. Rogers also includes the compatibility of the innovation with existing values, needs, and experiences as affecting an individual’s intention to adopt or use an innovation. Whether this also includes perceived safety is not specifically mentioned.
2.2 Resistance to innovation
2.2.1 Introduction
Almost 90% of products that are launched do not survive in the marketplace (Crawford & Di Benedetto, 2008). According to Cornescu and Adams (2013), consumer resistance plays a significant role in this, as resistance can inhibit or delay consumer adoption of an innovation. Because of this, the authors see consumer resistance as one of the most prominent reasons that an innovation fails in the marketplace. With regard to this phenomenon, consumer resistance slowing the adoption process is the best-case scenario (Cornescu & Adam, 2013).
According to Claudy et al. (2015), consumer behavior frameworks in diffusion of innovation studies have largely failed to distinctly account for reasons against adoption – despite the fact that it is important to understand consumer resistance towards innovations as an important factor in explaining and predicting adoption-related behavior (Claudy et al., 2015; Heidenreich & Handrich, 2015). For this reason this section discusses the resistance towards innovation.
17
2.2.2 Resistance to change and innovation
It is typical of human beings to strive for consistency instead of constantly searching for new behavior (Sheth, 1981). According to Claudy et al. (2015), resistance to innovations can be seen as a more specific form of the general human resistance to change. This can explain why the determinants of resistance to change in general, as described by Oreg (2003), show similarities with the determinants of resistance to innovation as described by the literature on innovation resistance. The factors described by Oreg (2003) that determine the resistance to change stem from routine seeking, emotional reaction to the change, cognitive rigidity, and short-term focus.
Sheth (1981) theorized resistance to both change and innovation. The resistance towards an innovation is determined by two factors: (1) habit with regard to an existing practice and (2) perceived risks associated with the adoption of the innovation (Sheth, 1981). Ram and Sheth (1989) elaborate on Sheth’s (1981) study in an influential article on consumer resistance towards innovations. The authors argue that this resistance comes from barriers that can be understood as taking two types, functional barriers (i.e. usage, value and risk bariers) and psychological barriers (i.e. tradition and norm-based barriers, as well as image barriers) (Ram & Sheth, 1989; Claudy et al., 2015). The functional (usage) barrier described is similar to the first determinant of resistance to change, “routine seeking.” The concept of routine seeking is a recurring factor in the literature with regard to resistance towards innovation.
Although resistance to change and innovation is a natural phenomenon, it is individually dependent. Consumers have different attitudes when judging innovations and an individual’s context determines the reasons for or against adoption (Caudy et al., 2015). It can therefore be concluded that when it comes to resistance to an innovation, as with the
18 diffusion of an innovation described by Rogers (2003), much of this resistance is a matter of personal perception and the perceived advantages and disadvantages of the innovation. 2.2.3 Resistance to technology As stated in the introduction, a study by König and Neumayr (2017) found a lack of trust in the technology of the autonomous car. Respondents expressed a fear of errors made by autonomous car technology that could lead to unsafe situations (König & Neumayr, 2017). According to Fishbein and Azjen (1975) and Dietvorst et al. (2014), this is a common phenomenon. In addition, their studies showed that humans have a more negative perception about mistakes when they are made by computers than when the same mistakes are made by human beings (Fishbein & Ajzen, 1975; Dietvorst et al., 2014). People are more likely to forgive a human for making an error than technology, even when on average, an algorithm-based computer makes fewer mistakes (Dietvorst et al., 2014).
Waytz et al. (2014) further discussed this phenomenon. They investigated a theoretical determinant of people's willingness to trust technology to perform
competently – the extent to which a nonhuman agent is anthropomorphized as having a human-like mind – in the domain of autonomous driving. Participants using a driving simulator drove either a normal car, an autonomous vehicle able to control steering and speed, or a comparable autonomous vehicle augmented with additional anthropomorphic features, including name, gender, and voice. Behavioral, physiological, and self-reported measures revealed that participants trusted the vehicle to perform more competently when it acquired more anthropomorphic features (Waytz et al., 2014).
19
2.2.4 Loss-aversion
In addition to the resistance to change and innovation, the distribution of focus within the human brain also plays an important role in decision-making. As part of the decision to adopt or not adopt an innovation consumers consider both the advantages and the disadvantages of the innovation. Kleijnen et al. (2009) conclude that when confronted with a new innovation, consumers focus on disadvantages. Unfavorable aspects have a stronger impact on consumers than positive ones, and attention will therefore be drawn to negative aspects. The studies of Claudy et al. (2015) confirm this, finding that perceived losses have a disproportionately greater influence on people’s decisions than potential gains. This phenomenon is called “loss-aversion” (Tversky & Kahneman, 1974).
2.3 Autonomous cars
2.3.1 Innovation in the automotive industry
Since the advent of the car as a method of transportation for the ordinary citizen, there have been several innovations to automobiles that are now considered normal. Many of these innovations have made cars safer than before, such as the three-point safety belt that is now mandatory (Volvo, 2017). Like innovations in general, innovations in the automotive industry have begun to follow each other more and more quickly (Silberg et al., 2014).
Similar to other industries, digitization, automation, and new business models are expected to revolutionize the car industry. According to Gao et al. (2016), these forces create four disruptive trends in the automotive sector that are driven by technology, one of which is “autonomous driving” (Gao et al., 2016).
20 2.3.2 Autonomous cars As stated in the introduction, the journey towards the autonomous car started in the 1990s with the application of cruise control (Audi-a, 2017). Through the use of cruise control a certain speed can be set, after which the car maintains this speed until the driver switches off the function or presses the brake pedal. After this innovation adaptive cruise control was developed. In adaptive cruise control, the car not only maintains a set speed, but also uses radar to calculate the distance and speed of a vehicle ahead of the driver’s car, meaning that when necessary, adaptive cruise control can cause the car to brake by itself. In addition to adaptive cruise control, other systems to aid driving have been developed. Some examples include the lane change merge aid, which can recognize cars in a driver’s blind spots, or the lane keeping aid, which recognizes the lines on the road and helps the car drive between these lines (NOS-b, 2017). All of these assistance systems are already on the market and are offered for cars ranging into the mid-price segment (ANWB, 2015).
Assistance systems were the first steps towards a complete self-driving car. At the higher end of the market, cars are offered with a combination of assistance systems. These cars already have a higher level of automation and can drive completely independently in certain situations. Examples of car brands that offer these systems are Tesla, Volvo, Mercedes-Benz and Audi.
Various suppliers provide the auxiliary systems that make this form of autonomous driving possible. These systems were not comparable with one another, and the problem arose that it was not clear which systems were and were not permitted for use on public roads. In response to this, the Society of Automotive Engineers developed a universal classification scale to determine the level of autonomy of a car (SAE International, 2016).
21
Below is the United States National Highway Traffic Safety Administration (NHTSA) description of the various autonomy levels (NHTSA, 2017), which is summarized in Figure 3. Level 0: The human driver does all the driving. Level 1: An advanced driver-assistance system in the vehicle can sometimes assist the human driver with either steering or braking/accelerating, but not with both simultaneously. Level 2: An advanced driver-assistance system in the vehicle can itself actually control both steering and braking/accelerating simultaneously under some circumstances. The human driver must continue to pay full attention (“monitor the driving environment”) at all times and perform the rest of the driving tasks.
Level 3: An automated driving system (ADS) in the vehicle can itself perform all aspects of driving tasks under some circumstances. In those circumstances, the human driver must be ready to retake control at any time that the ADS requests the human driver to do so. In all other circumstances, the human driver performs the driving tasks.
Level 4: An ADS in the vehicle can itself perform all driving tasks and monitor the driving environment – essentially, do all the driving – in certain circumstances. The human driver does not need to pay attention in those circumstances.
Level 5: An ADS in the vehicle can do all the driving in all circumstances. The human occupants are simply passengers and never need to be involved in driving.
22
Figure 3 Levels of autonomy (NHTSA, 2017)
Although Tesla has become a leading pioneer in the field of both alternative fuels and autonomous driving, it is Audi that has launched the car with the highest level of autonomous driving currently available on the market. The new 2017 Audi A8 has an autopilot feature with a Level 3 SAE autonomous driving system, although it is not yet permitted for use on public roads (Audi-b, 2017). 2.3.3 Benefits of the autonomous car Autonomous cars will be connected to the external world, for example through the use of radar, cameras, and the internet, in order to be able to drive themselves. This leads to cars that, under normal operating conditions, do not crash. Improved safety can therefore be seen as the first advantage of the autonomous car. Silberg et al. (2014) state that by using multiple technologies that can substitute for each other, such as radars, cameras, and sensors, system failures will not lead to unsafe operation (Silberg et al., 2014). The connectivity of autonomous cars leads to another safety improvement – a shorter response time to changes in traffic. Autonomous cars can detect changes before a human being can recognize them, allowing their adjustments to traffic to occur more quickly (Silberg et al, 2014).
23
The pursuit of improved safety has encouraged the NHTSA to draw attention to self-driving vehicles. Their goal is not to make autonomous cars as safe as traditional cars operated by human drivers, but rather to design a car that never crashes. According to John Maddox, the NHTSA’s associate administrator for vehicle safety, the autonomous car is the best solution to achieve this goal (Silberg et al., 2014). The opinion of the NHTSA about the autonomous car is also reflected in the following statement on their website:
“The safety benefits of automated vehicles are paramount. Automated vehicles’ potential to save lives and reduce injuries is rooted in one critical and tragic fact: 94 percent of serious crashes are due to human error. Automated vehicles have the potential to remove human error from the crash equation, which will help protect drivers and passengers, as well as bicyclists and pedestrians. When you consider more than 35,092 people died in motor vehicle-related crashes in the U.S. in 2015, you begin to grasp the lifesaving benefits of driver assistance technologies” (NHTSA, 2017).
As is apparent from the NHTSA statement, the argument for the autonomous car is particularly supported by the facts and statistics relating to current traffic statistics. In addition to those who lose their lives, almost 2.5 million people are admitted to a hospital in the US each year with injuries resulting from a traffic accident. As previously stated, vehicle crashes are the leading cause of death for Americans aged 4-34 (Silberg & Wallace, 2012). Volvo is fully committed to autonomous car technology and to making their Vision2020 (that “by 2020, no one will die or be seriously injured in a new Volvo”) a reality (Volvo Cars, 2018). Based on this, it can be concluded that both car manufacturers and the NHTSA see the autonomous car as the solution to the safety concerns regarding driving (Olarte, 2011; Silberg et al., 2014).
24
According to the literature surveyed in this study, the second advantage of autonomous cars is their improvement of time and fuel efficiency. Research carried out by the Dutch independent organization for applied scientific research (TNO) on behalf of the Dutch government, shows that over 20% of a car’s fuel consumption is determined by the driving style of the driver (TNO, 2013). Regarding time-saving measures, the real-time traffic information available to an autonomous car is not only useful for actual driving, but for other purposes as well; it makes it possible to give every driver a reliable and predictable route from origin to destination. This can help the car avoid unexpected congestion as much as possible, which is the most the most time-consuming and therefore most costly form of traffic jams (Silberg et al., 2014). In addition, the chosen route will always be the most time and fuel efficient. This technology is already offered by some navigation systems, but in autonomous cars this information sharing will extend even further.
Silberg et al. (2014) describe the advantage of information sharing for infrastructure usage. For example, autonomous cars could share information about the availability of parking spots. According to a report published by MIT Media Lab, in congested urban areas 40% of total gasoline use is expended by drivers looking for parking (Mitchell). Autonomous cars can reduce the time and fuel required in this search by assessing the availability of parking more quickly (Silberg et al., 2014). Guller et al. (2014) confirm this finding and add that the usage of shared information can prevent a car from waiting unnecessarily for a red traffic light by connecting with the traffic lights (Guller et al., 2014). These types of features are not directly available through autonomous cars themselves but are rather made possible by the connectivity of autonomous cars and can therefore be implemented simultaneously with autonomous vehicles (Silberg et al., 2014).
25 2.4 Consumer willingness to adopt the autonomous car As mentioned in the introduction, various studies have been conducted to investigate the level of acceptance of the autonomous car and reasons for consumer resistance towards the innovation. In this section an overview is given of the studies conducted thus far on the perception of consumers about the autonomous car. In the 1990s, Bekiaris et al. (1997) conducted one of the first studies on consumer perceptions of the autonomous car. It turned out that drivers appreciate the appearance of driver-assistance systems with warning functions, but it showed a clear rejection of “automatic driving” as well (Bekiaris et al., 1997). Although more recent studies have found a more positive attitude towards autonomous cars, the more sophisticated the systems of autonomous driving become, the more skeptically people react to these systems (König & Neumayr, 2017).
The market research institute Multiscope conducted a market survey among 1,020 Dutch consumers in 2015. Although this study asked about the autonomous car, possible reasons for consumer resistance were not discussed. Multiscope concluded that 38% of Dutch consumers have negative feelings about autonomous cars (Multiscope, 2015). According to Sander Klous of the international accountancy and consultancy organization KPMG, an equal percentage plans to never even get into an autonomous car (KPMG, 2017). KPMG also conducted market research on the opinion of Dutch consumers about the self-driving car that concluded that the Dutch doubt the adoption of the self-driving car. Just as in the research of Multiscope, the study of KPMG only examined main points of resistance, the underlying reasons for this resistance were not investigated further (KPMG, 2017).
König and Neuhmayr (2017) did investigate the main underlying reasons for
consumer resistance to self-driving cars. Their study was conducted using a quantitative self-
26
completion online questionnaire, mostly completed by people from Austria. König and Neumayr’s study (2017) illustrates that consumers assign value to the potential benefits of autonomous cars. However, despite this, respondents showed a higher degree of concern towards the autonomous car than interest in its benefits. These results reflect a concern for legal issues, followed by a concern about technical errors and vulnerability to attacks by hackers, which could both lead to safety problems (König & Neumayr, 2017).
The results of the König and Neuhmayr (2017) study show similarities with the study of Kyriakidis et al. (2015). Kyriakidis et al. (2015) investigated user acceptance, concerns, and willingness to buy partially, highly, and fully automated vehicles. Using a 63 question internet-based survey, they collected 5,000 responses from 109 countries (40 countries with at least 25 respondents). The online survey conducted by Kyriakidis et al. (2015) found that respondents expressed concerns especially related to threats of hacking and legal issues, followed by safety and privacy concerns.
While the aforementioned studies examined public opinions in the Netherlands and in some other Western countries, Schoettle and Sivak studied public opinion about self-driving vehicles in China, India, and Japan (Schoettle & Sivak, 2014). For respondents in the Schoettle and Sivak study (2014), safety was considered the main advantage of the autonomous car, based on the possibility of reducing the number of accidents. Similar to the studies conducted in Western countries, respondents expressed strong concerns about safety issues as well. These were related to errors in the system or equipment, as well as a concern about self-driving technology not being able to drive as competently as a human being (Schoettle & Sivak, 2014).
In sum, while consumers recognize the positive aspects autonomous cars offer, their concerns overwhelm any benefits they perceive. These concerns ensure that consumers
27
resist autonomous cars instead of embracing them. Consumers are mainly concerned about the risk of hacking and the possibility that the car could make driving mistakes, as both aspects could lead to unsafe situations. The resistance can therefore mainly be characterized as resulting from safety concerns. This finding leads to the following main hypothesis: H1: A higher level of perceived safety will have a positive impact on willingness to adopt the autonomous car. 2.5 Influence of information 2.5.1 Introduction As stated by Rogers (2003), individuals seek knowledge and information during every stage of the innovation diffusion to reduce their uncertainty and reduce risk (Rogers, 2003; Kleijnen et al. , 2009). The actual decision outcome is influenced by the type, source and the amount of information individuals receive (Kleijnen et al., 2009). While both companies and consumers previously needed more information when making choices, the ever-increasing amount of information and knowledge available today now results in information overload (Bawden & Robinson, 2009; Kleijnen et al., 2009). As stated by Kleijnen et al. (2009), an overload of information can negatively influence the choices consumers make. Provision of too much information leads to poor decision-making and dysfunctional performance due to the limited capacity of human memory (Malhotra, 1982; Lurie, 2004).
28
2.5.2 Negative news
Mizerski (1982) states that unfavorable information is more influential than favorable information, which corresponds to the previously described “loss-aversion” phenomenon. An explanation for this phenomenon is that negative information has a stronger impact because it makes a stronger impression than positive information (Weinberger et al., 1981; Oreg, 2003). This is a result of the fact that the social environment of consumers has more positive than negative cues. One example of this is social media, which consists almost entirely of positive cues. This means that a negative cue attracts more attention and therefore has a stronger influence on decision-making (Weinberger et al., 1981).
The impact of negative information on decisions is moderated by different influences. Weinberger et al. (1981) found that negative information has a stronger influence on consumers with a higher level of uncertainty about the product. In addition, the literature on decision-making agrees that reduced decision timelines increase the impact of negative information on decision-making (Mizerski, 1982).
Weinberger et al.(1980) demonstrated the impact of negative product news information on decision-making. A network news broadcast imbedded an actual negative news story about an automobile. This influenced the opinions, beliefs, and intentions of the consumers immediately and effects persisted even after two weeks had passed. A direct response from the automotive company to neutralize these effects showed little to no success. The results of the study corresponded with a decrease in sales of the particular car model (Weinberger et al., 1980). These findings are in line with those of Mizerski (1982), who states that the stronger response to negative information leads to consumers not purchasing a particular product.
29 Today, autonomous cars are a very popular topic. Incidents involving cars equipped with forms of autonomous driving make the news, especially accidents involving these cars or possibilities of hacking (NOS-c, 2015; NOS-d, 2015; Janssen, 2016). It is highly plausible that this information has reached consumers and has had an influence on perceived safety, thereby affecting consumer willingness to adopt the autonomous car. This theory forms the basis for the second and third hypotheses of this study: H2: Reading negative safety news about autonomous cars will lead to lower perceived safety of autonomous cars. H3: Reading negative safety news about autonomous cars will lead to lower willingness to adopt the autonomous car. 2.5.3 Facts Consumer resistance towards a radical automotive innovation like the autonomous car is largely based on safety concerns. This is interesting because the autonomous car is a further evolution and combination of existing assistance systems that were originally developed to improve safety by partially eliminating human error (Audi-a, 2017). In addition, car manufacturers, governments and the NHTSA conclude that self-driving cars can be a solution to, rather than a cause of safety issues. This opinion is based on safety facts about the autonomous car and current traffic accident statistics (Olarte, 2011; Silberg et al., 2014). When considering a piece of information, the information’s source and type both affect the level of influence it has on a decision-making process (Kleijnen et al., 2009). While it is plausible that proven facts have a greater influence on decision-making than an average opinion, in a world in which verified facts are dismissed as “fake news” it becomes difficult
30 for consumers to determine which facts are actually true. Whether information is perceived as true and valuable depends on the person and is therefore based on each individual’s own perception (Bovee et al., 2003). According to Mizerski (1982), source plays a more important role for positive information than negative information. When positive information about a new product is received from a supplier, for example, the consumer is less inclined to regard the positive qualities as true (Mizerski, 1982). This means that the perceived credibility and influence of facts on a decision-making process is different for every decision maker. Nevertheless, it is plausible that consumers’ perceived safety of autonomous cars and their willingness to adopt this innovation will increase once they become aware of the same facts that car manufacturers, governments and the NHTSA considered when forming their opinions. This leads to the following hypotheses:
H4: Reading safety facts about autonomous cars leads to higher perceived safety of these vehicles. H5: Reading safety facts about autonomous cars leads to a higher willingness to adopt this innovation. H6: Reading negative safety news about autonomous cars has a lower impact on perceived safety when consumers read safety facts about autonomous cars as well.
31 2.6 Conceptual model Below the conceptual model of this study is exhibited. Figure 4 Conceptual model
32
3. Method
This chapter presents the method used in this research. This includes the research design, the independent and dependent variables, how the variables were measured and the pre-test used to decide which news article and what facts were shown to the participants.
3.1 General
The aim of this study is to understand the effect of perceived safety on the willingness to adopt autonomous cars and how negative safety news and safety facts influence both perceived safety and consumers’ willingness to adopt the innovation. Hypotheses were tested with the use of experiments in order to answer the research question, making this research a quantitative explanatory study. A survey was used for the data collection of the experiments in a cross-sectional manner.
Four different online surveys were designed and respondents were randomly assigned to one of these four, making this a true experiment (Walker, 2005). The surveys had a 2 (News/No News) x 2 (Facts/No Facts) between subjects factorial research design. For participants who saw news and/or facts, the perceived safety was measured before and after the information was presented. For participants who did not see news or facts, the perceived safety was measured once. The following table presents the 2x2 research design with the conditions created for the experiment.
33 Table 1 Scenario overview Qualtrics was used for conducting this survey. Both social media, like Facebook and LinkedIn, had been used to recruit participants. This study investigates the aforementioned influences on the willingness of Dutch consumers to adopt the autonomous car. Therefore, the target group is the Dutch consumer. As this study aims to examine all types of Dutch consumers and not all Dutch consumers know the English language well, the survey was conducted in Dutch. Multiple Dutch master’s students with an excellent level of English translated the scales as optimally as possible from English to Dutch.
After participants opened the web link of the survey, the survey began with an opening message that thanked them for their cooperation in the research and introduced the subject of study, the self-driving car. The term “self-driving car” is more commonly used than “autonomous car” among consumers, which is why it was chosen for the survey. A notion of informed consent was included, which stated that the information collected in the study and the consent to participate remained confidential. This was expected to lead to a lower level of social desirability in the responses (Saunders & Lewis, 2012). In addition, an email address was provided for participants to contact if they had additional questions or comments about the survey or the study in general.
34
A pre-test has been executed to determine suitable news and fact items for this experiment. Subsequently, a pilot study was conducted to ensure that the format of the collected survey data could be directly used for data analysis (Fan & Yan, 2010).
3.2 Pre-test
The pre-test was performed to be sure of maximum variation between the two independent variables “news” and “facts”. The goal was that the news item shown in the experiment would have the maximum possible negative impact on the respondents. During the pre-test three different authentic news items were shown. From these, the item that was perceived as having the most negative impact was chosen. The facts that were provided during the experiment had to have the maximum possible credibility, since source and type of the information influences decision-making process (Bovee et al., 2003; Kleijnen et al., 2009). During the pre-test three messages were presented, written by the automotive manufacturer Volvo, the NHTSA and Rijkswaterstaat (Rijkswaterstaat is responsible for the main road network in the Netherlands). Twelve master’s students determined the degree of impact that the negativity of the news items had as well as determining the credibility of the facts. The news and facts that were perceived as having the most impact were chosen for use in the experiment (for the chosen news and facts items, please see appendix 2). 3.3 Variables This section outlines and discusses the variables used for this study. 3.3.1 Independent variables
As previously mentioned, the independent variables are negative safety news and safety facts about the autonomous car. In addition, the variable “safety facts” is a moderator
35 for hypothesis 6. The possibility of showing or not showing both the news and/or the facts gives this experiment the characteristic 2x2 between subjects design, as illustrated in Table 1. To avoid any expectations that could influence the outcomes of this study the participants did not know in advance whether they would be confronted with a news item or with facts about autonomous cars. 3.3.2 Perceived safety
According to Bartneck et al. (2008) perceived safety has rarely been measured directly. Perceived safety is mostly measured indirectly, using methods other than questions in a survey. Despite this, according to Bartneck et al. (2008) there is a large number of questions that can be asked about the topic of safety. As a result of the need for a careful and studied set of baseline questions, Bartneck et al. (2008) created a semantic differential scaled questionnaire measuring anxiety, surprise and level of calm. These scales provides “a repeatable and reliable measure for assessing user’s perceived safety” when the three questions are used together (Bartneck et al., 2008, p. 78). These questions were used in this study and exhibited in Figure 5. An additional question was also added for which a five-point Likert scale was used that ranged from 1: “I strongly disagree” to 5: “I strongly agree (a = .83) (Payre et al., 2014). Figure 5 Questionnaire measuring perceived safety (Bartneck et al., 2008)
36
These scales were used for general questions about perceived safety. But according to Converse and Presser (1986), a survey must have specific questions as well. For example, a survey about satisfaction with housing should ask not only about the opinion of the house in general, but also about different rooms in the house. Scaled together, these types of questions will reflect the true opinion of the house. Specific questions can obtain more detailed data but a survey that consists entirely of specific and detailed questions is not advisable either because it takes more time to complete and the likelihood that questions would be left unanswered rises (Converse & Presser, 1986). For this reason, three questions that deal more specifically with factors on which perceived safety is based have been added to the survey. One such question is, “The self-driving car could cause safety consequences triggered by technical error.” These questions were previously used in the research of König and Neumayr (2017). 3.3.3 Willingness to adopt To measure the adoption intention the question, “Assuming you had access to such a device in the future, what is the probability that you would use it?” was used with a five-point Likert scale (α =.93; construct reliability = .85) (Kulviwat et al., 2007). Similar to the measurement of perceived safety, additional, more specific questions were used to measure the willingness to adopt the autonomous car. These questions were derived from the König and Neumayr (2017) study as well. All questions are exhibited in appendix 1, see appendix 3 for the codebook.
37
3.3.4 Control variable
All respondents of the four surveys were asked demographic questions, which form the control variables. Data about gender (nominal variable), age (ratio variable), and educational level (ordinal variable) was gathered. In addition, familiarity with autonomous cars (ordinal variable), and accident involvement was asked (Yes/No, nominal). 3.4 Sample In the interest of the analyzability of the data, the number of respondents needed to be greater than 200 (Fan & Yan, 2010). The average response rate from individuals is 52.7% (Baruch & Holtom, 2008), therefore more than 400 people had to be reached. A total of 295 surveys were begun, of which 255 (N=255) were completed within 15 days. As described in the Section 3.1, the participants were contacted through email and social media. This study used a non-probability convenience sample because the sampling frame was unknown and “Dutch consumers” is a large population. From the total included participants, 52% were male (N=133) and 48% were female (N=122). A total of 25% (N=64) were between 18-24 years old and formed the biggest group, followed by the groups between 25-34 and 55-64 years old, which each constituted 18% of the total sample (N=46). Out of all the participants, 48.6% (the largest percentage) held a Bachelor’s degree (N=124), the second largest group (20.4%) held a Master’s degree (N=52), and the third largest group (N=48) had completed secondary vocational education. Almost all participants were familiar with the concept of the autonomous car: 76.9% had seen or heard something about autonomous cars (N= 196), 13.3% had already immersed themselves in the technology (N=34), and 9.8% said they were not familiar with autonomous cars (N=25). The characteristics of the sample are summarized in Table 2.
38 Table 2 Characteristics of the sample 3.5 Data analyses After the data was collected, the data processing was initiated. First, we evaluated for missing data and recoded some of the data items. The validity and reliability of the data could then be checked to determine whether the measures were capturing what they were intended to and to examine the consistency of the measures. The scale means were then calculated and the correlation between variables was calculated. After performing these steps, the hypotheses were tested and the data was checked for additional results.
The aim of this study is to examine the direct effects of independent variables on the dependent variable, which in this case was the willingness to adopt the autonomous car.
39 Although hypotheses 1-5 focus only on direct effects between the variables, there may also be indirect effects between the variables in this study. The design of the conceptual model implies that perceived safety could act as a mediator between news and willingness to adopt or as a mediator between facts and willingness to adopt. Investigating perceived safety as a mediator was not a direct goal of this study, but such an investigation could yield interesting information. For this reason, Hayes’ PROCESS macro was chosen for the regression analysis to test the hypotheses. PROCESS uses an ordinary least squares or logistic regression-based path analytical framework for estimating direct and indirect effects in both mediator and moderator models (Hayes, 2012). This makes it possible to determine whether there is an effect between the variables and whether this is a positive or negative relationship (Hayes, 2012; Field, 2013).
In addition to testing the hypotheses, more results can be obtained from the data collected for this study. For example, it is interesting for car manufacturers and other stakeholders to know the willingness to adopt for different groups of consumers. The participants were divided into groups based on demographic information collected during the survey. This includes the gender, age and level of education of the participant. Subsequently, the average willingness to adopt of each group can be compared. If there are only two groups, for example, a t-test could have been conducted. However, apart from the control variable "gender,” there are more than two groups for these variables. For this reason, different one-way analyses or variance between groups (one-way ANOVA) were performed (Field, 2013).
40
4. Results
In this chapter the results of the study are presented. First, the general data characteristics, including the reliability and validity of the scales and the processing of missing data, are discussed. Following this, correlations are examined and hypotheses are tested. Finally, additional results are presented.
4.1 Statistical analysis
4.1.1 Missing Data
Of the 295 surveys begun, 255 surveys were completed. After analyzing the results it can be concluded that there was no missing data. This is probably due to forcing the respondents to answer each question before allowing them to go on to the next question.
4.1.2 Recoding
Some items of perceived safety were counter-indicative and were recoded. Additionally, control variables such as familiarity with autonomous cars, accident involvement, and perceived traffic safety were recoded to facilitate determining correlation.
4.1.3 Validity
Validity is the degree to which an instrument measures the construct it is intended to measure (Cronbach & Meehl, 1955; Field, 2013). A principal axis factoring analysis was conducted on the scales. The Kaiser-Meyer-Olkin measure verified the sampling adequacy for the analysis (KMO = .944). Bartlett’s test of sphericity χ2 (78) = 2750.07, p < 0.001, indicated that correlations between items were sufficiently large for principal axis factoring analysis. An initial analysis was run to obtain eigenvalues for each component in the data. Two components had eigenvalues above Kaiser’s criterion of 1 and in combination these
41
components explained 68.75% of the variance observed. In agreement with Kaiser’s criterion, examination of the scree plot revealed a leveling off after the second factor. Thus, these two factors were retained and rotated with an Oblimin with Kaiser normalization rotation. Table 3 lists the factor loadings after rotation. The items that cluster on the same factors suggest that factor 1 represents willingness to adopt, and factor 2 represents perceived safety.
Table 3 Rotated factor loadings
The results suggest four of the items that were meant to measure the perceived safety show high loadings on the factor of WillAdopt. This could be due to common source bias. Common source bias is a risk in situations where the common measurement contains a source of error that appears in both the independent variable and the dependent variable (Doty & Glick, 1998; Richardson et al., 2009). According to Meier and O’Toole (2012), common source bias can be a serious problem when researchers rely on the perceptions of individuals. In this research, common source bias could have manifested in participants rating their general feelings about autonomous cars instead of rating specific items that
42
measured their willingness to adopt this innovation or their perceived safety of autonomous cars. General formulated survey questions can reinforce common source bias while more specific questions can reduce the chance of this bias (Meier & O'Toole, 2012). For this reason it was decided to remove four questions, which are indicated in Table 3 in light gray. In addition to the fact that these questions exhibit high loading on willingness to adopt, they are more general and less specific than, for example, the item “The self-driving car can cause safety consequences triggered by technical error.”
4.1.4 Reliability
Reliability is the extent to which data collection techniques or analysis procedures result in consistent findings. Reliability enables examination of the consistency of the measures used in the research (Saunders & Lewis, 2009). Cronbach’s alpha was measured for the scales of perceived safety and willingness to adopt regarding internal consistency. Cronbach’s alpha of > .8 indicates a high level of internal consistency, while a value of > .6 indicates an adequate consistency (Field, 2013).
After conducting the reliability check, the item “Emotional State: Surprised” was also removed. This item had a low correlation with the total score of the scale and after the removal of the item the reliability of perceived safety changed from “poor” to a “questionable” (Cronbach’s Alpha = .57) as illustrated in Table 4. The developers of the three emotional state questions created the items specifically as a trio to measure perceived safety, making it plausible to remove this last emotional state question as well (Bartneck et al., 2008, p. 78). The corrected item-total correlations indicate that the items remaining show good correlation with the total score of the scale (both are above .30). Furthermore, none of the items would substantially affect reliability any further if they were removed.
43
The willingness to adopt scale has high reliability, with a Cronbach’s alpha value of .94. In addition, the corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .30). Removal of one of the items would not significantly affect the reliability.
4.1.5 Computing scale means
New variables were created as a function of existing variables by calculating the means of all items in order to test the hypotheses. Means, standard variations, skewness and kurtosis of the two factors are exhibited in Table 4. The distributions for perceived safety and the willingness to adopt are exhibited in Figure 6 and Figure 7. Table 4 Descriptive statistics for the two factors Figure 6 Distribution of perceived safety 0 5 10 15 20 25 30 1 1,5 2 2,5 3 3,5 4 4,5 5 P er ce n ta ge ( % ) Perceived Safety
44 Figure 7 Distribution of willingness to adopt 4.2 Correlation matrix Correlational research implies that data will be analyzed so as to look at relationships between variables rather than making statements about causality (Field, 2013). Multiple combinations of variables were checked for correlation, as seen in the SPSS correlation matrix in Table 5. It can be concluded that perceived safety correlates with willingness to adopt (r = .41, p < .01). This means that as people perceive the autonomous car to be safer, their willingness to adopt it is higher as well. The correlation matrix also shows a significant correlation between reading safety facts and perceived safety (r = .14, p < .05). This indicates that people who have read facts about the safety of the autonomous car are more willing to adopt it. Although it is not a significant correlation, it is interesting that reading negative safety news has a very small correlation with both perceived safety (r = - .05, p > .05) and willingness to adopt (r = -.02, p > .05), as this was not expected.
As the correlation matrix indicates, there is a significant negative correlation between both age and willingness to adopt (r = -.28, p < .01) and age and perceived safety of autonomous cars (r = -.13, p < .05). The older people are, the lower their level of perceived 0 2 4 6 8 10 12 14 16 18 20 1 1,5 2 2,5 3 3,5 4 4,5 5 P er ce n ta ge ( % ) Willingness to adopt
45 safety and the less willing they are to adopt the autonomous car. Another interesting finding is the negative correlation between gender and willingness to adopt (r = -.20, p < .01) (male = 1, female = 2), which indicates that women are less willing to adopt the autonomous car than men.
Finally, familiarity provides interesting findings for discussion. Familiarity with autonomous cars was found to correlate with both perceived safety (r = .21, p < .01) and willingness to adopt (r = .39, p < .01). This indicates that a person who is more familiar with autonomous cars has both a higher level perceived safety regarding such vehicles and a higher willingness to adopt them.
46 Ta bl e 5 Co rr el at io n m at rix
47
4.3 Hypotheses testing
A regression analysis was conducted to examine the hypotheses. The PROCESS macro written by Andrew F. Hayes was used (Hayes, 2012). To match the conceptual model with the models used in the PROCESS macro, the conceptual model was split in two. PROCESS model 7 for a moderated-mediation effect was used to test the first part of the conceptual model, which included hypotheses 1, 2, 3, and 6 (please see Figure 8). For the second part of the conceptual model, PROCESS model 1 was used for simple mediation. This part of the conceptual model also included hypothesis 1, as well as hypotheses 4 and 5 (exhibited in Figure 9). Figure 8 Regression model part 1 Table 6 Outcomes regression model part 1 As the results in Table 6 indicate, participants exhibiting relatively higher levels of perceived safety showed higher willingness to adopt the autonomous car (b1 = .566, p =
48
.000). This means that the first hypothesis can be confirmed. The direct effect of consumers reading negative news about the autonomous car on perceived safety is negative, but not significant (a1 = -.075, p = .414). Similarly, there is no evidence for a direct relationship between reading negative safety news and the willingness to adopt the autonomous car (c1’ = .002, p = .987). This indicates that both hypotheses 2 and 3 are rejected. Furthermore, there is no significant interaction between news and facts on perceived safety, which leads to the rejection of hypothesis 6. Figure 9 Regression model part 2 Table 7 Outcomes regression model part 2 (1/2) Table 8 Outcomes regression model part 2 (2/2)
For the second part of the conceptual model (exhibited in Figure 9) a simple mediation analysis was conducted using ordinary least squares path analysis. Presenting people with safety facts about the autonomous car indirectly influences their willingness to
