The behavioural intention to adopt autonomous vehicles in the Dutch market

(1)

The Behavioural Intention to Adopt

Autonomous Vehicles in The Dutch

Market

Bram van Bakel: 10782338 Supervised by M. Etter

MSc Business Administration - Digital Business Track MSc Thesis (15EC)

Academic Year 2017-2018 22-06-2018

(2)

1

Abstract

Autonomous Vehicles (AVs) are coming to the market and it will take no more than a few years for it to happen. Several researchers state that the widespread adoption AVs can benefit society in multiple ways. There are several factors that affect the degree to which consumers have the intention to adopt AVs. It is important to determine what these factors are, so that the widespread adoption can be stimulated and the benefits can be reaped by society. This study tries to find out which factors are influencing the behavioural intention to adopt AVs of potential Dutch consumers. Prior studies show that there is little consensus on which factors do and do not influence the behavioural intention to adopt new products and the AV in particular. This research distributed a survey among 216 Dutch respondents and found

evidence to support the direct relationship between three factors and the behavioural intention to adopt AVs. The innate customer innovativeness was found to have a positive effect, while the effort expectancy as well as the risk perception were both found to have a negative effect. The direct positive relationship between the familiarity and behavioural intention was not supported. Also, the moderating effects of income and age on the positive relationship between innate customer innovativeness and the behavioural intention were not found. Likewise, no evidence was found to support the moderating effects of age and gender on the negative relationship between effort expectancy and the behavioural intention.

(3)

2

Statement of Originality

This document is written by Bram van Bakel 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.

(4)

3

List of Tables

Table 1. The Cronbach’s Alphas ... 30

Table 2. Exploratory Factor Analysis ... 31

Table 3. Descriptive Statistics ... 32

Table 4. Correlation Matrix ... 34

Table 5. Results of The Hierarchal Multiple Linear Regression ... 35

Table 6. The Variance Inflation Factors and Tolerance Values ... 42

Table 7. Overview of Hypotheses and Outcomes ... 44

List of Figures

Figure 1. Conceptual Framework for Direct and Moderating Effects ... 24

Figure 2. Scatterplot Standardized Predicted Value and Standardized Residual ... 39

Figure 3. Histogram of The Standardized Residuals ... 40

Figure 4. Probability Plot of The Standardized Residuals ... 41

Figure 5. Scatterplot Familiarity (F) and Behavioural Intention (BI) ... 62

Figure 6. Scatterplot Innate Customer Innovativeness (ICI) and Behavioural Intention (BI) . 62 Figure 7. Scatterplot Effort Expectancy (EE) and Behavioural Intention (BI) ... 63

(7)

6

Chapter 1. Introduction

1.1 Theoretical Background

The way society transports is about to be fundamentally changed in the upcoming years. Jen-Hsun Huang, the CEO of an American computer hardware manufacturer company named Nvidia, stated that they are working together with Audi to bring a level 4 Autonomous

Vehicle (AV) to the market in 2020 (Ross, 2017). The widespread adoption of this technology could potentially lead to disrupting benefits for society (Anderson, Nidhi, Stanley, Sorensen, Samaras and Oluwatola, 2014; Fagnant and Kockelman, 2015; Howard and Dai, 2014). Nevertheless, there might be several factors that are influencing the degree to which the AV technology will be adopted. Howard and Dai (2014) stated that the public opinion toward AVs is becoming more and more important as the public will shape the demand for the technology. This arises the question what potential consumers currently think of the AV and what their intentions are regarding the adoption of this technology. Which factors are

influencing this behavioural intention to adopt AVs?

Until now, several researchers have examined the potential benefits of the AV as well as the consumer’s intentions toward adopting the new innovative product. For example, Fagnant and Kockelman (2015) state that the widespread adoption of AVs will result in a lower number of crashes due to the elimination of errors made by human drivers. Furthermore, less congestion and fuel consumption will be the result of using AV technology, because this technology enables vehicles to be more efficient in their driving behaviour (Bullis, 2011; Fagnant and Kockelman, 2015; Tientrakool, Ho and Maxemchuk, 2011). Finally, the travel behaviour of consumers will improve through increasing mobility (Anderson et al., 2014) and stimulating car- and ridesharing (Fagnant and Kockelman, 2015).

(8)

7 Moreover, previous research on the adoption of new products, innovations, technologies and AVs has identified some factors that are influencing the behavioural intention to adopt AVs. Venkatesh, Morris, Davis and Davis (2003) developed the Unified Theory of Acceptance and Use of Technology (UTAUT), which combines multiple models and theories into one unified model to describe the relationship between several factors and the intention to use as well as the actual usage of new technologies. In addition to that, Arts, Frambach and Bijmolt (2011) have researched the adoption intention toward new innovations, while Choi and Ji (2015) investigated the behavioural intention to adopt AVs in particular. These researches and their findings together with several other papers will be discussed in the literature review.

Nevertheless, for some factors there is little consensus about their significance and certain combinations of factors are not tested. These factors will be tested in this thesis and consist of familiarity (Arts et al., 2011; Veryzer, 1998), innate customer innovativeness (Arts et al., rts1990), effort expectancy (Arts et al., 2011; Choi and Ji, 2015; Rogers, 2003; Taylor and Todd, 1995; Venkatesh et al., 2003) and risk perception (Choi and Ji, 2015; Martins, Oliveira, and Popovič, 2014; Veryzer, 1998). Furthermore, the moderating effects of income, age and gender will be tested. Also, prior studies show that research regarding the adoption of AVs has not been done extensively in the Dutch market. Therefore, the research question of this thesis is:

Which factors influence the behavioural intention of potential Dutch consumers to adopt Autonomous Vehicles?

(9)

8

1.2 Methodology

Accordingly, this thesis will develop and test a model to identify which factors are influencing the behavioural intention of potential Dutch consumers to adopt AVs. This model is based on the theories and findings of previous research. The data will be collected through a self-reported survey distributed among potential Dutch consumers that are 18 years old or older, since that is the age that Dutch consumers are able to acquire their full driver license. This thesis will use e-mail and social media channels to reach its respondents. After that, the data will be tested for reliability and validity to ensure that the data can be used for analysis. Finally, the hierarchical multiple linear regression method is used to test the hypothesized relationships and make conclusions regarding the research question.

1.3 Contribution

This thesis compares the results of prior studies and tests the factors for which contradictory relationships regarding adoption intentions have been found or for which the significance is found to be questionable. Therefore, the results of this thesis will contribute to literature by clarifying which factors are affecting the degree to which consumers have the intention to buy a new product and the direction and significance of this relationship. In addition to that, the outcome of this research will help practitioners designing, manufacturing and selling AVs to adapt their promotion strategy with the influential factors in mind to forecast, stimulate and maximize adoption. Furthermore, this thesis will expand literature by testing the relationships only for the Dutch market. Future research can build upon this work and compare the Dutch market with other markets to research any possible differences. Moreover, the results of this thesis will enable future studies to compare the relationships over time. The relationships found in prior studies can be compared to a more recent time period and possible changes over time can be identified.

(10)

9

1.4 Thesis Structure

The next chapter will explain what the AV is and what its potential benefits are. Furthermore, the literature regarding AVs and the behavioural intention to adopt new products, innovations and technologies will be reviewed. Also, several factors will be identified that could

potentially influence the behavioural intention to adopt AVs. The third chapter elaborates on how the data is collected and what its characteristics are. In addition to that, the methodology will be discussed, which will bring a better view on how the research is performed. The results will be analysed in the fourth chapter and discussed in chapter five. This will be followed by listing the managerial implications in chapter six. After that, the conclusions regarding the research question are summarized in chapter seven. Finally, the eight chapter consists of the thesis its limitations and the points of interest for future research.

(11)

10

Chapter 2. Literature Review

This chapter consists of the literature review and will discuss the theories regarding the behavioural intention to adopt new products, innovations, technologies and AVs, as well as the factors influencing this intention. First, a new innovative product based on a new technology, the AV, will be defined and explained, followed by its benefits. After that, the factors that are potentially influencing the behavioural intention will be considered and the related hypotheses are formulated. Also, several excluded factors will be briefly discussed. Finally, the conceptual framework will be presented.

2.1 The Autonomous Vehicle

The AV is a product that is currently in development and not yet available on the market. The National Highway Traffic Safety Administration (NHTSA) (2016) defines the Autonomous Vehicle (AV) as follows: “an automated vehicle system is a combination of hardware and software (both remote and on-board) that performs a driving function, with or without a human actively monitoring the driving environment” (p. 10).

Vehicles equipped with AV technology have the potential to fundamentally change

transportation and substantially affect safety, congestion, energy use and land use (Anderson et al., 2014; Howard and Dai, 2014). This is in line with Fagnant and Kockelman (2015), who, besides vehicle safety and congestion, also mention travel behaviour as a potential benefit. Both the type and magnitude of these potential benefits are dependent on the level of automation (Anderson et al., 2014). The Society of Automotive Engineers (SAE) classifies the AV technology into six levels of automation:

(12)

11 - Level 0 (No Automation): the driver is responsible for performing all driving tasks. - Level 1 (Driver Assistance): the driver is assisted in either acceleration or steering by

a driver assistance system and is responsible for performing all other driving tasks. Nevertheless, the driver must continue to be engaged with the driving tasks and the monitoring of the environment.

- Level 2 (Partial Automation): the driver is assisted in both acceleration and steering by one or more driver assistance systems, but is responsible for performing all other driving tasks. Nevertheless, the driver must continue to be engaged with the driving tasks and the monitoring of the environment.

- Level 3 (Conditional Automation): the automated driving system is able to perform most driving tasks. The driver should be able to take back control over the vehicle when requested, but it is not required to monitor the environment.

- Level 4 (High Automation): the automated driving system is able to perform most driving tasks under certain conditions, also when the driver does not take back control over the vehicle when requested. It is not required for the driver to monitor the

environment.

- Level 5 (Full Automation): the automated driving system is able to perform all driving tasks under all conditions. It is not required for the driver to monitor the environment.

This thesis researches the adoption intentions regarding “Autonomous” Vehicles, which in general means that the vehicle is able to drive autonomously and consequently without any human interaction. Therefore, this thesis will refer to the High Automation level (level 4) and the Full Automation level (level 5) as AV or AV technology, since these two levels are the most advanced levels of automation where the driver does not have to perform any of the

(13)

12 driving tasks nor has to monitor the environment. The other levels will be considered as partly automated or partly Autonomous Vehicles.

2.2 The Potential Benefits of The Autonomous Vehicle

Several researchers have found that the automation of vehicles can potentially benefit society by fundamentally changing the way society transports. First, AV technology could enhance safety. Fagnant and Kockelman (2015) argue that a significant part of crashes involves some kind of error made by the driver. Examples of these driver errors are drug and alcohol involvement, distraction, fatigue, speeding, aggressive driving, slow reaction times and inexperience. This is in line with the statements of Anderson et al. (2014), who mention that a large number of the total crashes are caused by driver errors. The AV would not make these human errors, simply because the execution of the driving tasks does not involve any human interaction. This suggests that the use of AV technology will reduce the number of crashes by eliminating the crashes that are caused by human errors.

The second benefit is related to congestion and fuel consumption. Researchers are developing the AV technology to be able to perceive and eventually maybe even predict the braking and acceleration decisions of the leading vehicle, which allows for smooth and precise braking and controlled speed adjustments of following vehicles. This in turn leads to fuel savings, less break wear and the reduction of shockwave propagation that destabilizes traffic (Fagnant and Kockelman, 2015). In addition to that, AVs enabled to make smart parking decisions do not have to search for parking spots, which will increase efficiency and fuel savings (Bullis, 2011). Also, the choice of more efficient routes and keeping shorter gaps between vehicles by AVs is expected to increase the efficient use of existing lanes and intersections (Fagnant and Kockelman, 2015). Tientrakool et al. (2011) even argue that decreasing the average safe

(14)

inter-13 vehicle distance through the use of sensors and vehicle-to-vehicle communication in AVs will increase the highway capacity. On the other hand, Anderson et al. (2014) argue that the effect of AV technology on congestion is uncertain, because a decreased cost of driving through the use of AVs might lead to an increase in the overall miles travelled by each vehicle, which in turn could increase congestion. Therefore, the effect of using AVs on congestion stays unclear and needs further investigation.

The change in travel behaviour is the third benefit. The AVs make it possible to increase mobility for people that are currently unable or unwilling to drive (Anderson et al., 2014; Fagnant and Kockelman, 2015; Litman, 2017). For example, people that are too old, too young, blind or disabled and therefore cannot drive, will be able to do so by using AVs. Also, AV technology could improve the sharing of cars and dynamic rides by allowing a person to rent a car in real-time based on per-minute or per-mile cost structures, like an on-demand taxi (Fagnant and Kockelman, 2015). Furthermore, Anderson et al. (2014) argue that due to the driver’s ability to engage in alternative activities while in the car, the willingness to travel longer distances to and from work might increase. This could cause people to locate

themselves further from the urban centre, which might lead to more dispersed and low-density patterns of land use around the metropolitan areas.

Finally, a more recent report provides an additional benefit. Litman (2017) writes that AVs can reduce stress and improve productivity, since the AVs will be designed as bedrooms or offices (NYT, 2017). Therefore, consumers will be able to use their time effectively while traveling. For example, consumers could go resting or working while being in their car, which will increase the productivity of the consumer.

(15)

14

2.3 The Behavioural Intention to Adopt

The literature shows that there are some interesting benefits that result from the use of AVs. These benefits can only be reaped if the AV is actually adopted on a widespread scale. Since this product is still in development, nothing can be researched about the actual adoption. What can be researched is the behavioural intention to adopt AVs.

First, a clear definition for the behavioural intention to adopt AVs must be determined. In the last few decades, prior studies have allocated different names and definitions to the

willingness of consumers to buy or use a new product or innovation. For example, Venkatesh et al. (2003) name it the “behavioural intention to use”. For this thesis, the definition of

Venkatesh et al. (2003) will be slightly adjusted to “the behavioural intention to adopt”, which represents the degree to which the consumer intents to buy or use the AV. The behavioural intention to adopt must be distinguished from the actual adoption behaviour. Im et al. (2003) name this phenomenon “new-product adoption behaviour” while Arts et al. (2011) use the term “consumer innovation adoption”. All of these definitions describe the degree to which a consumer actually buys or uses a certain product.

It is a common approach to use the self-reported behavioural intention to adopt a new

technology for predicting if people will actually adopt this new technology (Van Ittersum and Feinberg, 2010). This relationship has also been tested repeatedly. For example, Venkatesh et al. (2003) found a significant positive relationship between the behavioural intention and the adoption behaviour. Nevertheless, this relationship is out of scope for this thesis and will not be tested, since the AV is not yet on the market at the time of writing this thesis and

measuring the usage or actual adoption behaviour is therefore practically impossible. Thus, the factors that in prior studies are found to have a relationship with the actual adoption

(16)

15 behaviour are only tested for the behavioural intention to adopt AVs. However, research about the actual usage or adoption of AVs are interesting topics for future studies.

2.4 Factors Potentially Influencing The Behavioural Intention to Adopt

As mentioned earlier, the AV could bring some significant benefits to society, but what determines if consumers would even want to adopt this product? In the last two decades, several researchers have examined the way in which consumers adopt or have the intention to adopt new products, innovations and technologies (e.g., Choi and Ji, 2015; Im et al., 2003; Veryzer, 1998; etc.). In addition to that, prior studies have examined the public opinion on AV technology through surveys (e.g., Kyriakidis, Happee and de Winter, 2015; Howard and Dai, 2014). The literature regarding the public opinion on AVs as well as the adoption

intentions and actual adoption of new products, innovations and technologies was reviewed to determine relevant factors which could potentially influence the behavioural intention to adopt AVs. The results suggest that there are several factors that are directly influencing the consumer’s intention to adopt new products, innovations, technologies and the AV. These factors consist of familiarity, innate customer innovativeness, effort expectancy and risk perception. Also, the moderating effects of income, age and gender are discussed in prior studies. Furthermore, several factors that were found to be influencing the behavioural intention are excluded from this research. The directly related factors as well as the moderating and excluded factors will be further elaborated below.

2.4.1 Familiarity

The first factor that will be discussed is the familiarity factor. The familiarity factor is defined by the number of related experiences acquired by the consumer, where product-related experiences include a broad range of experiences (Alba and Hutchinson, 1987). In

(17)

16 addition to that, familiarity is also defined as what consumers perceive to know or what

consumers actually have stored in memory about a certain product (Rao and Monroe, 1988). It can be argued that the familiarity regarding product knowledge is included in the broad range of product-related experiences, since acquiring information about a certain product is also an experience related to that product.

Veryzer (1998) researched which key factors affect the customer evaluation of discontinuous new products, where discontinuous products are products that “involve advanced capabilities that do not exist in current products and cannot be achieved through mere extension of an existing technology” (p. 137). Veryzer (1998) found that the lack of familiarity with the new product is one of the key factors affecting the adoption of this new product. In most cases where this lack of familiarity existed, the result was that consumers were more likely to resist the product. Also, the product familiarity seemed to influence the speed in which the

consumer adopted the product. The effect of familiarity on new product adoption is confirmed by Li and Lin (2015), who found that when a consumer knows and understands the product well, it will positively influence the adoption of this product. However, the findings of Veryzer (1998) and Li and Lin (2015) are related to the actual adoption of new products. On the other hand, Arts et al. (2011) studied the adoption intention and found that consumers are more likely to have adoption intentions when the familiarity with a product is higher. This thesis will research the positive effect of familiarity on the behavioural intention to adopt. Therefore, the hypothesis regarding product familiarity is:

H1: The Familiarity with Autonomous Vehicles has a positive relationship with the Behavioural Intention to Adopt Autonomous Vehicles.

(18)

17 At the moment of writing this thesis, the AV is not yet on the market, which would imply that the familiarity regarding this product is expected to be relatively low. Kyriakidis et al. (2015) studied the public opinion on AVs among 5000 respondents and asked in which year they expected AVs to drive on public roads. Kyriaskidis et al. (2015) found a median of 2030. Nevertheless, as mentioned earlier, Ross (2017) wrote that the AVs could already be brought to market in 2020. The difference in expectancy between manufacturers and consumers might be a sign of low familiarity with AVs among consumers. This low familiarity in turn is

expected to reduce the behavioural intention to adopt a new product according to theory. In sum, it is expected that the average familiarity of the sample will be relatively low.

2.4.2 Innate Customer Innovativeness

The second factor is the customer’s innate innovativeness. Innate customer innovativeness is defined by Im et al. (2003) as “a generalized unobservable predisposition toward innovations applicable across product classes” (p. 62). Im et al. (2003) investigated the relationship between innate customer innovativeness, personal characteristics and new product adoption behaviour. They found a weak direct relationship between innate customer innovativeness and new product adoption behaviour. Nevertheless, they also found that income and age in

combination with innate customer innovativeness influence the adoption of new products. Im et al. (2003) found that people who have a relatively high income, are young and have

innovative predispositions were more likely to adopt a higher number of new products. Furthermore, Im et al. (2007) found that innate customer innovativeness does not have a direct influence on new product adoption, but only indirectly through the modelling of other’s behaviours and engagement in word-of-mouth.

(19)

18 Both Im et al. (2003) and Im et al. (2007) did not find a strong direct relationship. This

contradictory with the findings of Arts et al. (2011), who did find a direct and positive influence of innate customer innovativeness on the adoption intention and behaviour of the respondents. In addition to that, Venkatraman and Price (1990) researched the same

relationship for three products and even found a negative relationship for one of the products. This implies that there is little consensus about the significance and direction of the

relationship between innate customer innovativeness and the adoption intentions toward new products. Nevertheless, the negative relationship found by Venkatraman and Price (1990) could be caused by the product they have chosen in their research. They write that the product for which they found the negative relationship might be perceived as less new than the other products that they have used in their research.

This thesis follows Arts et al. (2011) and Im et al. (2003) and tests the positive relationship between innate customer innovativeness and the behavioural intention to adopt, including the moderating effects of income and age. Therefore, the hypotheses regarding innate customer innovativeness, income and age are:

H2a: Innate Customer Innovativeness has a positive relationship with the Behavioural Intention to Adopt Autonomous Vehicles.

H2b: The positive relationship between Innate Customer Innovativeness and the Behavioural Intention to Adopt Autonomous Vehicles is moderated by Income, such

that this relationship is stronger for higher values of Income.

H2c: The positive relationship between Innate Customer Innovativeness and the Behavioural Intention to Adopt Autonomous Vehicles is moderated by Age, such that

(20)

19 According to the European Commissions’ Innovation Scoreboard (2017) , the Netherlands holds the fourth spot in the ranking of European countries that are most innovative, which makes the Netherlands one of the innovation leaders of Europe. Moreover, the Netherlands is one the fastest growing innovators with a performance increase of 10.4% from 2016 to 2017. This could imply that the Dutch people are more innate innovative than other countries, which according to the theory is expected to positively influence the behavioural intention to adopt AVs.

2.4.3 Effort Expectancy

The third factor is the effort expectancy factor, which is defined by prior studies using different terms for the same definition. Two alternative terms used for effort expectancy are perceived complexity (Rogers, 2003; Taylor and Todd, 1995; Arts et al., 2011) and ease of use (Davis, 1989; Choi and Ji, 2015). The latter one is a counter-indicative definition of effort expectancy and perceived complexity, which means that higher levels of the ease of use correspond to lower levels of effort expectancy and perceived complexity. In prior research, these definitions are used interchangeably. For this thesis, the effort expectancy from the Unified Theory of Acceptance and Use of Technology (UTAUT) will be used, which is defined by Venkatesh et al. (2003). Venkatesh et al. (2003) define effort expectancy as “the degree of ease associated with the use of the system” (p. 450). They found that the effort expectancy is influencing the behavioural intention to adopt new technologies and that this effect is moderated by age, gender and experience. The experience moderator is out of scope for this thesis, since the AV is not yet on the market and experience with the AV is therefore unlikely. Hoque and Sorwar (2017) tested the UTAUT model for the mobile health services and found a positive relationship between effort expectancy and the behavioural intention to

(21)

20 adopt. However, in this thesis the items were reverse coded, therefore the findings of Hoque and Sorwar (2017) are consistent with hypothesis 3a.

Other studies also researched the relationship between effort expectancy (i.e., also perceived complexity and ease of use) and the behavioural intention to adopt. Rogers (2003) argues that when a consumer perceives an innovation as more complex, it will negatively affect the adoption in the sense that the customer will less likely adopt the innovation. This is in line with Taylor and Todd (1995), who state that complexity is expected to negatively affect the attitude toward adoption, which in turn negatively influences the behavioural intention to adopt. Choi and Ji (2015) find a positive relationship between the ease of use and the behavioural intention, which is consistent with the above stated findings since this term is counter-indicative. On the other hand, Arts et al. (2011) found that complexity has a positive effect on the adoption intention, while consumers were more likely to actually adopt an innovation when they perceived the innovation as less complex. They write that in the intention stage, before adoption, complexity might be underestimated and usability of the innovation might be overestimated by the consumer, which explains the positive effect of complexity on the intention to adopt. The reason that the results of Choi and Ji (2015) and Taylor and Todd (1995) differ from the findings of Arts et al. (2011) could be that a different research design was used. Choi and Ji (2015) and Taylor and Todd (1995) both used a survey design to collect the data and make conclusions regarding this data. On the other hand, Arts et al. (2011) performed a literature study where they collected many different texts regarding the behavioural intention to adopt and the actual adoption and made conclusions based on the collective of these papers. Another reason could be the difference in the time period.

(22)

21 The difference in the results show that there is little consensus about the significance and the direction regarding the effect of effort expectancy on the behavioural intention to adopt. This thesis will follow Choi and Ji (2015), Rogers (2003) and Taylor and Todd (1995) and tests the negative relationship between effort expectancy and the behavioural intention to adopt. Also, the moderating effects of age and gender stated by Venkatesh et al. (2003) will be tested. Therefore, the hypotheses regarding effort expectancy, age and gender are:

H3a: Effort Expectancy has a negative relationship with the Behavioural Intention to Adopt Autonomous Vehicles.

H3b: The negative relationship between Effort Expectancy and the Behavioural Intention to Adopt Autonomous Vehicles is moderated by Age, such that this

relationship is stronger for lower values of Age.

H3c: The negative relationship between Effort Expectancy and the Behavioural Intention to Adopt Autonomous Vehicles is moderated by Gender, such that this

relationship is stronger for women.

Since the AV is a high-technology innovation, consumers might expect having to exert a relatively high level of effort when using this product. According to theory, this is expected to negatively affect the behavioural intention to adopt.

2.4.4 Risk Perception

Finally, the last factor that will be researched in this thesis is risk perception. Risk perception or perceived risk is described as the degree of uncertainty related to the potential negative consequences when using or buying a product (Featherman and Pavlou, 2003). Peter and Ryan (1976) state a more formal description of risk perception, which is “the expectation of

(23)

22 losses associated with purchase and, as such, acts as an inhibitor to purchase” (p. 185). This latter description implies that risk perception has a negative effect on the willingness or intention to purchase a product.

The relationship between risk perception and the behavioural intention to adopt a new product has been researched by several prior studies. Martins et al. (2014) added risk perception to the UTAUT model of Venkatesh et al. (2003) and found that this increased the predictive power of the UTAUT model in explaining the behavioural intention. Martins et al. (2014) found a negative and significant relationship between risk perception and the behavioural intention. This same relationship was found by Veryzer (1998), who researched the risk and uncertainty associated to new discontinuous products. He found that resistance was encouraged when consumers were uncertain about the risks related to a certain new discontinuous product. On the other hand, Choi and Ji (2015) also researched the negative relationship between risk perception and the behavioural intention, but they could not find any evidence to support the significance of the relationship. However, Choi and Ji (2015) mention in their limitations that the results might be biased toward younger men’s opinions, since their sample consisted of more young people and more men. It might be that younger men are less sensitive to risk, which could explain that Choi and Ji (2015) did not find a significant relationship.

Thus, also on this factor it seems like there is not a lot of consensus about the significance of its influence on the behavioural intention. This thesis will follow prior studies and also tests the negative relationship between risk perception and the behavioural intention to adopt. Therefore, the hypothesis regarding risk perception is:

H4: Risk Perception has a negative relationship with the Behavioural Intention to Adopt Autonomous Vehicles.

(24)

23 One phenomenon that could play a role in the risk perception of a consumer regarding AVs is called the social dilemma. This phenomenon is identified by Bonnefon, Shariff and Rahwan (2016) and occurs when the AV has to choose between two worst-case scenarios, such as running over a pedestrian or sacrificing itself and the driver to save this pedestrian. They find that people would like others to buy AVs that sacrifice the driver for the greater good, but would themselves prefer to buy an AV that saves the driver at all cost. This creates a social dilemma and could result in a higher risk and uncertainty associated to the AV, which in turn is expected to have a negative effect on the behavioural intention to adopt AVs. Moreover, NOS (2018) reported the first deadly accident involving an AV while this thesis was in development. An AV from Uber that was doing a test drive hit a women who randomly crossed the street while the crosswalk was near. Later on, she died in the hospital because of her injuries. These kinds of events might influence the degree to which consumers perceive AVs as risky, which according to theory is expected to have a negative effect on the

behavioural intention to adopt AVs. However, the effect of the social dilemma and events like the deadly accident on the risk that consumers perceive associated with AVs will not be researched in this thesis. This is because this thesis only researches the direct and moderating effects of several factors on the behavioural intention to adopt AVs. Therefore, the above mentioned effects on risk perception are out of scope for this thesis. Nevertheless, this can be done in future studies.

2.4.5 Excluded Factors

The four above mentioned factors and the related moderators were not the only influencing factors found by prior studies. However, not all of these factors were fitting this research. For example, user-product interaction problems (Veryzer, 1998) was one of the factors that would be interesting to test. However, this factor measures the degree to which consumers

(25)

24 experience problems by interacting with a new product. Nevertheless, the AV is not yet on the market and it is therefore very unlikely that consumers will have had any experiences

regarding the interaction with AVs. Therefore, this factor will be hard to measure and does not really fit the design of this study. Other factors include system’s performance expectancy or perceived usefulness (Choi and Ji, 2015; Venkatesh et al., 2003), transparency (Choi and Ji, 2015), facilitating conditions (Venkatesh et al., 2003), compatibility (Arts et al., 2011) and more. However, considering the scope of this thesis and the size of the model these factors were excluded. The included factors were chosen based on the relevance, level of consensus in the literature and the measurability. The excluded factors can be tested in future research.

2.5 The Conceptual Framework

The results of prior studies show an inconsistency in the significance and/or direction of the factors listed above. Although they could be explained by differences in the research design and samples as mentioned before, it is still interesting to look at these relationships. The research conducted in this thesis aims to reduce the inconsistency and bring more clarity to the literature. This thesis will research if and how the earlier stated factors are influencing the behavioural intention to adopt AVs through testing all eight hypotheses in the following model presented in figure 1.

(26)

25

Chapter 3. Data and Methodology

This chapter will elaborate on how the research is performed to give clear insights regarding the research process. First, the development of the measurements will be presented. Second, the process of collecting the data is discussed and some characteristics of the dataset are given. Finally, the chosen method to analyse the data is described.

3.1 Measurement Scales

This study measured five main constructs and three constructs which are related to the

demographics. The five main constructs are familiarity, innate customer innovativeness, effort expectancy, risk perception and behavioural intention. All of these constructs consist of three items. The larger part of these items are taken from existing literature, while for some

constructs items were developed for this paper. The three remaining constructs are income, age and gender, which are measured through collecting the demographics regarding each respondent.

To measure familiarity, one item is taken from Rao and Sieben (1992) and two other items are developed and implemented for this thesis. The innate customer innovativeness was measured by three items taken from Goldsmith and Hofacker (1991), who originally developed eleven items. The risk perception was measured by using two items of Choi and Ji (2015), while one item was developed and implemented for this thesis. The effort expectancy and behavioural intention are both measured using three adapted items from the ten item list developed by Davis (1989). All five main items were measured using a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). In correspondence with other studies (e.g., Venkatesh et al., 2003), the responses of the three items are summed up and divided by three to get the final average score per construct. The results of the items regarding effort expectancy were reverse

(27)

26 coded to ensure proper interpretation, since these items are counter-indicative. The

demographics were measured as follows. The income variable was measured using an 11-point scale with intervals of €10.000 (1 = less than €10.000, 11 = more than 100.000). The gender variable was measured as a binominal variable (0 = Male, 1 = Female). To measure age, the respondents had to select their age from a dropdown menu. All the items that were used in the survey can be found in Appendix B.

3.2 Data Collection and Description

This study is based on the collection of cross-sectional data, since the time period of the thesis only consists of six months. Hence, collecting time series data to find changes over time will probably not result in finding any significant increases or decreases in the data due to the short time period. To collect the cross-sectional data, a self-reported survey was used. This research design is chosen, because it is common in the field of this research (e.g., Choi and Ji, 2015; Venkatesh et al., 2003) and using a survey provides a practical and convenient way to reach a high number of respondents. The non-probability method named convenience

sampling is used to collect the data, which in practice means that the sample was taken from a network that was relatively easy to access. The survey was distributed online through the use of social media and e-mail, which in particular means that the survey was posted on Facebook and sent via WhatsApp and e-mail. The network of the author and the network’s network was used to exponentially distribute the survey. The population of this study consists of the potential Dutch consumers of AVs and therefore people that are not able to get their full driver license (i.e., people that are younger than 18 years old) are excluded from the sample. Furthermore, the survey was conducted in Dutch, since the population only consists of Dutch consumers. The English and Dutch version of the survey can be found in Appendix B and C

(28)

27 respectively. The time period in which the survey was distributed runs from the 9th of April until 19th of April in 2018.

The data of 251 respondents was collected, however the response rate is unknown since the survey was distributed online, which caused the number of reached people to be unknown. Out of the 251 responses, 35 contained missing data or were not filled in at all. The cases that contained missing data were studied and it was concluded that the data was missing at

random. Therefore, the cases with missing data and the ones that were not filled in at all were eliminated from the sample which reduced the dataset to 216 respondents. Among these 216 respondents, 44.9% were male and 55.1% were female. The age ranged between 18 and 79 year old and the average age was approximately 31.

3.3 Data Analysis

First, the data of each variable was tested for normality and outliers. The normality of the variables was assessed by studying the histogram and probability plot, analysing the skewness and kurtosis and finally by performing a Kolmogorov-Smirnov test. The standardised scores of the variables were used to determine the presence of outliers.

Second, the reliability and validity of the constructs based on the Likert scale were tested before analysing the data. The Cronbach’s Alpha (Cronbach, 1951) was used to determine the reliability and the Principal Axis Factoring (PAF) analysis was executed to conclude the validity (Costello and Osborne, 2005).

Third, the hierarchical multiple linear regression analysis (Cohen, Cohen, West and Aiken, 2013) was used to analyse the data, since the model that is tested consists of multiple

(29)

28 independent variables. The hierarchical multiple linear regression was performed in three steps. The first regression included only the independent variables that are hypothesized to have a direct relationship with the behavioural intention. Therefore, the equation looks as follows:

𝐵𝐼 = 𝛽₀+ 𝛽₁∗ 𝐹 + 𝛽₂∗ 𝐼𝐶𝐼 + 𝛽₃∗ 𝐸𝐸 + 𝛽₄∗ 𝑅𝑃 + 𝜀_𝑖 (1)

Where, BI is behavioural intention, F is familiarity, ICI is innate customer innovativeness, EE is effort expectancy and RP is risk perception. The second regression adds the demographics to this equation to estimate their direct effects. This changes the equation and makes it look like the following:

𝐵𝐼 = 𝛽₀+ 𝛽₁∗ 𝐹 + 𝛽₂∗ 𝐼𝐶𝐼 + 𝛽₃∗ 𝐸𝐸 + 𝛽₄∗ 𝑅𝑃 + 𝛽₅∗ 𝐼𝑁𝐶 + 𝛽₆ ∗ 𝐴𝐺𝐸 + 𝛽₇∗

𝐺𝑁𝐷 + 𝜀_𝑖 (2)

Where, INC is income, AGE is age and GND is gender. The third and last regression adds the interaction terms of income times innate customer innovativeness, age times innate customer innovativeness, age times effort expectancy and gender times effort expectancy. This will enable this study to determine if there are any moderating effects present. The equation gets expanded to the following:

𝐵𝐼 = 𝛽₀+ 𝛽₁∗ 𝐹 + 𝛽₂∗ 𝐼𝐶𝐼 + 𝛽₃∗ 𝐸𝐸 + 𝛽₄∗ 𝑅𝑃 + 𝛽₅∗ 𝐼𝑁𝐶 + 𝛽₆ ∗ 𝐴𝐺𝐸 + 𝛽₇∗ 𝐺𝑁𝐷 + 𝛽₅∗ 𝐼𝑁𝐶 ∗ 𝐼𝐶𝐼 + 𝛽₆∗ 𝐴𝐺𝐸 ∗ 𝐼𝐶𝐼 + 𝛽₆∗ 𝐴𝐺𝐸 ∗ 𝐸𝐸 + 𝛽₇∗ 𝐺𝑁𝐷 ∗ 𝐸𝐸 + 𝜀_𝑖 (3)

Finally, the assumptions of the hierarchical multiple linear regression model were tested to conclude if the results are correctly interpreted. The results regarding the normality, outliers, reliability, validity, hierarchical multiple linear regression and its assumptions will be

(30)

29

Chapter 4. Results

This chapter discusses the results from the analysis performed. First, the normality and the outliers considering the variables will be discussed. Second, the results of the reliability and validity tests are presented. Third, the descriptive statistics regarding the variables will be analysed. Fourth, the correlations between the independent variables and the dependent variable are discussed using the correlation matrix. Fifth, the outcomes of the hierarchical multiple linear regression analysis are considered and the hypotheses are evaluated. Finally, the assumptions of the hierarchical multiple linear regression are tested.

4.1 Normality and Outliers

The variables were tested for normality and outliers. After using a variety of techniques mentioned before to analyse the normality of the variables, it was concluded that none of the variables were normally distributed. The histograms and probability plots of several variables indicated normality. However, the tests regarding skewness and kurtosis indicated that most of the variables did not meet the requirements to be considered normally distributed. Also, the Kolmogorov-Smirnov test rejected the hypothesis of the data being normally distributed for every variable. Nevertheless, normality of the variables is not an assumption of the

hierarchical multiple linear regression analysis and hence the data is not changed.

The analysis of the standardised scores of the variables resulted in the finding of six outliers. Five outliers were found in the data related to the income variable and the remaining outlier was related to the age variable. All outliers were studied and the conclusion was made that due to the size of the sample these outlaying values are not unusual. Therefore, the outliers are kept in the sample.

(31)

30

4.2 Reliability and Validity of the Measurement Scales

The Likert scale based constructs used in this study were tested for reliability and validity to determine if these instruments are consistently measuring the constructs which were intended to be measured. This is an important step in the process since some of the items were

developed and implemented for this thesis.

The reliability is tested by using the Cronbach’s Alpha (Cronbach, 1951), which for each construct is presented below in table 1. The table shows that all the constructs have a

sufficiently high Cronbach’s Alpha (i.e., higher than 0.70). Also, the reliability test indicated that for every construct the items within that construct showed a good correlation with the total score of the scale (i.e., the corrected item-total correlation was above 0.30). In addition to that, removing any of the items within a certain construct would not result in a significant improvement of the Cronbach’s Alpha of that construct. Thus, based on these three results it is concluded that the constructs are reliable.

Table 1. The Cronbach’s Alphas

Construct Cronbach's Alpha

Familiarity 0.857

Innate Customer Innovativeness 0.869

Effort Expectancy 0.872

Risk Perception 0.898

Behavioural Intention 0.951

The Principal Axis Factoring (PAF) analysis (Costello and Osborne, 2005) was used to test the validity of the constructs. The Kaiser-Meyer-Olkin (KMO) and Bartlett’s test are used to test if the assumptions of the PAF analysis are met. The KMO measure of sampling adequacy was 0.826 and the Bartlett’s test of sphericity χ2(105) = 2283.585, 𝑝 < 0.001. The

(32)

31 met. After that, the eigenvalues for every component in the data was obtained and five of them showed an eigenvalue which was bigger than Kaiser’s criterion of 1. These five

components together explained 74.61% of the variance. The factor loadings after rotation are presented below in table 2. The table shows that the items of each construct are clustering around only one factor, which suggests that the constructs are valid.

Table 2. Exploratory Factor Analysis

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Scale Items (Familiarity) (Innate Customer Innovativeness) (Effort Expectancy) (Risk Perception) (Behavioural Intention) Familiarity F1 -0,86 -0,02 -0,04 0,00 -0,06 F2 -0,90 0,04 0,04 0,06 0,04 F3 -0,68 0,00 -0,02 -0,03 0,06 Innate Customer Innovativeness ICI1 -0,03 0,81 0,10 0,10 -0,03 ICI2 0,01 0,86 -0,08 -0,06 0,02 ICI3 0,01 0,81 -0,06 -0,05 0,03 Effort Expectancy EE1 0,08 -0,03 0,75 0,04 0,03 EE2 -0,02 0,00 0,90 -0,07 0,03 EE3 -0,02 0,01 0,80 -0,03 -0,10 Risk Perception RP1 0,05 0,00 0,05 -0,81 0,00 RP2 0,07 -0,02 -0,04 -0,96 0,02 RP3 -0,09 0,01 0,04 -0,80 -0,06 Behavioural Intention BI1 0,01 0,05 0,01 0,01 0,93 BI2 -0,04 -0,03 -0,02 0,01 0,91 BI3 0,00 0,00 0,00 0,01 0,93

(33)

32

4.3 Descriptive Statistics

The descriptive statistics provide an overview of several important characteristics of the variables used in the research. This subchapter will be used to interpret and discuss the findings resulting from these statistics. The descriptive statistics are presented in table 3.

Table 3. Descriptive Statistics

Statistic Variable Count Mean Median

Standard

Deviation Minimum Maximum

Familiarity 216 3,64 3,67 1,56 1 7

Effort Expectancy 216 3,00 3,00 1,34 1 7

Innate Customer Innovativeness 216 2,74 2,33 1,53 1 7

Risk Perception 216 4,33 4,67 1,50 1 7

Behavioural Intention 216 4,16 4,33 1,74 1 7

Gender 216 0,55 1,00 0,50 0 1

Income 216 2,63 2,00 2,26 1 11

Age 216 30,84 23,00 15,62 18 79

The count represents the number of observations made for each variable. The number of observations is equal for every variable since the results of respondents containing missing data were eliminated from the sample. If the means and medians are analysed, it can be seen that the means of familiarity, risk perception and behavioural intention are fairly in the middle, since the Likert scale goes from 1 to 7. This suggests that on average the respondents are neutral considering the before mentioned variables. However, the means of effort

expectancy and innate customer innovativeness seem to be more on the lower end of the Likert scale. This suggests that on average the respondents are not very innovative in nature and do not expect to exert a lot of effort when using the AV. The mean of the gender variable shows that the sample consisted of just a little bit more females compared to males. Males were coded with a zero and females with a one. Therefore, the mean tells us that the sample contains 55% females and consequently 45% males. For the income and age variable, it is more useful to take a look into the medians, since both these variables had one or more

(34)

33 outliers. The median of income is 2.00, which equals to a yearly income of €10.000 - €19.999. Furthermore, the mean of income is 2.63 and will therefore also give the same result. The Central Bureau of Statistics (2018), which is an authority that collects and announces statistics relevant for the Dutch population, reports that the average income of the Dutch population in 2016 was €30.900. This number is significantly higher than the results found in the sample of this research. One of the reasons that this occurs might be because the sample is quiet young and therefore probably does not earn that much income yet. Namely, the mean age is 30.84 and the median age is 23.00. Moving on, the standard deviations of all variables, except the age variable, are not too big nor too small. The age variable has a standard deviation of 15.62, which might be the result of the outlier or the spread of the sample. The minimum and

maximum of each variable are a result of the scales used to measure each variable. The age variable however has a minimum of 18 since younger respondents were not included in the sample and a maximum of 79, which is also the outlier.

These results give a clear picture of the basic features of the sample and the data seems to be a reliable source to conduct the research upon. However, it is not possible to make conclusions that are only based on the descriptive statistics. Therefore, the thesis will continue by

analysing the correlation matrix and hierarchal multiple linear regression.

4.4 Correlation Matrix

The Pearson correlation matrix is presented in table 4. The correlation matrix provides information about how the variables correlate, how strong this correlation is and in what direction the correlation is moving. The interpretation of the correlation matrix will be discussed below. Also, the results of the correlation matrix are compared to the hypotheses that were formulized in chapter 2.4.

(35)

34 Table 4. Correlation Matrix

Variable 1 2 3 4 5 6 7 8

1. Behavioural Intention -

2. Familiarity 0.32** -

3. Innate Customer Innovativeness 0.27** 0.35** -

4. Effort Expectancy -0.50** -0.41** -0.28** -

5. Risk Perception -0.46** -0.23** -0.10 0.29** -

6. Income -0.20** -0.04 -0.11 0.04 -0.01 -

7. Age -0.31** -0.23** -0.35** 0.30** 0.03 0.53 -

8. Gender -0.15* -0.27** -0.20** 0.30** 0.27** -0.19 0.05 - ** Correlation is significant at the 0.01 level (2-tailed).

* Correlation is significant at the 0.05 level (2-tailed).

The first column represents the variable names and the second column represents the

correlations with the dependent variable, namely the behavioural intention to adopt AVs. The matrix shows that all variables have a significant correlation with the behavioural intention at the 1% level, except for gender, which is significant at the 5% level. Furthermore, the

independent and moderating variables do not show any strong correlations between

themselves. Therefore, for this research multicollinearity will not be a problem, which will be discussed in more detail in subchapter 4.5.3.

The familiarity (0.32) and innate customer innovativeness (0.27) seem to have a weak positive correlation with the dependent variable. This is in line with hypothesis 1 stating a positive relationship between familiarity and behavioural intention. Also, hypothesis 2a which states the positive relationship between innate customer innovativeness and behavioural intention is in line with this finding. When the correlations of effort expectancy (-0.50) and risk

perception (-0.46) are studied, it can be seen that both of these factor have a moderate negative correlation with the behavioural intention. Also, these findings are consistent with hypothesis 3a and 4 that expect a negative relationship between both these factors and the behavioural intention.

(36)

35 The demographics age (-0.20), income (-0.31) and gender (-0.15) also have a significant correlation with the dependent variable. However, it is not possible to conclude anything about the remaining hypotheses that describe the moderating effects of the demographics. Thus, the next step is to analyse the results of the hierarchical multiple linear regression.

4.4 Hierarchical Multiple Linear Regression

The hierarchical multiple linear regression was performed in three steps. In the first step, the familiarity, innate customer innovativeness, effort expectancy and risk perception are

regressed on the behavioural intention to research their direct effects. The second step adds income, age and gender to this model to determine the direct effects of the demographics. The last step includes the interaction terms to test for the effects of the moderating variables. The results are presented in table 5. The table shows the unstandardized beta coefficients, where significant coefficients appear in bold. Also, the R-squared and the change in R-squared are included. The interpretation of the results are discussed below, after which these results will be compared to the hypotheses stated in subchapter 2.4.

Table 5. Results of The Hierarchal Multiple Linear Regression

Variables Regression 1 Regression 2 Regression 3

Familiarity 0.06 0.06 0.07

Innate Customer Innovativeness 0.14* 0.10 0.14

Effort Expectancy -0.45*** -0.43*** -0.22

Risk Perception -0.38*** -0.41*** -0.40***

Income -0.08 -0.20*

Age -0.01 0.03

Gender (1, female; 0, male) 0.22 0.01

Income x Innate Customer Innovativeness 0.03

Age x Innate Customer Innovativeness 0.00

Age x Effort Expectancy -0.01

Gender x Effort Expectancy 0.07

R² 0.38 0.41 0.43

R² Change 0.38*** 0.04** 0.01

(37)

36 The first model is statistically significant F (4, 211) = 31.85; p < 0.001 and explains 37.7% of the variance in the behavioural intention. This model will be used to determine the direct effects of the independent variables (i.e., familiarity, innate customer innovativeness, effort expectancy and risk perception) on the dependent variable (i.e., behavioural intention). The first hypothesis states that familiarity is positively related to the behavioural intention. The regression shows that familiarity is not significantly related to the behavioural intention and therefore no evidence is found to support hypothesis 1, which is therefore rejected. Hypothesis 2a expects innate customer innovativeness to have a positive relationship with the behavioural intention. The beta coefficient of innate customer innovativeness is 0.14 and is significant at the 5% level. This means that when the innate customer innovativeness increases with one unit, the behavioural intention will increase with 0.14 unit. Consequently, this finding

provides evidence to support hypothesis 2a and hence is confirmed. Moving on, hypothesis 3a formulates a negative relationship between the effort expectancy and the behavioural

intention. The beta coefficient of effort expectancy is significant at the 0.1% level and equals -0.45, which means that if the effort expectancy goes up by one unit, the behavioural intention will go down by 0.45 unit. Hence, it can be stated that evidence is found to support hypothesis 3a and consequently this hypothesis is confirmed. The last hypothesis that formulates a direct relationship is hypothesis 4, which states the negative relationship between risk perception and the behavioural intention. Also, the beta coefficient of risk perception is significant at the 0.1% level. However, the beta coefficient of risk perception equals -0.38, which means that if the risk perception goes up by one unit, the behavioural intention goes down by 0.38 unit. Consequently, also for hypothesis 4 evidence is found to support its statement, which is therefore confirmed.

(38)

37 The second model, which includes the demographics, was also found to be statistically

significant F (7, 208) = 21.07; p < 0.001. The second model explains 41.5% of the variance in the behavioural intention, which is a change of 3.8%. The change in R-squared of 0.038 is statistically significant F (3, 208) = 4.55; p < 0.01. None of the demographics show a significant direct relationship with the behavioural intention. However, these relationships were not in the aim of this research and therefore these results will not be further discussed.

The third model as a whole is again found to be significant F (11, 204) = 13.73; p < 0.001 and has an explaining variance of 42.5%. However, the change in the R-squared is not significant. This model will be used to determine the moderating effects of the demographics (i.e.,

income, age and gender). Hypothesis 2b states that income moderates the positive relationship between innate customer innovativeness and behavioural intention in such a way that this relationship is stronger for higher values of income. However, the beta coefficient of the interaction term between income and innate customer innovativeness shows that there is no significant moderating effect of income on this relationship. Therefore, no evidence is found to support hypothesis 2b and the hypothesis is therefore rejected. Hypothesis 2c expects that age is moderating the positive relationship between innate customer innovativeness and behavioural intention in such a way that this relationship is stronger for lower values of age. Again, the interaction term’s beta coefficient is not significant, which in turn does not provide any evidence to support hypothesis 2c. Hypothesis 2c is therefore also rejected. No

moderating effect of age on the relationship between innate customer innovativeness and behavioural intention is found. Hypothesis 3b is again related to the age variable and states that age is moderating the negative relationship between effort expectancy and behavioural intention in such a way that this relationship is stronger for lower values of age. Nevertheless, also the interaction term between age and effort expectancy does not show any significance

(39)

38 regarding its beta coefficient. Therefore, no moderating effect of age on the relationship between effort expectancy and behavioural intention is found and hypothesis 3b is not supported, which means the hypothesis is rejected. Finally, hypothesis 3c formulates that gender is moderating the relationship between again effort expectancy and behavioural intention in such a way that this relationship is stronger for females. Unfortunately, also hypothesis 3c is not supported and thus rejected. This is because the beta coefficient of the interaction term between gender and effort expectancy does not show any significance. This means that no evidence is found to support any of the hypothesized moderating effects studied in this research.

4.5 Assumptions of The Hierarchical Multiple Linear Regression

In this subchapter, the assumptions of the hierarchical multiple linear regression are tested. The first assumption is the need of a linear relationship between the independent and

dependent variables. The second assumption is that the residuals are normally distributed. The third assumption states that there should not be any multicollinearity present among the independent variables. The fourth and last assumption is that the data should be

homoscedastic. The results will be discussed below.

4.5.1 Linear Relationship Between Independent and Dependent Variables

The scatterplots are used to determine the linear relationship between the independent and dependent variables. All the scatterplots needed to test this assumption can be found in Appendix A. For every scatterplot, the relevant independent variable will be on the x-axis and the dependent variable will be on the axis. Furthermore, the AVG tag on the x-axis and y-axis represents the fact that the variable used in the graph is the average score of the related construct. Also, the scatterplots include a best fitting line to provide a more clear picture of

(40)

39 the relationship. If the scatterplots including the best fitting line (i.e., figure 5, 6, 7 and 8) are analysed, it can be concluded that for every independent variable there is a linear relationship with the dependent variable. All scatterplots show a linear fit with the data points scattered around the linear line. Therefore, the assumption regarding linearity of the model is met.

4.5.2 Normally Distributed Residuals

The normal distribution of the residuals is the second assumption. The scatterplot with the standard predicted value on the x-axis and the standardized residuals on the y-axis is used to analyse if the errors are normally distributed. If normality is the case, the dots should be scattered around the line and no patterns should be visible. Figure 2 presents the scatterplot mentioned above. The figure shows that the dots are scattered around the line. However, it seems like there is a little negative trend in the dots. Therefore, further investigation is necessary to determine if the residuals are normally distributed.

(41)

40 Two other tools to assess the normality of the residuals are the histogram and the probability plot of the standardized residuals. Both of these graphs will be presented in figure 3 and 4 respectively. If normality of the residuals is the case, the histogram should be shaped like a bell and the dots in the probability plot should generally follow the line. Figure 3 shows that the data is generally following the normality distribution and is shaped like a bell. Also, the probability plot presented in figure 4 shows that the dots are broadly following the line. Therefore, based on figure 2, 3 and 4 we can consider the assumption regarding the normality of the residuals as met.

(42)

41 Figure 4. Probability Plot of The Standardized Residuals

4.5.3 Multicollinearity

The third assumption assumes that there is no multicollinearity present in the data. This means that none of the independent variables are strongly correlating between themselves (i.e., have a correlation of |0.80| or lower). The assumption of multicollinearity can be tested by studying the correlation matrix. In this thesis, the correlation matrix was already discussed in

subchapter 4.4 and the results were presented in table 4. The correlation matrix showed that there were several significant correlations between the independent variables. However, these independents variables were only weakly correlated (i.e., the variables have a correlation between |0.2| and |0.4|). So, based on the correlation matrix, it can be concluded that there is no multicollinearity present in the data.

Another method to test for multicollinearity is to analyse the Variance Inflation Factors (VIFs) and the Tolerance values of each variable in every regression. The VIFs should not exceed 10 and the average VIF should not be much bigger than 1. Also, the Tolerance values

(43)

42 should be greater than 0.2. The VIFs and Tolerance values will be listed in table 6. The first and second regression show that all the VIFs are below 10 and the average VIF is around 1. Also, the Tolerance values for both regressions are above 0.2 for all variables. However, when the third regression is analysed, it can be seen that some of the variables show a VIF higher than 10 and also the average VIF exceeds 1. In addition to that, for several variables the tolerance level is below 0.2. The results of regression three indicate that multicollinearity is present. Nevertheless, when regression three is investigated more deeply, the problem can be identified. The occurrence of multicollinearity in regression three is the result of the interaction terms used in the model. Logically, the effects of the initial variables and the interaction terms where this initial variables are also present are showing high correlations. Nevertheless, this is a natural result which is inevitable, since some of the variables in regression three are used twice.

Table 6. The Variance Inflation Factors and Tolerance Values

Variables Regression 1 Regression 2 Regression 3

VIF Tol. VIF Tol. VIF Tol.

Familiarity 1,31 0,76 1,34 0,75 1,35 0,74

Innate Customer Innovativeness 1,17 0,86 1,28 0,78 7,62 0,13

Effort Expectancy 1,29 0,77 1,40 0,72 6,53 0,15

Risk Perception 1,11 0,90 1,16 0,86 1,18 0,85

Income 1,50 0,67 6,14 0,16

Age 1,71 0,59 18,21 0,05

Gender (1, female; 0, male) 1,24 0,80 7,45 0,13

Income x Innate Customer Innovativeness 6,00 0,17

Age x Innate Customer Innovativeness 9,32 0,11

Age x Effort Expectancy 17,32 0,06

Gender x Effort Expectancy 11,67 0,09

Average VIF 1,22 1,37 8,43

*p< 0.05; **p< 0.01; ***p< 0.001

So, after analysing the correlation matrix in table 4, it can be concluded that the assumption related to multicollinearity has been met. However, the VIFs and Tolerance values in table 6 show a problem of multicollinearity in the third regression. But, since the multicollinearity is

The behavioural intention to adopt autonomous vehicles in the Dutch market