Consumer adoption of self-driving cars : a diffusion of innovation perspective

(1)

Consumer Adoption of

self-driving cars: a diffusion of

innovation perspective

By Zara Pardjinjan, 11419547

MSc Business Administration, Entrepreneurship & Innovation track Faculty of Economics & Business, University of Amsterdam

Supervisor: dr. Wietze van der Aa 18 August, 2017

(2)

Statement of originality

This document is written by Zara Pardjinjan who declares to take full responsibility for the contents of this document.

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 creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

The self-driving car is moving from the realm of a science fiction fantasy to becoming a reality in the very near future. Numerous car companies are currently working hard on introducing the first truly autonomous (level 4 and up) vehicles to the consumer market within the next few years. Although several technological, legal and ethical challenges first need to be overcome, it has frequently been noted that the main determinant of mass market success of the self-driving car will be the consumer acceptance. The study uses the model of diffusion of innovation to determine and examine important factors that will determine the adoption of the self-driving car. Several differences in concerns and expected benefits were found between the early adopters and the early majority. The research also showed that as expected, the more people know about self-driving cars, the more benefits they perceive in adopting the technology. Additionally, the more knowledge people have about self-driving cars, the less concerned they are about potential risks of adopting the technology. Also, both knowledge and innovativeness have a positive effect on the intention to adopt the self-driving car. Several ideas for future research are presented.

(6)

1. Introduction

The advent of self-driving cars, or autonomous vehicles becomes less of a futuristic scenario and more of a reality every day. The changes could be potentially revolutionary as they will change the way vehicles are built and used, completely reshape existing infrastructure and cities, as well as the interaction between humans and machines (Silberg et al., 2012).

Although first steps have been taken towards understanding the customer acceptance and willingness to adopt the technology (Schoettle & Sivak, 2014; König, & Neumayr, 2017; Kyriakidis et al., 2015), there is still a lot to be learned. Deeply divergent scenarios have been painted for the adoption of the self-driving technology. This shows that still too little is known about the customer opinion and intention to adopt self-driving cars. Due to the novelty of the technology, there is no extensive body of literature and model for researching the topic. There have, however, been studies of exploratory nature that have attempted to gauge the public opinion on self-driving cars.

The following research attempts to study the adoption and diffusion potential of self-driving cars by adopting existing theories and concepts. By taking first steps towards studying the adoption of this technology in an academically grounded way, it is setting a groundwork to further built upon. Such development should eventually lead to better predictive models, thereby decreasing the uncertainty that currently surrounds the diffusion of the self-driving car. Hence, the diffusion of innovation model by Rogers (2003) is used.

The theory states that adoption of new technologies is said to follow a particular curve. In that process generally adoption occur starting with the innovators and ending with the laggards, but crossing the chasm between the early adopters and early majority is the true determinant of mass market success for disruptive innovations like the self-driving car (Moore, 2014). Additionally, people will decide to seek knowledge about a technology and adopt it only if they see a relative advantage (Rogers, 2003).

The rapidly growing body of literature, academic and non-academic, discusses several advantages of the self-driving car. It is said to eliminating the driver from the driving equation, can improve mobility for elderly and disabled, has the potential to substantially improve safety, increase human efficiency, fuel efficiency, and the list goes on (Beiker, 2012; Silberg et al.,

(7)

2012; Douma, & Palodichuk, 2012). However, the more important question is whether the people, and particularly the first few adopter categories, perceive the value of such benefits. Hence, the following question will be answered:

Research question 1: How do the expected benefits of self-driving cars differ for the early adopters when compared to the early majority?

This question is expected to give better insight into the current attitudes about self-driving cars for both adopter groups. Additionally, it will show which benefits are currently not being perceived by the people and serve as guidelines for car manufacturers to focus on during their communication efforts.

Closely related to the benefits, are the concerns, as they will also be a great determinant in the decision process when people are weighting the relative advantages and disadvantages of adoption. Hence, the following question will be answered:

Research question 2: How do the concerns about self-driving cars differ for the early adopters when compared to the early majority?

Studying the biggest concerns for these adopter groups, valuable insights can be gained on what the biggest concerns currently are that could potentially hinder the full adoption of self-driving cars. These can similarly serve as guidelines for car manufacturers to focus their attention on.

According to Rogers (2003), the innovation-decision process starts with the knowledge stage, where an individual is figuring out what an innovation is and how it works. The information seeking process in itself is an indication that an individual is considering adoption of technology, but increased knowledge is also said to increase the chance of adoption and is even more important with complex technological innovations. To get insights into the current level of knowledge of potential customers, the following question will be looked into:

Research question 3: How does knowledge about self-driving cars affect the interest in adopting the technology?

In order to answer the third research question, three different hypothesis will be tested. It is assumed that information seeking activities will decrease uncertainty about the adoption of a

(8)

technology because people will perceive more benefits and see less potential risks because they understand the technology better (Rogers, 2003). Hence, the following two hypotheses are tested:

H1: The more knowledge people have about self-driving cars, the more benefits they perceive in adopting the technology.

H2: The more knowledge people have about self-driving cars, the less concerned they are about potential risks of adopting the technology.

Individuals who indicate to have deep knowledge about the technology, can be assumed to have engaged in information-seeking activities by themselves. According to Rogers (2003), this indicates a first step towards an interest to potentially adopt. Since the innovators and early adopter are generally the first in the technology adoption lifecycle to do so, it is assumed that a person’s innovativeness will have a positive moderating effect on the relationship.

H3: The more knowledge people have about self-driving cars, the more likely they are to be interested in adopting the technology, and this effect gets strengthened by innovativeness.

Several differences in concerns and expected benefits were found between the early adopters and the early majority. The research also showed that as expected, the more people know about self-driving cars, the more benefits they perceive in adopting the technology. Additionally, the more knowledge people have about self-driving cars, the less concerned they are about potential risks of adopting the technology. Also, both knowledge and innovativeness have a positive effect on the intention to adopt the self-driving car. Several ideas for future research are presented.

(9)

2. Literature review

The following chapter provides an extensive review of literature, theories and concepts related to the topic at hand. It is divided into the following sections: diffusion of innovation, crossing the chasm, self-driving cars and diffusion of self-driving cars.

2.1 Diffusion of innovation

For over 40 years the process of adopting new innovations has been studied. Although, the origins of diffusion research can even be traced back to the beginning of the 20th century. In “The laws of imitation” Gabriel Tarde (1903) stated that invention and imitation were the keys of cultural change. In the years that followed, several researchers further contributed to the advancements in diffusion research (Ryan & Gross, 1943; Menzel & Katz, 1955; Coleman et al., 1957). However, the topic did not get real traction until 1962 when Everett M. Rogers published the first edition of his book ‘Diffusion of Innovation’. In his book Rogers synthesized research from over 508 diffusion studies and built a theory to explain the adoption of innovations among individuals and organizations (Rogers, 1962). The concepts described by Rogers have ever since served as prominent literature for understanding the (technology) innovation adoption process. Additionally, they have shown to be applicable to several fields (Haider & Kreps, 2004; AlGahtani, 2003; Kilmon and Fagan, 2007; Stuart, 2000; Tabata & Johnsrud, 2008). Meyer (2004) estimated that the model has been used in thousands of studies across several disciplines including marketing, communications, economics, public health, education, and technology.

Rogers (2003, p. 11) defines diffusion as the “process by which an innovation is communicated through certain channels over time among the members of a social system”. In this definition an innovation is referred to as “an idea, practice, or project that is perceived as new by an individual or other unit of adoption” (Rogers, 2003, p. 12), where the focus is on technological innovations and “innovation” and “technology” are used as synonyms. Notably, with diffusion studies the focus is not on “objective” measures of newness of an innovation but on perceived newness by a particular individual.

Based on Rogers’ definition of diffusion (2003, p. 11), it is said that the described process can have a different rate of adoption under different circumstances. How fast an innovation is adopted is highly dependent on the perceived attitudes of individuals about this innovation. How

(10)

an innovation is perceived is influenced by several of its attributes. According to Rogers (2003), the relative advantage is the strongest predictor of the rate of adoption of an innovation. If an innovation is perceived to be far better than the idea it supersedes, then an individual is more likely to adopt an innovation. What constitutes ‘better’ and which factors are taken into consideration is different for each individual (it could be related to economic gains, status, safety, convenience etc.). In the Technology Acceptance Model (TAM) it has equally been recognized that perceived usefulness of a technology or system is one of the most important adoption determinants (Davis, 1989).

It is generally implied that a technological innovation will have some degree of benefit or advantage for the potential users (Rogers, 2003). However, these benefits or advantages are often not clearly visible to the potential adopters. They will be doubtful whether the innovation represents a superior alternative to their current practice, which creates uncertainty in their minds. This uncertainty mostly revolves around the expected consequences of adopting an innovation. Potential adopters will weight these consequences against the possible efficacy of the innovation in solving a perceived problem or catering to a felt need. In case such potential benefits are expected, an individual will be motivated to exert effort in learning about the innovation (Rogers, 2003). These information-seeking activities will reduce the uncertainty about the possible consequences of the innovation to a tolerable level for the individual. Based on the obtained information, the individual is now ready to make a decision: adopt or reject an innovation.

Following this logic, Rogers (2003) describes the innovation-decision process as “an information-seeking and information-processing activity, where an individual is motivated to reduce uncertainty about the advantages and disadvantages of an innovation” (p. 172). This process involves five steps, which are typically said to follow an order: (1) knowledge, (2) persuasion, (3) decision, (4) implementation, and (5) confirmation.

The innovation-decision process starts with the knowledge stage, where an individual is figuring out what an innovation is and how it works. This knowledge can be awareness-knowledge, which indicates that an individual is aware of the existence of the innovation. But it can also be how-to-knowledge, which is knowledge about how to correctly use the technology. According to Rogers (2003), this knowledge is an essential component in the

(11)

innovation-decision process. Providing sufficient how-to-knowledge can increase the chance of adoption and is even more important with complex technological innovations. Finally, there is the principles-knowledge, which is related to the functioning principles about how and why an innovation works. In the process of information-seeking and learning about a new technology, individuals are generally not as much concerned with the technological aspects of an innovation, but rather with the expected advantages and disadvantages that will occur in their situation (Rogers, 2013). Hence, information about technology should not only focus on providing a how-to experience but also a know-why experience (Seemann, 2003).

The second stage in the innovation-decision process is persuasion, which refers to an individual forming a favorable or unfavorable attitude toward an innovation. According to Rogers (2003), this stage is more based on feelings than knowledge. The formation of a certain attitude is affected by friends, peers, colleagues etc. Certainly there are large amounts of expert opinions and scientific evaluations available about new technologies. However, individuals more often seek out and are convinced by the subjective opinions of trusted people in their environment (Sherry, 1997). After the persuasion stage an individual decides whether to adopt or reject the innovation. In case of acceptance, the next stage is implementation and the individual actually starts using the innovation. Finally, in the confirmation stage an individual is seeking reinforcement to be confident that they have made the right decision.

2.2 Crossing the chasm

Effectively Rogers’s (2003) theory seeks to explain how individuals perceive new ideas and innovations and how they deal with the risk and uncertainties associated with them. The diffusion process starts by a few individuals adopting the innovation, communicating this within their network and hereby lowering the uncertainty, and causing more and more people to adopt the innovation. It further states that the propensity of individuals to adapt a specific innovation differs. Particularly, diffusion theory assumes that a population can be broken down into five adopter categories based on their innovativeness: (1) innovators, (2) early adopters, (3) early majority, (4) late majority and (5) laggards.

Innovators are characterized by their venturesomeness and eagerness to try out new ideas. They are generally curious and are able to understand and apply complex technical knowledge. This curiosity leads them to seek out innovations which are so new that adopting

(12)

them is considered daring and risky. Consequently, innovators must be able to deal with this level of risk and accept an occasional setback when an innovation turn out to be unsuccessful and unprofitable. This means they must also possess sufficient financial resources to cover the potential losses caused by such unsuccessful innovations. Innovators are said to have an important gatekeeping role as they bring innovations in from outside the system’s boundaries. With just 2.5% they represent the smallest portion of the population.

Early adopters are characterized by their prominent leadership position when it comes to giving advice and information about an innovation. They represent the next 13.5% of the population and are thus one of the first to try out new technologies, but are not too far ahead from the average individual in innovativeness. This makes them to be role models and opinion leaders for the rest of the society. To maintain such respectable role with their colleagues and the society, they are expected to continuously make sound innovation decisions. Within the system, the role of early adopters is to decrease uncertainty about a new technology by trying it out and communicating their experiences through their network.

The early majority is more deliberate and takes longer than the innovators and early adopters to adopt a new idea. They represent the next 34% of the population and thus they still adopt new technologies just about before the average person in the social system. This gives them an important role because they are the link between the few early birds and the greater masses.

The late majority is more skeptical when it comes to new ideas and technology. They represent the next 34% of the population and thus adopt new technologies after the average person in the social system. They are cautious and will only adopt new technologies with low uncertainty because they have seen several peers and people around them use it successfully.

Laggards are the last 16% of the population to adopt new ideas or technology. They have traditional views and often mostly interact with individuals who have similar values. They are cautious and their point of reference is often the past. By the time laggards adopt a technology, a new version of that technology might already be out there or in development.

Technology adoption is said to occur gradually and in the order of most innovative to the least innovative individuals in the society, forming an S curve when plotted over time (Rogers,

(13)

2003). Using the technology adoption lifecycle has been found particularly useful for understanding, predicting and stimulating the adoption of innovations that do not force a significant change of behavior by the customer. However, in his book ‘Crossing the Chasm’, Moore (2014) further builds on the existing theory by making it specific to disruptive or discontinuous innovations. According to Moore (2014, p. 7), these are the type of innovations that “require us to change our current mode of behavior or to modify other products and services we rely on”. With this type of innovations there exists a chasm between the early adopters and the early majority (Figure 1).

Figure 1: Chasm in the Technology Adoption Lifecycle

Several disruptive innovations in the past have not been able to reach mainstream adoption because they were not able to cross the chasm. This means the early majority did not see sufficient value and benefits to make true changed in their behavior or the product/service they use. According to Moore (2014), the main reason for such chasm is that the two groups have very different expectations when they adopt a new innovation.

The early adopters wish to get an edge by being the first ones in their environment to use a new technology. They expect radical change and are prepared to leave their old ways for it. Because they are the first ones to use a technology, they are tolerant to any bugs and glitches that might be associated with the new technology. The early majority, on the other hand, expects the new innovation to be an improvement on their current state. They want a new technology to

(14)

complement or enhance the current state, rather than eliminating it. They are not as tolerant to bugs because they expect the technology to work properly and effortlessly fit in their life.

For companies it is said to be of high importance to understand the needs, expectations and concerns of each adopter group and apply communication strategies accordingly (Moore, 2014; Kotler & Zaltman, 2001; Engel, Blackwell & Miniard, 1986). However, it should be noted that such generalizations carry the risk of oversimplifying the complex human nature. Assigning a person to an adopter category based on their general personal innovativeness is a useful classification, but might not be a good predictor for their adoption behavior for all products and services. This means one might score very high on innovativeness and be qualified as an innovator, and exhibit suiting adoption behavior for one product category, but behave like the late majority on adoption of some other product category based on their personal interests. Hence, practitioners should go beyond applying generalized strategies for each adopter group. They should try to understand the subtle differences in needs and concerns for each adopter group for their particular product or service.

2.3 Self-driving cars

2.3.1 Levels of autonomy

In 2013 the National Highway Traffic Safety Administration first introduces a policy outlining automation levels 0 to 4 for vehicles (NHTSA, 2013). Building on this policy, the Society of Automotive Engineers (SAE International, 2016) has adopted an additional level and has thereby created a classification that is being used universally. The levels that have been defined are based on the level of intelligence of a driving system, with level 0 being completely unautomated, and level 5 being completely automated:

 Level 0: Not automated; A human driver is required for all aspects of driving.

 Level 1: Supportive automation; The car can sometimes assist the driver in one aspects of driving, like steering and accelerating. This is mostly done with parking support or adaptive cruise control.

 Level 2: Partial automation; At this level, two or more aspects of driving are assisted or taken over by the system. An example of this is adaptive cruise control, lane-keeping assist and automatic emergency braking. Although at level 2 the car can fully take over certain tasks of the driving task for extended periods, the driver is still fully responsible and is required to always supervise the operation of the autonomy and take over control when the system fails to see danger ahead.

(15)

 Level 3: Conditional automation; At this level, the car is able to drive completely independently on certain parts of the journey. However, it will give warnings at moments when it is not capable of driving independently. At that point the driver is urged to take over the control within a short amount of time.

 Level 4: High automation; This level of automation means that he vehicle is able to drive itself for the entire journey, even when unmanned. However, the system is limited by the operational design domain, meaning it can only do the tasks it was designed for. Hence, it can only drive on specific roads and under specific traffic and weather conditions.

 Level 5: Full automation; The vehicle is at the level of a human driver and is able to adjust to every driving scenario.

Most of the cars on the road currently have level 1 automation as the great majority of the car manufacturers offers such features (Silberg et al., 2012). They are capable of assisting with maintaining speed and direction, but the driver is still responsible for all the critical tasks like braking, switching lanes, making turns etc. Several top-of-the-range models on the road today have semi-automated driving systems that classify as level 2. These include Tesla’s Autopilot system, BMW’s Active Driving Assistance Plus and the Mercedes Benz Drive Pilot (Godsmark, 2017). The Tesla Autopilot system is said to be the most advanced, because it measures the torque of the steering wheel to detect whether the driver is still paying attention (Cole, 2017).

Currently, there are no vehicles of level 3 available to the broader public. However, several car manufactures are testing such models in specific operational design domain sites. It is said that Google had already reached level 3 automation in 2012, but has decided not to commercialize such models. Google staff members were given the opportunity to test these models but it soon became evident that people were putting too much faith in the system and not responding to the handoff request fast enough (Godsmark, 2017). Multiple other manufacturers have similarly experiences the challenges of people disengaging, slow reorientation and all the safety risks associated with it (Sputnik, 2017). Several car companies, such as Volvo and Ford, have indicated that they will skip level 3. This is because they believe there is no way to ensure the safety of the driver, and that would decrease the development of the next step (Cole, 2017). These manufacturers are aiming to perfect their models until they reach level 4 qualification before launching them to the public.

Moving from level 3 to level 4 is a big step as this is the point where the responsibility of the full driving task is switched from the human to the car. This is when we step into the sphere

(16)

of ‘autonomous’ or ‘self-driving’ cars. Hence, when this report refers to the term “self-driving car”, it is defined as a car with a least a levels 4 automation system. According to SEA International (2016), it is highly likely that a number of manufacturers will have level 4 autonomous systems operating on specific locations within a decade. Indeed many of the manufacturers are currently working on self-driving vehicles. The first ones will be expected at the end of 2018, with Tesla, Google, the Nvidia and Bosch alliance, and the Chinese search machine Baidu all in the race for the first release. These will be followed by the larger brands of car manufacturers in 2020, including Audi, Toyota, GM, Nissan and Volvo. Ford and BMW are expected to trail, aiming for delivery in 2021 (Driverless Future, 2016; Muoio, 2017).

2.3.2 Self-driving technology

For a vehicle to operate at level 4 automation, it needs to know exactly what its position is, what is happening in its surroundings and predict what the possible movements of other drivers could be. To be able to determine all these factors, the vehicle needs additional equipment which is lacking in a standard vehicle. One of the most important of these is the laser (Silberg et al., 2012; James, 2001). This laser is used in the same manner a radar installation is used. It sends out a signal, and awaits the return of this signal to measure the distance between itself and the object. However, where a radar uses sound waves to measure distances, a laser uses light, giving it a much faster response and thus allowing for faster and more accurate calculations of an object’s position. Naturally, these lasers need to be angled in 360 degrees, to be able to give a clear image of the vehicle’s surroundings (Kilic et al., 2015).

In some vehicles currently under development, such as the self-driving car of Google, a radar system is also added merely for measuring the objects and drivers in front of the car to be able to adjust its speed and allow for a safer participation in traffic (Waymo, 2017; Withwam, 2014). Furthermore, the vehicle requires an orientation sensor (Lee et al., 2013). This sensor can be compared to the sensor in a smartphone. This sensor is able to detect the position of the car, whether it’s tilted, which direction it faces, etc. This allows for better detection of the underground of the vehicle, like driving uphill or on dirt roads. One other sensor required is the position sensor. This one is located in the wheel hub of the car, and measures the rotation of the wheel (Straub et al., 2017). With this information, the vehicle is able to calculate its own position more accurately. Naturally, the data supplied by these sensors and measuring equipment is

(17)

useless if it is not processed properly (Nvidia, 2017). To be able to do so, strong processors are required. These processors are loaded with all the data required to calculate where the vehicle is going, if an object in front of the car is a bicycle or a car, and how fast it is going. The coding required for things are immense, and are continuously being updated in order to improve the workings of the self-driving car.

2.3.3 Technological challenges

Before self-driving vehicles are safe to be introduced to the consumer market, there are several technological challenges that need to be overcome. The first one is detection, which needs to be perfect if the vehicles want to reach level 4 of autonomous driving.

The vehicle needs to be able to operate in every scenario possible, be it rain, snow, or sun. As mentioned in the previous paragraph, a lot of the detection is done by lasers radars and cameras. These detection methods work sufficiently as long as the weather is good and the car is in an urban area. Since water and snow have different reflective and sound absorbing properties, the weather conditions are said to drastically decrease detection (Malinka et al., 2016). With heavy rain, the laser will have little problem, but the radar system will function less since the soundwaves get dampened by the water. On the other hand, the laser will have much more problems with heavy snow, since the reflective properties of the snow are much higher than what the laser is used to. This could cause it to detect objects which aren’t actually there and react based on this information (Sun et al., 2006). If the system suddenly detects an object in front of the car which turns out to be a snow mount, the car will break heavily, causing potentially dangerous situations. These are extreme conditions, but the same challenges are present with bright and sunny weather.

In May 2016, Joshua Brown was driving a Tesla Model S with the autopilot engaged. While the car was driving, he had not touched the steering wheel for 37 minutes. Because of the bright sunlight, the autopilot failed to detect a trailer that was making a turn, and the Tesla crashed into it, killing the driver (Thompson, 2017; Dikmen & Burns, 2016). This real life example shows that the current sensory technology is not mature enough to become a standard component in millions of cars (Simonite, 2017). Until such level of maturity is reached, the entire system needs a high degree of redundancy to assure that if one of the detection systems fails, there are back-up systems to compensate.

(18)

Other than physical objects that need to be detected, the vehicle also needs to know the rules and laws of the roads and area its driving upon. Until there is a highly adaptive mapping system (GIS) that tells all smart cars how fast they are going, where they need to stop, where pedestrian crossings are etc., the vehicle needs to be able to read all the signs (Brewster, 2017). At this moment, this appears to be quite difficult. The normal signs can be read easily by an image recognition system, but testing has shown that these systems can be confused with a simple adjustment to the sign (Fingas, 2017).

Perfecting the detection systems is relatively easy compared to the challenge that manufacturers face with perfecting the artificial intelligence of the system. One of the main challenges in this area is the behavioral analysis. A human driver is generally well capable of forming context and predicting what other drivers might do. If a ball rolls in front of the car, an average person knows there is a possibility that this ball is followed by a child trying to catch it, and thus would slow down quickly before an accident occurs. At the current level of intelligence, a vehicle might detect the ball rolling in front of the car, but does not know enough to also expect the child following the ball (Hamers, 2016; Lee et al., 2013). Another challenge with the current state of the artificial intelligence is dealing with traffic controllers. The vehicle must first detect a controller and realize that the general traffic signs are overruled by the controller. Secondly, the vehicle must read and understand the signing of the traffic controller. Although the system can be taught the general hand signaling, slight deviations in human motion can confuse the system and cause dangerous situations (Cools et al. 2013).

A final challenge that should not be overlooked is the security. With so many computer systems and intelligence, the possibilities of hacking has become greater (Glancy, 2012). In 2015 it was already demonstrated that it was possible to hack the digital system of a Chrysler Jeep (The Guardian, 2015). If everything within a vehicle becomes automated and connected, this means it is susceptible to more forms of hacking. A hacker could gain control over essential functions like the braking, the speed, steering and locking and unlocking of a car (Press, 2017). This opens up opportunities for several forms of dangerous and criminal behavior.

2.3.4. Ethical challenges

A challenge that has been topic to many debates is the ethics behind the artificial intelligence of a self-driving car (Goodall, 2016; Bonnefon et al., 2015; Belay, 2015). To demonstrate the ethics

(19)

of the system, the trolley problem is often discussed (Nyholm & Smids, 2016; Shashkevich, 2017). The trolley problem is an ethical dilemma presented by Thomson (1985) in which you are said to stand next to a train track with a train (trolley) approaching fast. On the track there are 5 workers, who do not notice the train that approaches. If the train is left unstopped, these people will certainly die. However, you are standing next to a lever, which will allow you to switch the train to another track. On the other track stands only one worker who will certainly die. To many utilitarians it seems like a simple numbers game: five lives are worth more than the one. This case becomes more difficult if you realize that you are making the conscious choice to kill the one. The ethical challenge becomes even more complex if the age and health of the workers are taken into account. A self-driving car might very well be presented with similar situations and it needs to be programmed the “right” decisions.

Then there is also the aspect of the driver of the vehicle. A car is generally designed to protect the driver, but if a choice must be made between hurting several others or crashing the car into another object, the choice is not so clear-cut anymore. Although legally and statistically there might be a best possible decision (ArXiv, 2015; Beiker, 2012), the ethical solution might be of a complete different nature. Additionally, experts believe that this is a problem that has to be handled properly, because people are not willing to drive in a car which might consciously kill them (Silberg et al., 2012; Martens & Jenssen, 2012).

2.3.5 Legal framework

Another important aspects for consumer market adoption of the self-driving car is the legal framework. The Vienna Convention on Road Traffic, signed in 1968, states that a vehicle needs to have a driver, and that the driver is required to have hands on the steering wheel at all times (United Nations, 1968). This treaty greatly decreases the testing abilities with self-driving vehicles on actual roads. However, on March 10th of 2017, the Department of Motor Vehicles in California signed for a change in those laws (State of California, 2017), allowing the companies in Silicon Valley to test their vehicles on the actual infrastructural system. Similarly, European law makers are moving to a more lenient legal framework. Germany, Sweden, Finland and the Netherlands have already allowed some testing in restricted environments (Plucinska, & Posaner, 2016). For full autonomous driving (level 5) to be possible, law maker need to not only allow self-driving cars on the road but also support vehicle (V2V) and

(20)

vehicle-to-infrastructure (V2I) interaction. Only a mandate on a national or international level that makes it mandatory for all cars currently on the road to be equipped with V2V software can make big scale adoption of the self-driving car possible.

Another legal aspect that is being discussed more and more now that level 4 vehicles are closer to reality, is the liability issue. The central question is when is the vehicle responsible to do something, when is the driver responsible, and at what point exactly does the liability shift? In the example that was given earlier, Joshua Brown was in a vital accident in a Tesla Model S with the autopilot engaged. However, what is often omitted in this story, is the fact that the driver was given seven warning signals to put his hands back on the steering wheel, six of them with auditory signals. The driver ignored all signals. Tesla was sued for the accident, but the judge ruled in favor of Tesla, stating that the autopilot worked accordingly and gave plenty of indications that the driving task was handed off to the driver (Knight, 2016). As long as a level of fully autonomous vehicles (level 5) is not reached, there will always have to be a point where responsibility is shifted between the vehicle and the driver. But when the driver is responsible, when the vehicle should be responsible and where there should be an overlaps is as of now still fuzzy.

2.4 Diffusion of self-driving cars

The previous paragraph discussed several technological, ethical and legal challenges that the self-driving car still faces. However, the main barrier to market success for self-driving cars currently is with consumer acceptance, which is progressing at a much slower pace (Fitchard, 2012; Tingvall, 2014; Waytz, et al., 2014). However, opinions about if and when mass adoption of self-driving cars will take place differ greatly.

2.4.1 Adoption scenarios

As previously mentioned, new technologies usually follow an S-curve (Rogers, 2003; Christensen, 1992). KPMG did an extensive study about self-driving cars, in which they interviewed 25 industry leaders and experts about their expectations (Silberg et al., 2012). Although they agreed that they expect an S-curve type of adoption for self-driving cars, three different scenarios of adoption were generally given.

(21)

The first scenario is the one of aggressive adoption. This is a scenario envisioned by among others the CEO of Tesla, Elon Musk. In a recent TED talk, he restated his vision: a level 5 Tesla model will be driving around within the next two years. For an aggressive adoption scenario, it’s important that the systems are sufficiently optimized before launching and that the suppliers can make people see the benefits of self-driving cars. If this is the case, it is important that the systems are also immediately embraced by the proper authorities, where after they issue a mandate that includes aftermarket components, obligating people to install system to make their vehicles “smart vehicles”. The popularity of the system will cause even greater technological developments, and will increase the vehicle to environment abilities. This will then make more and more people adopt the system, reaching the critical mass within less than five years (Silberg et al., 2012).

A less aggressive scenario of adoption is the base case scenario (Parente & Prescott, 1994). In this case, the systems are sufficiently optimized before launch, and consumers embrace the benefits of self-driving cars. However, it is not immediately embraced by the proper authorities, causing a delay of adoption. Eventually, as more and more people start using self-driving vehicles, the authorities are obligated to issue a V2V mandate, but not necessarily with aftermarket components. This will cause another plateau of adoption, because the technology of self-driving vehicles cannot use its full potential. Finally, private companies will improve the aftermarket systems, allowing for a greater level of adoption. This this scenario, it is however not necessarily assumed that self-driving cars will reach 100 percent adoption. The IEEE (2012), for example, has stated that they expect 75% of all vehicles to be completely autonomous by 2040.

Then finally there is the conservative scenario (Davila et al., 2003). In this case, because the systems fail to deliver on the initial promises, the greater mass of the consumers will not adopt the system. Because of this lack of adoption, the authorities will not issue a mandate. Self-driving cars will not be able to utilize the full extent of their design. They will not be able to communicate with other (non-selfdriving) vehicles. The lack of interest from the general consumers will cause a delay in the technological development. Eventually, there will be a small rise in adoption as the development increases, but it will never reach the general populace (Hall & Khan, 2003). One of the supporters of this scenario is the American Automobile Association. They have done a studies in 2017 which concluded that 78% of the respondents would be afraid

(22)

to ride in a self-driving vehicle (Stepp, 2017). These fears are not easily assuaged, thus the initial embrace by the consumers, and hence mass adoption, will never occur.

2.4.2 Theoretical framework

The deeply divergent scenarios that were painted in the last paragraph show that little is known about the customer opinion and intention to adopt self-driving cars. Due to the novelty of the technology, there is no extensive body of literature and model for researching the topic. There have, however, been studies of exploratory nature that have attempted to gauge the public opinion on self-driving cars.

The following research attempts to study the adoption and diffusion potential of self-driving cars by adopting existing theories and concepts. By taking first steps towards studying the adoption of this technology in an academically grounded way, it is setting a groundwork to further built upon. Such development should eventually lead to better predictive models, thereby decreasing the uncertainty that currently surrounds the diffusion of the self-driving car.

Particularly, the diffusion of innovation theory of Rogers (2003) and related concepts will be used. At the current stage, the adoption of semi-automotive cars is possible, but speaking of adoption of self-driving cars is a relatively hypothetical situation. Hence, not all concepts might be valid or applicable. Additionally, studying the intention to adopt a technology has the risk that people might say they would adopt a technology, but will not do so in reality. However, according to Ajzen’s Theory of Planned Behavior (1991), studying intentions are still one of the best predictors of human behavior. Hence, the general concepts in the theory of diffusion of innovation have still been found useful and insightful for their predictive abilities (Helitzer et al., 2003).

2.4.3 Research on benefits and concerns

As was explained in section 2.1 and 2.2, adoption of new technologies is said to follow a particular curve. In that process generally adoption occur starting with the innovators and ending with the laggards, but crossing the chasm between the early adopters and early majority is the true determinant of mass market success for disruptive innovations like the self-driving car (Moore, 2014). Additionally, people will decide to seek knowledge about a technology and adopt it only if they see a relative advantage (Rogers, 2003).

(23)

The self-driving car is said to have several advantages over the current mode of driving. It eliminating the driver from the driving equation, can improve mobility for elderly and disabled, has the potential to substantially improve safety, increase human efficiency, fuel efficiency, and the list goes on (Beiker, 2012; Silberg et al., 2012; Douma, & Palodichuk, 2012). However, the more important question is whether the people, and particularly the first few adopter categories, perceive the value of such benefits. Hence, the following question will be answered:

Research question 1: How do the expected benefits of self-driving cars differ for the early adopters when compared to the early majority?

This question is expected to give better insight into the current attitudes about self-driving cars for both adopter groups. Additionally, it will show which benefits are currently not being perceived by the people and serve as guidelines for car manufacturers to focus on during their communication efforts.

Closely related to the benefits, are the concerns, as they will also be a great determinant in the decision process when people are weighting the relative advantages and disadvantages of adoption. Hence, the following question will be answered:

Research question 2: How do the concerns about self-driving cars differ for the early adopters when compared to the early majority?

Studying the biggest concerns for these adopter groups, valuable insights can be gained on what the biggest concerns currently are that could potentially hinder the full adoption of self-driving cars. These can similarly serve as guidelines for car manufacturers to focus their attention on. 2.4.4 Knowledge

According to Rogers (2003), the innovation-decision process starts with the knowledge stage, where an individual is figuring out what an innovation is and how it works. The information seeking process in itself is an indication that an individual is considering adoption of technology, but increased knowledge is also said to increase the chance of adoption and is even more important with complex technological innovations. To get insights into the current level of knowledge of potential customers, the following question will be looked into:

(24)

Research question 3: How does knowledge about self-driving cars affect the interest in adopting the technology?

In order to answer the third research question, three different hypothesis will be tested. It is assumed that information seeking activities will decrease uncertainty about the adoption of a technology because people will perceive more benefits and see less potential risks because they understand the technology better (Rogers, 2003). Hence, the following two hypotheses are tested:

Individuals who indicate to have deep knowledge about the technology, can be assumed to have engaged in information-seeking activities by themselves. According to Rogers (2003), this indicates a first step towards an interest to potentially adopt. Since the innovators and early adopter are generally the first in the technology adoption lifecycle to do so, it is assumed that a person’s innovativeness will have a positive moderating effect on the relationship.

Based on this hypothesis, the following framework in figure 2 was drawn.

(25)

3. Data and method

The following chapter will discuss the methodology used for conducting the research, divided in two sections: data collection and methods.

3.1 Data collection

For conducting the empirical research, a structured questionnaire (see appendix 1) was developed in Qualtrics. The focus when designing was to make the survey clear and generally possible to answer within five minutes in order to keep the dropout rate as low as possible. To create ideal conditions for not just the quantity but also the quality of the responses, the ‘force responses’ option was selected in Qualtrics to assure that no questions could be left unanswered. First, a pretest was conducted where three people with different educational levels were asked to fill in the questionnaire. The feedback showed that the questionnaire was mostly clear and could be completed within reasonable time. Hence, no noteworthy changes were made except for minor simplifications in the terminology used.

All responses were collected online as this is time and cost efficient, allows for an appealing visual design and ease of use after collection (Lumsden, 2007). A convenience sampling method was used. Although no structured sample stratification was applies during the distribution of the survey, special attention was put to including respondents with diverse characteristic in order to avoid selection biases and extreme underrepresentation of specific groups. Hence, people in different age ranges, with different educational backgrounds and from both rural and urban areas were included. To further assure the representativeness of the sample, it was aimed to obtain more than 200 respondents (Saunders et al., 2012). A total of 221 responses were collected, of which 30 were not fully completed. This yielded 191 valid responses.

3.1.1 Innovativeness scale

To assure that the respondents do not get influenced by their concern or excitement about self-driving cars, a general question about innovativeness was asked first. Consumer innovativeness for years has caught the attention of researchers and practitioners alike. Steenkamp & Baumgartner (1999) defined innovativeness as “a predisposition to buy new and different products and brands, rather than remain with previous choices and consumer patterns”. Other

(26)

consumer related research defines innovativeness as “a propensity to adopt new ideas and techniques earlier than one’s peers” (Von Fleckenstein, 1974) or “the degree to which an individual is receptive to new ideas” (Midgley & Dowling, 1978).

To assure high validity and reliability of the instrument, an existing scale was chosen for measuring the innovativeness variable. Several scales have been designed to measure the innovativeness of individuals (Price & Ridgway, 1983; Jackson, 1976; Kirton, 2011; Hunter et al., 2012). As the research is focused on studying the customer adoption of self-driving cars, the focus is on capturing consumers’ behavioral expectations for adopting a product in the future, rather than actual adoptive behavior. The most appropriate scale under these conditions was found to be the ‘Innovativeness Scale’ by Hurt et al. (1977), which measures the degree of innovativeness and willingness to change. Items are focused on aspects such as personal value for tradition, curiosity and risk taking. Moreover, the scale is based on the diffusion of innovation model and allows assignment of respondents to the five adopter categories (Berwick 2003, Rogers 2003).

To make the scale as suitable to the research at hand as possible, modest changes in terminology were applied to some of the items. Mostly, general terms such as “an idea” or “an innovation” were replaced by “technology”. The original scale asks respondents to rate 20 items. However, to avoid survey fatigue and increase the completion rate of the survey, the shorter 10-item version of the scale suggested by Hurt et al. (1977) was used to measure innovativeness. The respondents were asked to indicate their level of agreement or disagreement with each of the ten items. Their responses were scored as follows: 1 = strongly agree, 2 = agree, 3 = somewhat agree, 4 = neither agree nor disagree, 5 = somewhat disagree, 6 = disagree, 7 = strongly disagree. Using this scoring, individuals with the highest scores are said to be most innovative, whereas individuals with the lowest scores are said to be least innovative. Three out of the ten items were counter-indicative and had to be recoded.

The Innovativeness Scale was found to have a high reliability, with Cronbach’s Alpha = 0.877. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above 0.30). Also, none of the items would substantially affect reliability if they were deleted. The ten items were added together to form the variable InnoSUM.

(27)

3.1.2 Intended adoption

The current study is testing the intended adoption of highly automated vehicles which qualify as levels 4 and 5 (SAE International, 2016). To assure that all respondents have the same definition and understanding of ‘self-driving cars’, an explanatory text was added before the question “How interested would you be in owning/using a completely self-driving vehicle?”.

To truly measure an interest or disinterest in adopting the self-driving car, a bipolar scale was used. In order to avoid redundancy in possible answer choices and assure meaningful differences between scale values, a 4-point scale was used. Additionally, this scale makes it faster to read and complete the questionnaire. Respondents were asked to answer according to the following format: (1) not interested at all, (2) slightly interested, (3) moderately interested and (4) very interested. The variable was named IntAdop.

3.1.3 Expected benefits

Based on the findings of numerous studies which were discussed in the literature review, the most prominent expected benefits of self-driving cars were selected to be tested: fewer crashes, improved emergency response to crashes, lower insurance rates, lower vehicle emissions, less traffic congestion, shorter travel time and increased human productivity.

Because a respondent can either expect or not expect a benefit to occur, a bipolar scale was used. Although it was considered to add an “I don’t know” option, this is likely to have put the respondent in a “fence sitter” position, particularly because it is a topic that many people might not know much about. Eliminating this ‘easy way out’ is believed to make the participants think for a few seconds longer about their answer choice, thereby engaging them as more active survey participants. If a person does not really know, they probably do not see much value in a particular benefit.

In order to avoid redundancy in possible answer choices and assure meaningful differences between scale values, a 4-point scale was used. Additionally, this scale makes it faster to read and complete the questionnaire. Respondents were asked to rate the likeliness of occurrence of each of the seven benefits according to the following format: (1) very unlikely, (2) somewhat unlikely, (3) somewhat likely and (4) very likely. The sum of all the expected benefits was named BenSUM.

(28)

3.1.4 Concerns

Based on the findings of numerous studies which were discussed in the literature review, the most prominent concerns about self-driving cars were selected to be tested: safety consequences of equipment failure or system failure, legal liability for drivers/owners, system security/hacking, data privacy (location and destination tracking), interacting with non-selfdriving vehicles and self-driving vehicles getting confused by unexpected situations.

Because a concern can be somewhere on a continuum, a unipolar scale was used. Similar to the expected benefits, it was considered to add an “I don’t know” option, but the option was rejected for the same reasons. In order to avoid redundancy in possible answer choices and assure meaningful differences between scale values, a 4-point scale was used. Additionally, this scale makes it faster to read and complete the questionnaire. Respondents were asked to rate their level of concern about the six different aspects of self-driving cars according to the following format: (1) not at all concerned, (2) slightly concerned, (3) moderately concerned and (4) very concerned. The sum of all the concerns was named ConSUM.

3.1.5 Knowledge

Respondents were asked to rate their familiarity based on the following four categories: “I have deep knowledge of the technology and the latest developments”, “I have basic knowledge of the technology and the latest developments”, “I have heard about the concept of self-driving cars, but do not know much about it” and “This survey is the first time I heard about it”. The highest level of knowledge was ranked with a four, and the lowest with a one. The variable was named Knowledge.

3.1.6. Moderator

In order to test the interaction effect of innovativeness on the relationship between knowledge on self-driving cars and interest to adopt, a few new variables had to be created. First, the variables knowledge and InnoSUM were centralized, creating two new variables. Next, these two variables were multiplied together, and the product of both was named Moderator.

3.1.6 Demographic variables

Age, gender and educational level were included in order to gain insights into the demographic distribution of the sample.

(29)

3.2 Methods

For the statistical analysis, the package program of SPSS (Statistical Package for the Social Sciences) version 20 was used.

3.2.1 One-way anova

The first two research questions are about the differences in expected benefits and concerns between the early adopters and the early majority. This will be done by comparing the means using One-way ANOVA’s.

The first research question looks into how the expected benefits of self-driving cars differ for the early adopters compared to the early majority. For the first research question, the independent variable that was selected is Chasm and the dependent list includes all the perceived benefits that were posed as well as the variable BenSUM.

The second research question looks into how concerns about self-driving cars differ for the early adopters compared to the early majority. For the second research question, the independent variable that was selected is Chasm and the dependent list includes all the concerns that were posed as well as the variable ConSUM.

Additionally, the variables were checked for homoscedasticity using Levene’s test. 3.2.2 Pearson correlation

The third research question is about the effect of knowledge about self-driving cars on the interest in adopting the technology. In order to answer the third research question, three hypotheses were formulated.

This relationship is tested by means of a Pearson correlation test. The variables Knowledge and BenSUM are correlated.

(30)

This relationship is also tested by means of a Pearson correlation test. The variables Knowledge and ConSUM are correlated.

3.2.3 Linear regression

First, a multicollinearity check was performed with the variables Knowledge, BenSUM, ConSUM and InnoSUM.

For testing the third hypothesis, a linear regression was used. Here the dependent variable was IntAdop, and the dependent list included Knowledge, InnoSUM and moderator.

4. Results

The following chapter presents the results of the research, organized in two sections: descriptives and analysis.

4.1 Descriptives

4.1.1 Demographic characteristics

As mentioned in the previous chapter, a total of 191 valid responses were collected. Hereof 82 (42.9%) were male and 109 (57.1%) were female. Additionally, the respondents were asked about the highest degree or level of education they have completed. Out of the total sample 16 (8.4%) completed secondary school, 39 (20.4%) vocational school, 74 (38.7%) bachelor, 59 (30.9%) a masters and 3 (1.6%) a PhD. Hence, there is an overrepresentation of people who have completed higher education. The ages of the respondents range from 17 to 74, with an average of 35.05, standard deviation of 13.155 and a median of 29. Additionally, there is an overrepresentation of the ages 24, 25, 26 and 27.

4.1.2 Adoption categories

Based on the innovativeness scale that was used, it was found that the scores ranged from 17 to 70. Additionally, the mean innovativeness score for the sample is 45.64 and the standard deviation is 10.458. Based on this information, the method of adopter categorization as suggested by Rogers (2013) was used to distribute the respondents into adoption categories. Out of the total sample, 6 (3.1%) were identified as innovators, 32 (16.8%) as early adopters, 50 (26.9%) as the

(31)

early majority, 78 (40.1%) as the late majority and 25 (13.1%) as laggards. Based on this categorization, two new variables were created: Chasm, which includes the (1) early adopters and the (2) early majority and the variable AdoGroups, which divides the five adopter categories. 4.1.3 Intended adoption and knowledge

Table 1 shows the distribution of the indicated level of interest in adopting self-driving cars. Additionally, the mean of the sample is 2.76, meaning that on average people indicate to be between slightly and moderately interested in using self-driving cars.

Table 1: Interest in adopting self-driving cars Frequency Percent Not at all interested 25 13.1 Slightly interested 48 25.1 Moderately interested 66 34.6 Very interested 52 27.2

Total 191 100.0

When asked about their knowledge on self-driving cars, the great majority of the people indicated that they have heard about the concept, but do not know much about it (Table 2). For merely seven people this survey was the first time they heard about self-driving cars. Additionally, the mean of the sample is 2.42. This indicates that an average person has either basic knowledge on the technology and the latest developments or has heard about the concept, but does not know much about it.

Table 2: Knowledge on self-driving cars

Frequency Percent I have deep knowledge of the technology and the latest developments 13 6.8 I have basic knowledge of the technology and the latest developments 62 32.5 I have heard about the concept of self-driving cars, but do not know much about it 109 57.1 This survey is the first time I heard about it 7 3.7

(32)

4.1.4 Expected benefits and concerns

Table 3 shows the expected benefits for each of the adopter groups. It can be noted that for each expected benefit, the mean in almost a perfectly gradual way goes from being highest for the innovators to getting lower towards the laggards. With fewer crashes, for example, it can be seen that the mean is a perfect 4. This means all innovators think it is very likely that there will be fewer crashes with the adoption of self-driving cars. The late majority, on the other hand, scores closes to a 3, meaning they think this benefit is somewhat likely to occur.

Table 3: Expected benefits per adopter category Innovators Early

Adopters

Early majority

Late

majority Laggards Total

N=6 N=32 N=50 N=78 N=25 N=191

Mean Mean Mean Mean Mean Mean

Fewer crashes 4 3.34 3.18 3.06 2.8 3.14

Improved emergency

response to crashes 3.67 3.47 3.16 3.08 2.88 3.16 Lower insurance rates 3.67 2.91 2.76 2.68 2.64 2.76 Lower vehicle

emissions 3.83 3.44 3.16 2.9 2.68 3.06

Less traffic congestion 3.67 3.47 3.04 3 2.88 3.09

Shorter travel time 3.83 3 2.84 2.82 2.88 2.9

Increased human

productivity 3.67 2.94 2.68 2.64 2.6 2.73

Total of expected

benefits 26.33 22.56 20.82 20.18 19.36 20.83

Additionally, it can be seen that overall, the benefits that are expected most are fewer crashes and improved emergency response to crashes. What is least expected is the increased human productivity and lower insurance rates.

Table 4 shows the concerns for each of the adopter groups. It can be noted that for each concern, the mean is the lowest for the innovators and almost in a perfectly gradual way goes higher towards the laggards. With the concern about self-driving cars interacting with non-selfdriving vehicles, for example, it can be seen that the mean for innovators is 1.17. This means the innovators are close to not at all concerned, whereas the laggards score a 3 and are moderately concerned about this issue.

(33)

Table 4: Concerns per adopter category _Innovators Early Adopters Early majority Late

majority Laggards Total N=6 N=32 N=50 N=78 N=25 N=191 Mean Mean Mean Mean Mean Mean Safety consequences of

equipment failure or system failure

1.83 2.69 2.82 2.96 3.2 2.87 Legal liability for

drivers/owners 2.33 2.34 2.56 2.86 3.24 2.73 System security/hacking 2.33 2.84 3 3.26 3.44 3.12

Data privacy 1.83 2.75 2.52 2.97 3.44 2.84

Interacting with

non-selfdriving vehicles 1.17 2.19 2.44 2.77 3 2.57 Self-driving vehicles getting

confused by unexpected situations

1.5 2.03 2.7 2.9 3.04 2.68

Total of concerns 11 14.84 16.04 17.72 19.36 16.8

Additionally, it can be seen that for the total of all groups together, the factors that people are most concerned about is the risk of security and a system being hacked. They are least concerned about self-driving vehicles getting confused by unexpected situations and the interaction with non-selfdriving vehicles.

(34)

4.2 Analyses

A One-way ANOVA was performed to compare the means for the early adopters and the early majority on each of the potential benefits that were posed. The results in table 5 suggest there are only significant differences in means for the benefit of less traffic congestion (F = 4.957, p = 0.029). Although no significant results were found for all other benefits, the total of expected benefits does significantly differ between the two groups (F = 4.811, p = 0.031).

Additionally a homoscedasticity check was performed using Levene’s test. It was found that the p-values were not significant for any of the concerns, except for ‘fewer crashes’ (p = 0.016). Although this might have a weak effect on the F-test, the effect is not expected to be strong because the groups are relatively similar in size.

Table 5: ANOVA results for expected benefits

Sum of Squares df Mean Square F Sig. Fewer crashes Between Groups

Within Groups Total .523 36.599 37.122 1 80 81 .523 .457 1.14 4 .288 Improved emergency response to crashes Between Groups Within Groups Total 1.860 42.689 44.549 1 80 81 1.860 .534 3.48 6 .066

Lower insurance rates Between Groups Within Groups Total .417 67.839 68.256 1 80 81 .417 .848 .492 .485

Lower vehicle emissions Between Groups Within Groups Total 1.503 48.595 50.098 1 80 81 1.503 .607 2.47 4 .120

Less traffic congestion Between Groups Within Groups Total 3.587 57.889 61.476 1 80 81 3.587 .724 4.95 7 .029

Shorter travel time Between Groups Within Groups Total .500 66.720 67.220 1 80 81 .500 .834 .599 .441 Increased human productivity Between Groups Within Groups Total 1.294 64.755 66.049 1 80 81 1.294 .809 1.59 8 .210 Total of expected benefits Between Groups Within Groups Total 59.245 985.255 1044.500 1 80 81 59.245 12.316 4.81 1 .031

Consumer adoption of self-driving cars : a diffusion of innovation perspective