MSc Marketing Intelligence Thesis

MSc Marketing Intelligence Thesis

What Drives Consumers?

A Competitive Analysis of Brand Extension Success Drivers

on Consumer Preferences for Self-Driving Cars

University of Groningen

Faculty of Economics and Business

Department of Marketing

PO Box 800, 9700 AV Groningen (NL)

January 2017

by Robin Gringhuis

s2164795

Diephuisstraat 30a

9714 GX Groningen

+31650465688

robingringhuis1@gmail.com

3

ABSTRACT

Following extensive previous research on brand extensions success drivers, their influence is researched within the unique, still-to-come market environment of self-driving car retail to test their robustness within a highly competitive, high-involvement product category. Multiple studies have indicated the existence of a number of brand extension success drivers that are most influential concerning consumers’ evaluations of brand extensions. The most important drivers are included into a choice-based conjoint analysis in order to simulate a competitive market environment among 277 respondents, and investigate the drivers’ influence on consumers’ choices rather than evaluations. The results indicate that previously established knowledge on the drivers is more robust than expected, as all drivers are found to significantly affect consumers’ choice preferences. These include the fit between the parent brand and the extension product, the parent brand conviction, parent brand experience, quality of the parent brand and the relative brand familiarity. Moreover, the partially mediating effect of perceived purchase risk on the influence of the drivers is analysed and supported. Lastly, several moderation effects of consumer characteristics are tested.

4

PREFACE

5

Table of Content

1. INTRODUCTION ... 6

2. THEORETICAL FRAMEWORK ... 10

2.1 Brand Extension Success Drivers ... 10

2.2 Setting a Critical Reality Including Competition ... 13

2.3 Completing the Overview ... 14

2.4 Consumer Characteristics ... 16

2.5 Overall Brand Expectation ... 17

2.6 Conceptual Framework ... 18

3. METHODOLOGY ... 19

3.1 Research Method ... 19

3.2 Data Collection Procedure ... 19

3.5 Study Design ... 24

3.6 Covariates... 26

3.7 Moderation ... 26

4. RESULTS ... 27

4.1 Sample Characteristics ... 27

4.2 Main Effect Model ... 28

4.3 Covariate Model ... 31

4.4 Mediation Model ... 33

4.5 Moderation Model... 35

4,6 Total Effects ... 38

4.7 Hypothesis Testing Summary ... 39

5. DISCUSSION ... 40

5.1 Brand Extension Success Drivers ... 40

5.2 Perceived Purchase Risk ... 42

5.3 Consumer Characteristics ... 42

5.4 Managerial Implications ... 43

5.5 Limitations and Future Research ... 44

5.6 Conclusion... 46

6. REFERENCES ... 47

6

1. INTRODUCTION

It is highly common for companies to be present in a variety of product categories using the same established brand name. In fact, new product introduction is most often performed using brand extensions (Carter & Curry, 2011), contributing around 80% of the total product introduction (Keller, 1998). Brand extensions are defined as using an already established brand name to launch new products within and beyond original product categories, and are among the most widely used branding strategies (Völckner & Sattler, 2006). By extending brands, firms are able to tap into their brand equity investment (Court et al., 1999). It is widely deemed as a profitable strategy, as brand extensions are assumed to require less inputs, such as trade deals, advertising, or price promotions, compared to completely new brands (Collins-Dodd & Louviere, 1999).

Brand extensions are applied within a wide range of industries, ranging from FMCG products including foods and cosmetics, to high-end retail such as consumer electronics (Estes et al, 2011). A recent example of brand extensions is the case of self-driving or autonomous cars. These are defined by the US Department of Transportation’s National Highway Traffic Safety Administration (NHTSA) as “vehicles in which operations of the vehicle occur without direct driver input to control

the steering, acceleration and braking and which are designed so that the driver is not expected to constantly monitor the roadway while operating in self-driving mode.” An estimated 33 corporate

groups are currently investing in driverless R&D, ranging from large automotive industry players (Volkswagen, Volvo, BMW etc.) to leading technology brands (Google, Nvidia, Microsoft etc.) (CB Insights, 2016).

Another automotive brand involved is the Ford Motor Company. “We are no longer just an auto

company, we are also a mobility company”, said Ford CEO Mark Fields at a press conference in

7 Next to automotive brands, it is highly probable consumers will use cars by technology brands such as Google in the near future. Google is known to be one of the pioneers in self-driving technology and has been testing its prototypes since 2009, currently counting over 2 million miles of test rides (Google, 2016). Owning services such as Google Earth, Google Maps and Waze (a community-based traffic and navigation app), Google can utilize an enormous database during the development process. As Google is typically recognized as a highly innovative software producer, the link towards previously out-of-scope but highly technology-dense products such as self-driving cars is easily made.

These are but two examples of varying brands aiming to extend their product portfolio with self-driving cars. Although a commercial launch is expected to still take at least five years, mainly due to regulatory issues (Jiang et al., 2015), it is interesting to imagine brands such as Ford soon directly competing with Google over car sales. The entrance of many new competitors could potentially have a large disruptive impact on the automotive industry, depending on which brands are more preferred by consumers.

The drivers of brand extension success have been extensively investigated in previous marketing research literature by measuring consumer preferences (e.g. Aaker & Keller, 1990; Swaminathan et al., 2001). Findings of previous studies highlighted multiple important factors which drive the success of brand extensions. Among these are the perceived fit between the core brand and the brand extension (e.g. Kim & Roedder John, 2008; Park et al, 1991) as well as the perceived quality of the parent brand (e.g. Collins-Dodd & Louviere, 1999; Erdem & Swait, 2004). Völckner & Sattler (2006) identified and compared 10 extension success determinants and identified the fit between the core brand and the brand extension’s product category to be the most influential factor. Other research, e.g. by Aaker & Keller (1990) and Bottomley and Holden (2001), typically highlights parent brand quality as the other most important driver. These drivers were measured using consumer evaluations, therefore higher fit or parent brand quality lead to brand extensions being higher evaluated by consumers.

8 merely another driving variable, it sets a critical reality in which all relevant brand extension success drivers should be considered, therefore seriously questions the robustness of previous findings.

When including competition, the dynamics between the drivers change and choice becomes a more valid measure rather than evaluations (Milberg et al, 2013). Also, the large majority of previous brand extension research has been conducted in FMCG or durable goods settings, involving products that do not require particularly high commitment by consumers. Please refer to Appendix 1 for a table of previous brand extension research examples.

This study measures the preferences of consumers regarding the purchase of a specific brand of self-driving vehicle before the commercial launch is even realized. It addresses the following research question:

“Which self-driving car brands are perceived as the most preferred choice by

potential consumers, and what drives their choice?”

This research builds on previous brand extension literature, by further increasing the understanding of brand extension success drivers through applying a highly competitive setting and analysing how consumers’ choices rather than evaluations are affected. A competitive setting is realized by applying a Choice-Based Conjoint Analysis (CBC) in which respondents pick their preferred choice out of a set of different alternatives, where previous literature rather used isolated brand extensions that evaluated using Likert scales to measure consumers’ preferences. Furthermore, it is tested if the empirically accepted drivers behave similarly when applied to an industry which is perceived as highly innovative, involving higher commitment by consumers than the more common brand extension contexts of FMCG or durable goods retail. In this case, the self-driving cars retail industry is considered highly innovative as products are still under development, not yet commercially available on the market at the moment of writing, and entail a high level of technological advances.

The managerial relevance lies in understanding which aspects of a brand extension, in this case a self-driving car, make it a preferred choice by consumers or not. It provides insights in which factors have the highest influence on consumers’ choice, to understand exactly where consumers’ preferences originate from and what drives them. This information can be leveraged by managers to improve future firm performance by optimizing their brand extension success.

9 actual choice. Participants in the online questionnaire were continuously asked to select their most preferred option regarding self-driving cars.

10

2. THEORETICAL FRAMEWORK

In this section, an overview of previous scientific literature on brand extensions is presented. Firstly, an overview of previous brand extension research findings is presented. Secondly, the separate drivers are discussed in more detail. Third, recent research criticising previous findings is discussed. Fourth, two other factors influencing consumers’ choice are briefly discussed. Fifth, the moderating role of consumer characteristics on the relationships is elaborated on. Sixth, an overall expectation concerning brand preference is formulated, and lastly, an overview of all relationships and hypotheses is presented through a conceptual model.

2.1 Brand Extension Success Drivers

Despite the perceived benefits, such as leveraging brand equity, lower advertising and better access to distribution, brand extensions also face disrepute due to a relatively high percentage of failures. In fact, brand extensions of FMCG products experience failure rates of 80%, as stated by De Chernatoney (2003). This can be detrimental for brands, as a bad product performance can negatively influence the perceived quality of the overall brand (Aaker & Keller, 1990). This sparked a growing attention among market researchers for the potential drivers of brand extension evaluation and success. Multiple success factors have since been identified, providing practical insights for managers of firms to reduce brand extension failure rates.

Among the most influential of these studies is that by Völckner & Sattler (2006), who identified 10 extension success determinants that were found to be of significant (p < .10) influence in at least one other empirical study and filtered these down to ultimately state 5 most important (i.e. strongest parameter estimates) drivers. They found the overall most influential drivers to be: 1) the perceived fit between the parent brand and the product category of the brand extension; 2) the parent brand conviction; 3) the parent brand experience; 4) the perceived retailer acceptance; and 5) the perceived marketing support.

Next to these five drivers, other studies show large empirical support for the quality (i.e. strength) of the parent brand as a driver of consumers’ evaluations of brand extensions (e.g. Aaker & Keller, 1990; Bottomley & Holden, 2001). This driver was also found by Völckner & Sattler (2006) as the sixth most influential driver.

11 are not yet commercially available on the market at the moment of writing. As these factors indicate to what extent consumers perceive extensions as supported in terms of distribution and availability in stores (retailer acceptance) or advertising and receiving competent marketing support (marketing support), it does not make sense to assess these before actual advertising and distribution of the extension product has taken place.

2.1.1 Fit

The most influential driver found by Völckner & Sattler (2006) is the fit between the parent brand and the extension product, indicating how well the original brand and the extension category can be linked in the minds of consumers. In other words, fit entails the global similarity between the parent brand and the extension product. This supported findings of previous research (e.g. Aaker & Keller, 1990; Broniarczyk & Alba, 1994; Bottomley & Doyle, 1996; Martin & Stewart, 2001) also linking perceived fit to consumers’ evaluations. Aaker & Keller (1990) defined a high fit as a high perceived ability of the parent brand company to make a product within the brand extension’s product category, based on the usefulness of the skills and resources needed to manufacture the product(s) the parent brand is originally associated with to develop, refine and manufacture the new extension product. Next to the product category aspect of fit, research by Broniarczyk & Alba (1994) focused on the associations consumers have with the parent brand, i.e. how the parent brand image can be leveraged for brand extensions. In this regard, a high fit translates to a high relevance of the extended parent brand associations for the extension product. Both for the product category as well as the image fit, a higher perceived fit led to brand extensions being evaluated more positively by consumers.

Based on the literature mentioned above, it is expected that a higher overall fit between the parent brand and the extension product, being a self-driving car, will result in that brand’s product being a more preferred choice among consumers compared to lower-fit brands. Hence, the following hypothesis is constructed:

H1: A brand extension’s fit is positively associated

with consumers’ brand preferences.

2.1.2 Parent Brand Conviction

12 more favourably. Furthermore, Kirmani, Sood & Bridges (1999) referred to this phenomenon as the ownership effect. In their research, consumers owning a products of a certain brand generally tended to evaluate new extension products of the respective brand higher than non-owners. This happened for both upward (e.g. more innovative or of higher quality) and downward (e.g. cheaper or lower-quality) extension products for non-prestige brands. For prestige brands, however, only upward extensions were evaluated positively due to owners’ desire to maintain brand exclusivity. As mentioned in chapter 1 of this report, self-driving cars are regarded as of high quality and highly innovative, therefore qualify as upwards extensions within this framework and are therefore likely to be evaluated similarly regardless of the brand being (non-)prestige.

Following the above lines of reasoning, it is expected that parent brands which experience higher conviction among consumers, will find their self-driving cars more likely to be the preferred choice compared to competitors that are less trusted among consumers.

H2: A consumer’s parent brand conviction is positively associated

with consumers’ brand preferences.

2.1.3 Parent Brand Experience

The parent brand experience relates to how often consumers have previously considered, purchased or used other products of the same parent brand (Völckner & Sattler, 2006). According to research by Swaminathan et al. (2001), for more successful brand extensions, high consideration, purchase or usage rates associate positively with parent brand choice, especially among prior first-time users of the parent brand, and ultimately on market share. For prior users of the parent brand, this effect also exists the other way around, where negative reciprocal effects apply for unsuccessful brand extensions. In other words, when consumers lack concrete information about the quality of a product, they tend to use the parent brand as a quality indicator, where high previous positive experience is positively associated with the extension’s perceived quality.

Although the results above where achieved within an FMCG retail environment, a similar effect is expected to be applicable to the case of self-driving cars, where positive recent experience with the parent brand should lead to that brand’s extension product being a more preferred choice. This leads to the following hypothesis:

H3: A consumer’s parent brand experience is positively associated

with consumers’ brand preferences.

2.1.4 Quality of the Parent Brand

13 serviceability, durability, performance and incidence of defects (Garvin, 1984). However, all of these measurable brand features together often do not yet account for the total attitude. When assessing consumer preferences, researchers often included a general abstract component that could not be explained by precise brand-specific features and values (e.g. Srinivasan, 1979). Rather than attitude, Aaker & Keller (1990) conceptualized this driver as the consumer’s perception of the overall parent brand quality, based on earlier research by Zeithaml (1988) who concludes that “perceived quality is at a higher level of abstraction than a specific (set of) attribute(s) of a product.” This relationship is found (e.g. Aaker & Keller, 1990) to be positive, as a higher perceived parent brand quality leads to a brand’s extension product being regarded as qualitatively better. Therefore, self-driving cars of higher perceived quality parent brands are expected to be the more preferred choice.

H4: The perceived quality of the parent brand is positively associated

with consumers’ brand preferences.

2.2 Setting a Critical Reality Including Competition

As mentioned in section 1 of this report, more recent research criticises the robustness of previous findings and disconfirms the generalisability of the drivers mentioned earlier. Although they hold in non-competitive settings, consumers rarely make brand extension purchase decisions without regarding competitors’ products (Milberg et al., 2013). Furthermore, previous research measured brand extension success drivers mainly by consumers’ evaluations (e.g. Aaker & Keller, 1990; Bottomley & Holden, 2001; Smith & Park, 1992; Völckner & Sattler, 2006). Together, this leads to biased results as consumers tend to evaluate isolated brand extensions more positively than during comparisons with competing products (Kardes et al., 2002; Posavac et al., 2006). In order to improve the realism of brand extension research, Milberg et al. (2013) investigated the effects of several drivers on both consumers’ evaluations as well as choices, building on earlier findings of Milberg et al. (2010). When including competition, fit and parent brand quality were found to be weaker determinants of consumers’ evaluations and preferences, while an extension’s relative brand familiarity proved a crucial driver.

2.2.1 The Roles of Relative Brand Familiarity and Perceived Purchase Risk

14 choice, consumers are faced with the risk of making a bad choice. Risk should be taken into consideration as it directly affects consumers’ behaviour, preferences and choices (Campbell & Goodstein, 2001; Zhou & Nakamoto, 2007). Brand extensions are commonly perceived as reducing risk for firms, as they build on established brand equity; and for consumers as they lower uncertainty through brand familiarity when making purchase decisions (DelVecchio & Smith, 2005). However, increased competition brings higher risk for brands, due to consumers relying on brand familiarity to reduce their own decision-making risk. (Milberg et al., 2013). In case of relatively more well-known competitors, consumers regarded their choice for a certain extension as riskier, regardless of that extension’s fit and parent brand quality, and show a higher preference for familiar brands. Hoyer and Brown (1990) also observed that less experienced consumers showed higher purchase probabilities with more familiar brands.

The findings of Milberg et al. (2013) were derived within a consumer durables retail environment, where perceived purchase risk was found to be partially mediating consumers’ choice of extension product. Previous research on perceived risk applications established a fundamental understanding that, in general, products that have higher monetary and emotional value, are more complicated and more involving are considered riskier than simple convenience products of lower value and lower involvement (e.g. Deering & Jacoby, 1972; Toh & Heeren, 1982; Wu, 2001). As self-driving cars are higher in price, involvement and more complicated than FMCG or consumer durables products, they can thus be regarded as riskier. Because of this high risk perception, consumers are expected to show more cautious decision-making, leading to more precise brand preferences. Based on the literature above, the following hypotheses are therefore constructed:

H5: An extension’s relative brand familiarity is positively associated

with consumers’ brand preferences.

H6: The effects of the brand extension success drivers on consumers’ brand preferences

are partially mediated by the perceived purchase risk.

2.3 Completing the Overview

15 other product attributes. The main attribute distinguishing self-driving cars from already existing products is the self-driving feature. Therefore, the level of autonomy of the cars is chosen as the second product attribute. Thirdly, CBC analysis allows very well for the inclusion of price as a distinct product attribute to assess consumers’ willingness-to-pay or discover price thresholds (Eggers & Sattler, 2011). No other attributes are used as they would merely take away explanatory power from the brand, which is the primary focus of this research.

2.3.1 Level of Autonomy

Self-driving cars are those cars in which preferably multiple aspects of a safety-critical control function, such as braking or steering, are controlled automatically without any input necessary by the driver (NHTSA, 2013). Therefore, this leaves out vehicles that merely provide warnings to drivers, such as dashboard lights indicating a brake or oil temperature issue. Different levels of vehicle automation exist. These levels were defined in 2014 by the SAE International (former Society of Automotive Engineers) in a classification system involving six (0-5) distinct levels. These range from 0 being an automated system that has no vehicle control at all, merely warnings (i.e. the driver is manually needed for every driving act performed) to 5 where no human intervention is required other than setting a destination and starting the system. Due to the non-focal nature of this attribute, the distinction used for the CBC is limited to partially self-driving (e.g. including cruise control, parking assist and automated lane switching) versus fully self-driving vehicles.

2.3.2 Price

16 2.4 Consumer Characteristics

When conducting research on consumers’ brand preferences, it is important to acknowledge that individuals respond differently to the introduction of new products (Gatignon & Robertson, 1985). For example, people might differ in their response to high or low-fitting extension products. Early adopters are likely to be influenced less by a lower extension fit than later adopters. The main characteristic distinguishing early from late adopters is the degree to which they are innovative or rather risk-averse. Furthermore, each specific product category bears a certain perceived degree of uncertainty and risk among consumers. Both of these characteristics are expected to moderate the effect of the abovementioned attributes on consumers’ choice. The separate characteristics are discussed next.

2.4.1 Level of Innovativeness

One of the most prominent individual traits of consumers regarding innovation is their comfort with taking risk (Rogers, 1983). While brand extensions are often seen as sharply reducing risks for producers as well as consumers (Smith & Park, 1992), they are far from risk-free, for the obvious reason that they differ from previous products and thus carry a degree of uncertainty for those individuals that purchase them. The extent to which consumers have a higher or lower tolerance for risk was previously studied by Klink & Smith (2001), among others. In general, consumers with a higher risk tolerance, or higher level of consumer innovativeness) were found to be more receptive to lower-fit (i.e. signalling higher risk) extension products. On the other hand, less innovative consumers are more strongly influenced by the higher risk of lower fit extension products. As a result, the latter consumers typically exhibit resistance towards purchasing the extension product (the late adopters), while the former group shows a higher purchasing tendency despite the risk involved (the early adopters).

Based on the above literature, the following hypotheses are derived concerning the moderating role of consumers’ level of innovativeness:

H7a: The influence of the car brand on consumers’ preferences

is significantly moderated by a consumer’s level of innovativeness.

H8a: The influence of a higher level of autonomy on consumers’ preferences

is positively moderated by a consumer’s level of innovativeness.

H9a: The influence of price on consumers’ preferences

17 2.4.2 Level of Perceived Risk

The very essence of brand extension success drivers holds that these drivers are associated with the degree of perceived risk consumers experience with an extension product. As they lack experience with the product, consumers face a significant level of perceived uncertainty and risk making a purchase decision. By invoking the beliefs formed through previous experience with products of the same parent brand, consumers make assumptions about the quality of the brand extension product. In other words, experience with other products of the same brand are used as a surrogate for experience concerning the new extension product (DelVecchio, 2000). Moreover, a company engaging in brand extension uses its brand name as an implicit ‘bond’ for the product’s quality (Wernerfelt, 1988). The development of a parent brand requires substantial investments and acts as a collateral for the extension product. It is generally reasoned that no company would risk this accumulated brand investment by releasing products of inferior quality carrying the same brand name. In short, every brand extension is subject to a certain level of perceived risk, which differs per consumer based on beliefs and previous experience, influencing consumers’ decision-making.

Based hereon, the following hypotheses are derived for the moderating role of consumers’ level of perceived risk:

H7b: The influence of the car brand on consumers’ preferences

is significantly moderated by a consumer’s level of perceived risk.

H8b: The influence of a higher level of autonomy on consumers’ preferences

is negatively moderated by a consumer’s level of innovativeness.

H9b: The influence of price on consumers’ preferences

is positively moderated by a consumer’s level of perceived risk.

2.5 Overall Brand Expectation

18 choice utility within the CBC for) traditional car manufacturers (e.g. Ford, VW, Volvo, Mercedes Benz, BMW or Tesla) over leading technology companies (e.g. Microsoft, Apple, Google, Nvidia, Uber) concerning their choice of a self-driving vehicle. This leads to the final hypothesis being stated as follows:

H10: Overall, consumers prefer traditional car manufacturers over leading technology

companies when choosing their preferred self-driving car brand.

19

3. METHODOLOGY

This section presents the method and design of the research conducted in this study. The research method is discussed and justified first. Next, the data collection procedure is presented. Furthermore, the study design and plan of analysis are elaborated on. Lastly, the moderating effect is discussed.

3.1 Research Method

The analysis method chosen is a Dual Response Choice-Based Conjoint (CBC). Within a CBC, participants are presented with a series of choice sets, consisting of different combinations of product attributes (McCullough, 2002). Within this study, consumers are shown different self-driving cars. By choosing their preferred option, their evaluation of the different attribute levels is registrered. The results of the tests are statistically analyzed to dissect product preferences (Eggers & Sattler, 2011). Next to product types, this method is applicable to any services or layout designs that can be dissected into multiple attributes (Hair et al., 2010). In this research, respondents are asked to select their preferred choice out of three alternatives, which contain different levels of the attributes brand, level of autonomy and price. Respondents are expected to prefer the alternative of which the total utility is maximized (Eggers & Eggers, 2011; Jun & Park, 1999). Furthermore, a dual response no-choice option was included in order to gain insights if respondents were actually willing to use their chosen self-driven car, or rather use a manually driven car instead.

3.2 Data Collection Procedure

The online survey is created using the software Preference Lab. It contains all choice sets, as well as a range of multi-item measures in order to measure the brand extension success drivers, and additional questions for several control variables. It starts by introducing the experiment through a brief explanation of self-driving cars. Next, several control variables are gathered as respondents are asked for their gender, age, nationality and whether or not they own a car. Subsequently, several multi-item measures are used for the evaluation of the different brands. This is done using a pick-any procedure opposed to full rating scales leading to binary (0/1) coded, rather than interval scaled, response data as follows:

𝑌𝑖𝑗 = {1 𝑖𝑛 𝑐𝑎𝑠𝑒 𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑒𝑛𝑡 𝑖 𝑐ℎ𝑜𝑜𝑠𝑒𝑠 𝑜𝑏𝑗𝑒𝑐𝑡 𝑗 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

20

BRAND COMPANY TYPE

TESLA MOTORS New car manufacturer

APPLE Leading tech company

GOOGLE Leading tech company

MERCEDES BENZ Traditional car manufacturer

FORD Traditional car manufacturer

ROBOCAB Benchmark brand

Table 1: Overview of the included self-driving car brands

A distinction here is made between car manufacturers and leading tech companies. In order to assure that respondents have associations with the brands concerned, the well-known brands of Apple and Google were chosen to represent the technology companies. For the same reason, Mercedes Benz and Ford were chosen as traditional manufacturers. Furthermore, Tesla Motors is included, being the only car-manufacturing company currently having partly (level 2-3) self-driving cars widely available. Furthermore, an unknown brand, Robocab, was added to the choice experiment. This is motivated by Hoeffler & Keller (2003) defining brand equity as “the differential effect that brand knowledge has on consumer response, […] compared to when a brand is not identified.”

Next, during the conjoint phase, respondents are made to choose their preferred option, as well as indicate whether they would actually purchase this option via the dual-response (0/1) no-choice option. Lastly, respondents are asked to answer several questions relating to the moderating variables. To collect the data, the survey is spread almost entirely via social media, specifically through Facebook and Whatsapp. Next to this, additional respondents are gathered through personal contact.

3.2.1 Fit

The fit between the parent brand and the extension product is measured using separate questions for product category fit (Aaker & Keller, 1990) and image fit (Broniarczyk & Alba, 1994). The corresponding questions are derived from the paper of Völckner & Sattler (2006), although transformed into a pick-any formulation. The data of both questions are combined to form an overall score for the fit of the extension product. Table 2 below presents the items measuring the fit.

Measure of Fit

 Please select all brands that have the people, facilities and skills to create self-driving cars. Please check all that apply.

 Please select all brands that currently offer products and/or services that are closely tied to self-driving cars. Please check all that apply.

21 3.2.2 Parent Brand Conviction

In order measure the conviction or trust consumers have in the brands of interest, the question is again derived from research by Völckner & Sattler (2006), based on the findings of DelVecchio (2000) and Kirmani, Sood & Bridges (1999) and transformed into a pick-any formulation. For the sake of interview length, all drivers except for the fit are measured within one question to maximize respondents’ attention. Table 3 below presents the item measuring the parent brand conviction.

Measure of Parent Brand Conviction  Please select all brands that you trust. Please check all that apply.

Tesla Motors Apple Google Mercedes Benz Ford Robocab None of the above Table 3: Measure of Parent Brand Conviction

3.2.3 Parent Brand Experience

As mentioned earlier, parent brand experience relates to purchase, usage or consideration rates of the parent brand by consumers. However, as consideration and purchase rates are considerably lower within the product category of cars, rather than the more commonly applied category of FMCG or consumer durables, this item is measured through recent usage rates. It is based on the question formulated by Völckner & Sattler (2006), derived from previous theory by Broniarczyk & Alba (1994) and Swaminathan et al. (2001) and transformed into a pick-any formulation. Table 4 below presents the item measuring the parent brand experience.

Measure of Parent Brand Experience

 Please select all brands that you have used in the last 12 months. Please check all that apply.

Tesla Motors Apple Google Mercedes Benz Ford Robocab None of the above Table 4: Measure of Parent Brand Experience

3.2.4 Quality of the Parent Brand

As the quality of the parent brand measures the overall attitude over consumers’ based on the quality of the products offered by the parent brand (Aaker & Keller, 1990; Sheinin & Schmitt, 1994), the questions formulated by Völckner & Sattler (2006) can be combined into one pick-any formulation. Table 5 below presents the item measuring the quality of the parent brand.

Measure of Quality of the Parent Brand

 Please select all brands that offer high-quality products and/or services. Please check all that apply.

22 3.2.5 Relative Brand Familiarity

This driver is measured directly through assessing the familiarity of the self-driving car brands among respondents, based on the literature by Milberg et al. (2013). A pick-any formulation creates a clear overview of the relative familiarity of the brands in the mind of respondents. Table 6 below presents the item measuring the relative brand familiarity.

Measure of Relative Brand Familiarity

 Please select all brands that you are familiar with. Please check all that apply.

Tesla Motors Apple Google Mercedes Benz Ford Robocab None of the above Table 6: Measure of Relative Brand Familiarity

3.2.6 Perceived Purchase Risk

Lastly, the perceived purchase risk associated with the brands is measured to test for possible mediation effects. The perceive purchase risk is measured using a direct question derived from literature by Milberg et al. (2013), also transformed into a pick-any formulation. Table 7 below presents the item measuring the perceived purchase risk.

Measure of Perceived Purchase Risk

 When buying a self-driving car, which brands would you consider a risky choice? Please check all that apply.

Tesla Motors Apple Google Mercedes Benz Ford Robocab None of the above Table 7: Measure of Perceived Purchase Risk

3.3 Conjoint Attributes

The next step in the survey entails the CBC. In order to conduct the conjoint analysis, the several attributes depicted in the conceptual model first ought to be decomposed into distinctive levels. These are then randomly allocated to respondents. The number of levels should be a realistic and applicable number, typically two to five for each attribute in order to be perceived as reasonable and acceptable by respondents (Eggers & Sattler, 2011). Table 8 shows how the attributes brand, level of autonomy and price are divided into levels, as discussed earlier in chapter 2, to be used within the analysis.

ATTRIBUTE LEVELS

CAR BRAND Tesla Motors Apple Google

Mercedes Benz Ford

Robocab

LEVEL OF AUTONOMY Partially self-driving Fully self-driving

23 €12 €14 €16 €18 €20

Table 8: Overview of the Separate Levels per Attribute

Note that the price depicts a rental price per hour. Within the conjoint analysis, respondents were asked to imagine themselves going on a holiday in the year 2025. Rather than renting a manually-driven rental car, new self-driving cars are available at the destination. These cars are not rented for the entire holiday, but called upon via an app whenever needed. They drive themselves to the user’s destination upon request and automatically park themselves after. Users ought to pay an hourly rate for the car, which includes fuel costs, parking tickets, insurance and taxes. Should respondents not find any of the offered choice alternative attractive enough, the no-choice option presents driving a manually driven car for €49 per day. The rationale behind the dual response no-choice option entails that in order to improve the realism of the study, respondents should be able to state their preference while indicating that they find none of the presented options attractive enough, just as consumers can refrain from actually purchasing an alternative within a real-life scenario (Vermeulen et al., 2008).

3.4 Consumer Characteristics

Lastly, after the conjoint items, respondents are asked to answer several questions relating to the moderating variables, i.e. consumer-specific characteristics. These are measured separately for the Level of Innovativeness and the Level of Perceived Risk. In order to measure the Level of Innovativeness, i.e. the risk tolerance of respondents, 2 separate statements are derived from the research by Midgley & Dowling (1978), as also used in the study by Klink & Smith (2001). Respondents are asked to indicate whether or not they agree with two statements relating to their general attitude towards innovation on a 7-point scale. These separate measures are then manually added up into one Level of Innovativeness variable. Table 9 below presents the items measuring the Level of Innovativeness.

Measure of Level of Innovativeness  Overall I enjoy buying the latest products.

 I like to buy new products before others.

24 For the second consumer characteristic, a similar approach is applied based on the research by DelVecchio (2000) on consumers’ level of perceived risk concerning product categories. By answering 2 statements relating to the level of risk respondents associate with buying a self-driving car, and manually adding the two variables together, an overall score for every respondent’s Level of Perceived Risk is constructed. Please note the difference between this variable and the Perceived Purchase Risk. The Perceived Purchase Risk is a brand-specific variable indicating to what extent the separate brands are considered a risky choice, whereas the Level of Perceived Risk is a consumer-specific moderating characteristic indicating the amount of risk consumers feel concerning self-driving cars in general. Table 10 below presents the items measuring the Level of Perceived Risk.

Measure of Level of Perceived Risk

 If I choose self-driving cars, I would feel very uncertain of the level of quality that I am getting.  I prefer choosing a well-known self-driving car because I need the reassurance of an established

brand offering this product.

Table 10: Measure of Level of Perceived Risk

3.5 Study Design

25

Figure 1: Choice Set Example within the Survey

3.5.1 Plan of Analysis

In order to estimate the value appointed by the respondents to the separate attributes, i.e. the attributes’ utilities, the following equation is applied:

𝑈

= ∑ ß

𝑋

𝑘=1

Herein, U represents the total utility among all respondents for the preferred type of self-driving car (j), calculated using the sum of utilities (ß) of the separate explanatory variables (X; k=1, 2… K). The statistical software package LatentGold is used to estimate the utilities of the attributes as well as the moderation effects.

3.5.2 Model Fit

26 3.6 Covariates

In order to assess the influence of the separate brand extension success drivers, brand-specific covariates have to be created before carrying out the choice analysis. The pick-any formulation of the multiple-item measures results in binary-coded data in which a separate variable is made for each brand, resulting in 7 (6 brands plus the no-choice option) variables per driver. Using IF-statements, one dummy variable is created per driver in which only the value of the brand of interest is used for analysis. After doing so, the covariates (fit, parent brand conviction etc.) can be included into the analysis within LatentGold. The variable Perceived Purchase Risk is treated using the same method in order to be included into the LatentGold analysis.

3.7 Moderation

27

4. RESULTS

The results chapter presents the statistical outcomes of the research. First, the sample of respondents is described by its general characteristics. Next, the preference estimations are presented for the main model. Next, additional models are constructed including the covariates, mediator and the moderators to test their effects. Hereafter, an overview of the total effects of the separate car brands on consumers’ preferences is provided. In the last section, an overview of the hypothesis-testing results is presented.

4.1 Sample Characteristics

The total sample used for the analysis consists of 277 respondents. As can be seen in table 11 below, the sample is quite evenly distributed concerning respondents’ gender, with 47% male and 53% female respondents. More than half of the respondents (n=157; 57%) belong to the youngest age group of 18-24, and with 92 respondents (33%) in the second age group of 25-34, and an average age of 26.43, the overall sample can be described as relatively young adults. The youngest respondent (n=1) is 18 years old, the oldest respondent (n=1) is 73 years old. Out of the total 277 respondents, 89 (32%) own a car, where 188 (68%) do not.

GENDER FREQUENCY % MALE 130 47% FEMALE 147 53% AGE 18-24 157 57% 25-34 92 33% 35-44 16 6% 45-54 19 3% 55+ 3 1% CAR OWNERS YES 89 32% NO 188 68%

Table 11: Sample Characteristics and Distribution of Respondents

28 4.2 Main Effect Model

In the following section, the main effects of the CBC are presented and discussed. Firstly, the model performance is compared to the null model. Next, the utilities of the separate attributes are elaborated on, after which their relative importance is discussed.

4.2.1 Model Comparison

In order to estimate the main model (model 1), each attribute first needs to be specified within LatentGold. As all three attributes are nominal, they can be estimated as part-worth attributes. In order to compare the model with the null model, i.e. the model without parameters, a log-likelihood (LL) ratio test is conducted. Table 12 below presents the outcomes of the model comparison. Firstly, the results indicate that the main model clearly performs significantly better than the null model, as shown by the Chi-square (X²) statistic exceeding the critical value of 124.342 (df = 265; α = 0.05) as well as the hit rate (i.e. percentage of correctly predicted outcomes) being approximately double that of the null model. Also, the McFadden adjusted R², or rho-square, is calculated for the main model. This value is based on the difference in LL between the models compared and penalized for the number of parameters in the model, thereby accounting for

29 parsimoniousness, and it behaves better for multinomial logit models such as conjoint analyses than the common R² index. If the new model does not perform substantially better, there is no large difference in LL, resulting in a low R²adj. A McFadden R²adj value between 0.2-0.4 is generally

regarded as reflecting very good to excellent model performance (McFadden, 1974). Therefore, the analysis is continued for the main model.

MODEL LL X² NPAR DF R²ADJ AIC(LL) HIT RATE

NULL MODEL -6694.9432 - - - - - 33.33%

MAIN MODEL -4508.8482 1.014E+14 12 265 0.3247 9041.6963 65.47%

Table 12: Model Comparison of the Null & Main Model

4.2.2 CBC Results and Utility Levels

Depicted in table 13 below are the utility estimates of the separate attributes included in the main model, along with their respective Wald values and significance levels. The results show that car brand, level of autonomy and price all have a highly significant (p = 0.000) effect on the preference utilities at a 1% significance level.

ATTRIBUTE UTILITY WALD STATISTIC P-VALUE

CAR BRAND 459.5294 0.000*** TESLA MOTORS 0.5079 APPLE -0.0818 GOOGLE 0.0617 MERCEDES BENZ 0.5226 FORD -0.1287 ROBOCAB -0.8817 LEVEL OF AUTONOMY 14.7781 0.000*** PARTIALLY SELF-DRIVING -0.0715 FULLY SELF-DRIVING 0.0715 PRICE 314.9411 0.000*** € 10 0.5110 € 12 0.2702 € 14 0.1104 € 16 -0.0101 € 18 -0.2830 € 20 -0.5985 NONE OPTION -1.1073 587.0135 0.000***

Table 13: Overview of the Utility Estimates per Attribute Level

30 Closely following Mercedes Benz, Tesla Motors is the second most-preferred brand among respondents at a utility level of 0.5079. A slightly positive utility is allocated to Google, after which Apple, Ford and the benchmark brand Robocab all show negative utility levels. Concerning the level of autonomy, respondents prefer a fully driving (utility = 0.0715) over a partially self-driving car (-0.0715). The attribute price shows a linear distribution of utilities, as the highest utility (0.5110) is allocated to the lowest price level of €10, and utilities decrease as the price level increases.

Figure 3: Part-Worth Utilities of the Attributes in the Main Model

4.2.3 Attribute Importance

31

Figure 4: Relative Attribute Importance in the Main Model

4.3 Covariate Model

In the following section, the influence of the separate brand extension success drivers is investigated in order to test hypotheses 1-5. This is done by adding the covariates created for the drivers as attributes to the main model within LatentGold. The overall performance of the new model is first compared to that of the main model, after which an overview is given of the shift in relative attribute importance caused by adding the covariates.

4.3.1 Model Comparison

Table 14 shows the model performance of the new covariate model compared to the main model. Firstly, the log-likelihood decreased after extending the model. The Chi-square statistic further increased and is still highly significant. A direct model comparison can be made using the R²adj, the

AIC and the hit rate. As the R²adj increased further towards 0.4, the model performance clearly

increased by introducing the covariates. This is confirmed by the AIC. This value is based on the LL and is a measure used to compare the goodness-of-fit of models. A lower AIC represents higher model quality; thus the covariate model has a higher goodness-of-fit. Lastly, the increase in hit rate also confirms the increase in predictive performance of the covariate model compared to the main model.

MODEL LL X² NPAR DF R²ADJ AIC(LL) HIT RATE

MAIN MODEL -4508.8482 1.014E+14 12 265 0.3247 9041.6963 65.47%

COVARIATE MODEL

-4283.9654 5.306E+14 17 260 0.3576 8601.9307 68.18%

Table 14: Model Comparison of the Main & Covariate Model

4.3.2 CBC Results

The introduction of the brand extension success drivers as covariates causes a shift in utilities for the separate attribute levels, as depicted in table 15. Each attribute remains significant at a 1%

52,86% 5,38% 41,75% 0% 10% 20% 30% 40% 50% 60%

Car Brand Level of Autonomy Price per Hour

Re lat iv e im p o rta n ce Attributes

32 level, although the no-choice option is now insignificant. The utilities are observed to shift from the car brands to the brand-specific covariates. As these covariates are included to explain the attitude of consumers towards the brands, an ideal situation would present zero-level utilities for the car brands, as they fully shifted towards the explanatory covariates. Any utilities of the car brands are possibly explained by other drivers not included in the current model. Therefore, the individual brands’ utilities are not interpreted here. The utilities of level of autonomy and price remains approximately equal although the utility range of price has increased since the positive and negative utility extremes diverted. Finally, the five brand extension success drivers all show significant positive utilities on the choice preferences. Therefore, all five drivers are shown to positively affect consumers’ brand preferences. Therefore, hypotheses 1, 2, 3, 4, and 5 are all supported by the data.

ATTRIBUTE UTILITY WALD STATISTIC P-VALUE

CAR BRAND 66.1238 0.000*** TESLA MOTORS 0.2242 APPLE -0.2738 GOOGLE -0.3051 MERCEDES BENZ 0.2544 FORD -0.0087 ROBOCAB 0.1089 LEVEL OF AUTONOMY 15.8732 0.000*** PARTIALLY SELF-DRIVING -0.0764 FULLY SELF-DRIVING 0.0764 PRICE 352.3241 0.000*** € 10 0.5757 € 12 0.2847 € 14 0.1225 € 16 -0.0262 € 18 -0.3124 € 20 -0.6443 NONE OPTION 0.1336 2.0692 0.150 BRAND FIT 0.3556 114.4284 0.000*** BRAND CONVICTION 0.5935 105.9362 0.000*** BRAND EXPERIENCE 0.1454 4.6373 0.031** BRAND QUALITY 0.3129 18.4289 0.000***

RELATIVE BRAND FAMILIARITY 0.3088 9.8156 0.002***

Table 15: Overview of Separate Utilities per Attribute and Driver in the Covariate Model

4.3.3 Attribute Importance

33 the attributes in the covariate model is presented. Furthermore, the relative importance of the separate extension success drivers is depicted in figure 6. The highest importance (34.34%) is allocated to the fit between the parent brand and extension product, followed by the parent brand conviction (28.65%), the quality of the parent brand (15.10%), the relative brand familiarity (14.90%) and lastly the parent brand experience (7.01%).

Figure 5: Relative Atribute Importance in the Covariate Model

Figure 6: Relative Driver Importance in the Covariate Model

4.4 Mediation Model

Next, the role of the Perceived Purchase Risk (PPR) is investigated in order to test hypothesis 6. As mentioned earlier, this variable is expected to partially mediate the relationship between the brand extension success drivers and consumers’ brand preferences. In order to test for the presence of mediation, three separate models are tested.

52,86% 5,38% 41,75% 28,95% 7,90% 63,15% 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7

Car Brand Level of Autonomy Price

Re lat iv e Impo rta n ce Attributes

RELATIVE ATTRIBUTE IMPORTANCE

Main model Covariate model

34,34% 28,65% 7,01% 15,10% 14,90% 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4

Fit PB Conviction PB Experience Quality of the PB Relative Brand

Familiarity Re lat iv e Impo rta n ce

Brand Extension Success Drivers

34 Firstly, a binary logit model of the separate brand extension success drivers on the PPR is conducted within LatentGold, in order to check for a significant relationship. The results reveal that all five separate drivers significantly (p = 0.000; all at a 1% significance level) influence the PPR.

Secondly, a multinomial logit model (i.e. standard conjoint analysis) is conducted where the influence of PPR on the attributes Car Brand, Level of Autonomy and Price is analysed. The result shows a highly significant (p = 0.000) negative utility of PPR on the consumers’ brand preferences. In other words, less risky brands experience a higher preference among consumers in general. Thirdly, a multinomial logit model is conducted where PPR is included among the separate drivers in order to test for full (in case the drivers become insignificant after inclusion of PPR) or partial (in case the drivers still show significant utilities) mediation. The results are presented in Appendix 2, and show that after including PPR, all other drivers still show significant effects, as does PPR, which shows a negative utility (-0.3892; p = 0.000). Therefore, full mediation is not present and PPR is indeed found to be partial mediator of the brand extension success drivers. Therefore, hypothesis 6 is supported by the data. Next, this model is evaluated to compare whether the inclusion of PPR also improves the overall model performance.

4.4.1 Model Comparison

The latter model is compared using the aforementioned measures of model performance, as shown below in table 16. The McFadden R²adj increases when including PPR, indicating an

improvement of the model. This is confirmed by the lower AIC value indicating a better goodness-of-fit of the new model. Lastly, an increase in hit rate of 0.35% points out that a slight, but nevertheless apparent improvement is made. Therefore, the PPR is included for further analysis.

MODEL LL X² NPAR DF R²ADJ AIC(LL) HIT RATE

COVARIATE

MODEL -4283.9654 5.306E+14 17 260 0.3576 8601.9307 68.18% MEDIATION

MODEL -4262.6501 2.787E+14 18 259 0.3606 8561.3002 68.53%

Table 16: Model Comparison of the Covariate and Mediation Model

4.4.2 CBC Results

As mentioned, the utilities of the 5 drivers remain significant. The introduction of PPR mediates but does not cause major shifts among the drivers. An overview is included in Appendix 2. It does, however, influence the relative importance of the drivers.

4.4.3 Attribute Importance

As shown in figure 7, the PPR ranks 3rd _{concerning relative importance compared to the extension}

35 relatively equal, although quality of the parent brand has become slightly less important than relative brand familiarity.

Figure 7: Relative Driver Importance after the Introduction of PPR

4.5 Moderation Model

This section presents the moderating effects of the consumer characteristics, in order to test hypotheses 7, 8 and 9. The moderation variables are included by adding the 24 created interaction variables as separate attributes into the CBC in LatentGold. Firstly, a comparison is made between the previous model and the model including moderators. Next, an overview is given of the attributes and significance levels, and the moderation hypotheses are tested.

4.5.1 Model Comparison

The introduction of the consumer characteristics as moderators further improves the model performance, as shown by the measures depicted in table 17. A further increase in McFadden R²adj

is realized, as well as a decrease in AIC, both indicating that the model shows higher performance than previous models. This is confirmed by another 0.72% increase in hit rate, as 69.25% of respondents’ choices are now correctly predicted.

MODEL LL X² NPAR DF R²ADJ AIC(LL) HIT RATE

MEDIATION MODEL

-4262.6501 2.787E+14 18 259 0.3606 8561.3002 68.53%

MODERATION

MODEL -4193.3982 1.439E+15 44 233 0.3671 8474.7964 69.25%

Table 17: Model Comparison of the Moderation & Mediation Model

27,00% _25,12% 6,50% 12,03% 12,40% 16,95% 0% 5% 10% 15% 20% 25% 30% 35% 40% Fit PB Conviction PB Experience Quality of the PB Relative Brand Familiarity Perceived Purchase Risk Re lat iv e Impo rta n ce

Brand Extension Success Drivers

RELATIVE DRIVER IMPORTANCE

36 4.5.2 CBC Results

Table 18 presents the attributes and significance levels after inclusion of the moderators. The 3 attributes remain significant (p = 0.000), and the overall no-choice option remains insignificant. Also, the extension success drivers including PPR do not show considerable shifts.

In order to test hypotheses 7, 8 and 9, the moderating effects of the consumer characteristics are analysed. First, the interaction variables for Level of Innovativeness (“Innov” in table 18) are considered. The car brands Tesla Motors, Apple, Google and Mercedes Benz show insignificant results. Only Ford negatively interacts at a 10% significance level (p = 0.086). Although this result is quite weak, a slight interaction effect seems to be present. Thus, hypothesis 7a is partially supported. Next, a partially self-driving level of autonomy shows a highly significant (p = 0.005) negative interaction effect, indicating that consumers who score higher on Innovativeness prefer a higher level of autonomy. Therefore, hypothesis 8a is supported by the data. Furthermore, the interaction between Innovativeness and price is analysed, showing a significant (p = 0.014) negative interaction only for a price level of €16. As the other price levels do not interact significantly, no interpretation of the overall interaction between Innovativeness and price is possible. As a result, hypothesis 9a is not supported. Lastly, a highly significant (p = 0.000) negative interaction effect is observed concerning the no-choice option. Therefore, consumers scoring higher on Innovativeness are observed to have significantly higher intentions of actually purchasing their stated choice preference.

Next, the interaction of the attribute levels with consumers’ Level of Perceived Risk is investigated. Firstly, the car brands show insignificant results, except for Google interacting negatively at a 10% significance level (p = 0.095). Therefore, similar to consumers’ Level of Innovativeness, a slight interaction effect seems to be present, thus hypothesis 7b is partially supported by the data. Concerning the level of autonomy, a highly significant (p = 0.000) positive interaction effect is observed, indicating that consumers experiencing a higher level of perceived risk prefer a partially self-driving car. Following this observation, hypothesis 8b is supported. Next, only the €10 price level shows significant (p = 0.050) interaction effects with consumers’ level of perceived risk, indicating that a higher level of perceived risk makes consumers prefer a higher price level than €10. However, since other price levels cannot be interpreted, hypothesis 9b is only partially supported by the data. Finally, the no-choice option shows a significant (p = 0.049) positive interaction with the no-choice option, implying that a higher level of perceived risk leads consumers to have a higher preference for the option of choosing a manually driven car rather than their indicated choice preference.

ATTRIBUTE UTILITY WALD STATISTIC P-VALUE

37 TESLA MOTORS _0.2573 APPLE -0.2503 GOOGLE -0.2389 MERCEDES BENZ _0.1938 FORD _-0.0463 ROBOCAB 0.0845 LEVEL OF AUTONOMY 47.4422 0.000*** PARTIALLY SELF-DRIVING _-0.1819 FULLY SELF-DRIVING _0.1819 PRICE 257.0833 0.000*** € 10 0.6649 € 12 0.3571 € 14 _0.1522 € 16 _-0.0391 € 18 -0.3478 € 20 -0.7873 NONE OPTION -0.0855 0.6181 0.430 BRAND FIT 0.3213 87.0685 0.000*** BRAND CONVICTION 0.5450 85.8487 0.000*** BRAND EXPERIENCE 0.1625 5.5020 0.019** BRAND QUALITY 0.2833 14.6616 0.000****

RELATIVE BRAND FAMILIARITY 0.2873 8.2053 0.004****

PERCEIVED PURCHASE RISK -0.4003 43.5510 0.000****

INNOVXTESLAMOTORS 0.0161 1.3029 0.250 INNOVXAPPLE -0.0093 0.3595 0.550 INNOVXGOOGLE -0.0036 0.0623 0.800 INNOVXMERCEDESBENZ 0.0168 1.5189 0.220 INNOVXFORD -0.0255 2.9463 0.086* INNOVXLEVELOFAUTONOMY -0.0177 7.7594 0.005*** INNOVXPRICE10 -0.0085 0.3722 0.540 INNOVXPRICE12 0.0109 0.5960 0.440 INNOVXPRICE14 0.0211 2.0921 0.150 INNOVXPRICE16 -0.0363 6.0082 0.014** INNOVXPRICE18 0.0029 0.0344 0.850 INNOVXNONEOPTION -0.1152 56.2651 0.000*** RISKXTESLA -0.0254 1.6619 0.200 RISKXAPPLE 0.0081 0.1532 0.700 RISKXGOOGLE -0.0333 2.7844 0.095* RISKXMERCEDESBENZ 0.0137 0.5238 0.470 RISKXFORD 0.0299 2.0761 0.150 RISKXLEVELOFAUTONOMY 0.0582 43.4848 0.000*** RISKXPRICE10 -0.0374 3.8543 0.050** RISKXPRICE12 -0.0321 2.6836 0.100 RISKXPRICE14 -0.0188 0.8902 0.350 RISKXPRICE16 0.0134 0.4219 0.520 RISKXPRICE18 0.0188 0.7656 0.380 RISKXNONEOPTION 0.0426 3.8772 0.049**

38 4.6 Total Effects

As mentioned in section 4.3.2, the individual car brands’ utility parameters are not interpreted due to the shift in utility caused by the introduction of the brand extension success drivers. The total effect of the separate car brands on consumers’ choices is a combination of the utility residuals of the brands and the individual effects of the brand extension success drivers. In order to calculate the total effects, the effects of each separate driver, including PPR, are multiplied by the percentage of respondents who selected that brand within the pick-any formulated questions on the drivers. These values are then added up per brand to obtain the share of total effect as explained by the drivers included in the model. Lastly, the utility residuals are added per brand to derive the total effect of each brand on consumers’ choice preferences within the conjoint analysis. Ultimately, this results in the values depicted in figure 8, which are now suitable for interpretation. The total brand effects ought to be interpreted relative to the effect of the benchmark brand, since brand equity equals the differential effect of a familiar brand on consumers’ behaviour compared to an unidentified brand. The results reveal that the brands of Mercedes Benz (1.52) and Tesla Motors (1.49) have the highest differential effect on consumers’ choice. In other words, these are the most preferred brands among consumers when considering self-driving cars. Smaller effects are observed for Google (1.02), Apple (0.83) and Ford (0.80). These findings, together with the overall brand effects of the main model of section 4.2, shows that the car manufacturers Mercedes Benz and Tesla Motors, experience considerably higher preferences among consumers. Therefore, although Ford scores considerably lower, hypothesis 10 is supported by the data.

Figure 8: Total Brand Effects as a Differential Effect Compared to an Unknown Brand

1,49 0,83 1,02 1,52 0,80 0 0,5 1 1,5 2

Tesla Motors Apple Google Mercedes

Benz

Ford Robocab

39 4.7 Hypothesis Testing Summary

HYPOTHESIS VARIABLE RESULT

1 Fit Supported

2 Parent Brand Conviction Supported

3 Parent Brand Experience Supported

4 Quality of the Parent Brand Supported

5 Relative Brand Familiarity Supported

6 Perceived Purchase Risk Supported

7A Level of Innovativeness Partially supported

7B Level of Perceived Risk Partially supported

8A Level of Innovativeness Supported

8B Level of Perceived Risk Supported

9A Level of Innovativeness Not supported

9B Level of Perceived Risk Partially supported

40

5. DISCUSSION

This chapters presents a discussion of the research findings. First, the theoretical implications of the findings on the separate brand extension success drivers, the mediating variable and the interaction effects are elaborated on, followed by a discussion of the managerial implications. Furthermore, the limitations of this report and suggestions for future research are presented. The report ends with an overall conclusion.

5.1 Brand Extension Success Drivers

This study improves the scientific understanding of brand extension success drivers within a highly competitive environment by applying a choice-based conjoint analysis to the high-involvement product category of self-driving cars. The research results present that the hypothesized expectations concerning the success drivers were all supported. The drivers’ relative importance is especially notable, as well as the findings on relative brand familiarity, which are elaborated on later.

5.1.1 Fit

The results show that the perceived fit between the parent brand and the extension product has the highest importance (27.70% of the drivers included in the moderation model) in influencing consumers’ decision-making, also within a highly competitive setting. This finding questions previous research findings by Milberg et al. (2013), arguing that the inclusion of competition and choice significantly decreases the influence of fit. Also when researching consumers’ choice rather than product evaluations, the fit is found to be an important predicting factor. Within the product category of self-driving cars, possessing a unique character as the products are not yet commercially launched, consumers do seem to rely on the global similarity they perceive to exist between the brand and the product. The high importance of fit is in line with previous research by e.g. Völckner & Sattler (2006) highlighting that consumers prefer extensions products to which they can easily link the parent brands or make relevant associations based on the parent brand. Consumers do consider the usefulness of the skills and resources they associate with the parent brand in order judge that brand’s ability to develop and produce quality self-driving cars, which leads to the car manufacturers generally being preferred over the technology companies.

5.1.2 Parent Brand Conviction