THE EFFECTS OF ADVERTISING AND PREFERENCES ON ASSOCIATED BRANDS IN THE AUTOMOTIVE MARKET

THE EFFECTS OF ADVERTISING AND

PREFERENCES ON ASSOCIATED BRANDS IN

THE AUTOMOTIVE MARKET

An Empirical Investigation of a New Brand Choice Predictor

June 22, 2015

B.C.T.C. Peulen

MSc Marketing, Intelligence

THE EFFECTS OF ADVERTISING AND

PREFERENCES ON ASSOCIATED BRANDS IN

THE AUTOMOTIVE MARKET

An Empirical Investigation of a New Brand Choice Predictor

June 22, 2015

1st Supervisor: dr. ir. M.J. Gijsenberg

2nd Supervisor: dr. P.S. van Eck

Student: B.C.T.C. Peulen

MSc Marketing, Intelligence

Student number: S2582554

University of Groningen

Faculty of Economics and Business

Department of Marketing

Nettelbosje 2, 9747 AE Groningen, The Netherlands

ABSTRACT

A magnitude of literature has reviewed the associations consumers have with specific brands. Several well-known associations include consumers' view of Mercedes as a classy brand or Audi as a more sportive brand. This study will extend the current literature by introducing related car brands as a new association affecting other car brands. More specifically, this study reviews the effects of car brand A’s advertising and preference on car brand B’s preference in which car brands are related through similar country-of-origins and/or by sharing components. Hence, this study examines whether associated car brands affect each other in terms of advertising and preference. Additionally, two moderators: awareness of the dependent car brand and car brand category, are included to clarify the dynamics of the car brand associations. By using four consecutive analyses this study established that over a period of four years 77 out of 94 associated car brand combinations do not affect each other in terms of advertising and preference. However, 17 specific car brand combinations affect each other in varying degrees and signs. For instance, whereas Fiat's advertising has a negative effect on Alfa Romeo's preference, BMW’s advertising has a positive effect on Volkswagen’s preference. Moreover, when two associated car brands have similar country-of-origins, the effect of advertising and the effect of preference on an associated car brand's preference declines. These results imply that academics and marketing practitioners across the automotive industry should account for associated car brands, especially in terms of advertising and preference. After analysing and discussing the results, this study aims to provide marketing managers or the like with relevant implications for their specific car brands. That is, this study concludes by designating for 13 different car brand marketers in terms of advertising and/or preference a suggestion as to the best course of action: avoid, diminish, exploit or benefit from the effect of an associated car brand.

- 1 - 1. INTRODUCTION

Branding is the act of impressing a product, service, or business, on the mind of a consumer or set of consumers (Vaid and Campbell, 2003). For instance, Ford has offered dozens of vinyl wrap tattoos for buyers to use for personalizing their ford focus cars and BMW uses alphanumeric brand names to denote technological sophistication (Hoyer, Maclnnis, Pieters, 2012). These examples amplify the wide variety of branding techniques currently being employed by car brand marketers.

Marketing literature has conducted much empirical research in the field of brand and line extensions, and ingredient branding impacting consumer evaluations (e.g. Loken and John, 1993; Park, Jun, and Shocker, 1996; Simonin and Ruth, 1998; Völckner and Sattler, 2006). In contrast, the majority of marketing academics have not considered the precise impact of component sharing of automobile brands on consumer evaluations (Verhoef, Pauwels, and Tuk, 2012). At first, component sharing seems unrelated to these aforementioned constructs, however, all constructs have in common that in one way or another two brands are associated. Component sharing is applied in many industries, however, the automotive industry is particularly known for its use of component sharing (Verhoef, Pauwels, and Tuk, 2012). In the automotive industry, US car companies spent more than $50 billion on marketing in 2000, more than any other US industry. These figures resemble marketing costs of $2.900 per car sold (Chatterjee et al. 2002), indicating the magnitude and importance of marketing in the automotive industry. On the contrary, the truth is that the global auto market is now full of identical or near-identical cars sold under different brand names and at different prices (Fingleton, 2013), which is the direct result of the popularity of component sharing in this particular industry (Fisher, Ramdas, and Ulrich, 1999). For example, General Motors offered several models of cars based on a common platform and was roundly criticized for its look-alike car line-up (Desai et al., 2001). While platform sharing is increasingly common, the link between product development strategies (such as platform sharing) and brand perceptions and equity have been largely ignored by the brand strategy literature (Levin and Levin, 2000; Strach and Everett, 2006).

- 2 - BMW and Mini, which are possibly associated with each other in consumer perceptions through component sharing.

Another example of associations between brands is their common country-of-origin, which is strongly supported as an indicator of quality products (Bluemelhuber, Carter, and Lambe, 2007). Therefore, in addition to the sharing of components associations this thesis will contribute to the existing literature by including another form of associations between car brands: the country-of-origin association. For instance, Nagashima notes already in 1970 that Japanese respondents perceive Germany to be particularly good in the manufacture of luxury automobiles. Because consumers tend to hold stereotyped images of products made in different countries (Okechuku, 1994), this thesis questions whether these associations spillover to other country-related car brands. Consequently, this study is two-folded as it majorly studies two types of associations: through component sharing and through having a similar country-of-origin.

Whereas brand extensions – established brand names launching new products (Dens and Pelsmacker, 2010) – have (partly) similar brand names as their parent brand, most car brands are independent brands, merely connected to other car brands through financial or manufacturing configuration. Within the brand extension context consumers rely on parent brand attributes in initial brand extension evaluation (Bhat and Reddy, 2001). Hence, if introduction of a brand extension can produce such positive spill-over effects to existing products, it can be expected that advertising of the brand extension will also have a positive spill-over effect on sales of existing products (Balachander and Ghose, 2003). Consistent with the brand extension literature, the main question surrounding this study is whether these reciprocal spill-over effects also hold within the context of car manufacturers and their associated brands?

- 3 - research is studying whether and to what extent associated car brands which (1) share components (e.g. BMW and Mini) affect each other in terms of advertising and preference. More specifically, this study examines among different car brand combinations the effects of car brand A’s advertising and preference on car brand B’s preference. Additionally, this study contributes to the existing literature by studying whether associated car brands which have (2) similar country-of-origins (e.g. BMW and Audi) affect each other in terms of their advertising and preferences. Accordingly, this thesis is majorly divided in two main parts as it studies brand associations in two different contexts: whether car brands share components or have similar country-of-origins.

This study conveys important managerial implications as this study expects to find a mixture of significant and non-significant car brand combinations. These can be helpful in determining whether and which car brands should cooperate in future alliances, joint promotions or none at all. Additionally, this paper will study (3) the moderating effects of the car brand categories and the dependent car brand awareness on the relationship between the associated, in terms of component sharing or country-of-origin, car brands.

The remainder of this thesis is organized as follows. First, by summarizing the relevant literature this study will elaborate on specific subjects, such as the impact of component sharing and advertising on associated brands. Second, a conceptual model and hypotheses are formulated based on the literature review. Third, the research design and methodology are introduced. Fourth, the results of this study are presented and discussed. Finally, this paper concludes with theoretical managerial implications and study limitations.

2. LITERATURE REVIEW

This study will first review literature concerning component sharing, and second the country-of-origin literature. Third, by using information integration and signalling theory this study aims to research why associated car brands can affect each other. Fourth, the effects of advertising and preferences of car brands in general and in other related fields are reviewed.

2.1 Associated Car Brands & Sharing Components

- 4 - For example, PSA Peugeot Citroën have implemented a common platform policy, which has the goal to reach a 60 percent level of common parts among all cars assembled on the same platform, regardless of their brand—Peugeot or Citroën (Patchong, Lemoine, and Kern, 2003). To successfully manage such common platform policy a car manufacturer has to ensure that sharing of components will not affect consumer’s preferences or evaluations of the involved car brands. However, to the best of our knowledge, the consequences of component sharing, other than increased efficiency and cost reduction, have rarely been examined. This new field of interest is only touched upon by Verhoef, Pauwels, and Tuk (2012), indicating the necessity for academics to substantially address the consequences of component sharing on car brand preferences.

Platform sharing is an increasingly popular product development method where various products and the brands that are attached to them share the same architecture, components, technologies, and service procedures (Rechtin, 2002). The benefits, leveraging high research and development costs over multiple products and achieving production efficiencies (Olson, 2009; Vehoef, Pauwels, and Tuk, 2012), of such practices have resulted in more frequent brand alliances as marketers try to capitalize on the complementary features of different brands (Levin and Levin, 2000). In contrast, the focus on short-term profits and cash flow management may hamper the creation or preservation of strong brands (Keller, 1999; Rechtin, 2002; Ward and Ryals, 2001). Although platform sharing as a way to increase efficiency is commonly used (Olson, 2008), a question arising is whether sharing knowledge and resources would also be applicable in terms of marketing activities. Delorenzo (2013, p.102) commented in his critical review of auto manufacturers using platform sharing that:

“Auto manufacturers are confusing vehicle architecture symmetries – the use of fewer, common platforms for global manufacturing efficiency – with a delusional push for the commonality of brand image wrangling. They think a common message will save money.”

2.2 Sharing Components & Brand Alliances

- 5 - Qu, and Ruekert, 1999, Simonin and Ruth, 1998). Component sharing can be considered as a specific form of ingredient branding (Verhoef, Pauwels, and Tuk, 2012). There are different forms of ingredient branding where some brands are conspicuously presented together (e.g. Apple and BMW) and some are well-known brands (e.g. Intel) concealed in a product. However, the aforementioned definition reflects merely a marketing perspective, since two or more brands connected through a common platform can also constitute a brand alliance. White (2004) illustrative notes that not that long ago, one of the magic words in the auto industry was “alliance”. Accordingly, to avoid ambiguous use of terms this paper will use the term brand alliance to imply two or more car brands which share merely components or share a common manufacturing platform. Since these principles have the same underlying reasoning: sharing of components, this study will only address car brands which share components.

Despite the fact that brand alliances offer significant competitive advantage potential (Bluemelhuber, Carter, and Lambe, 2007), Rindfleisch and Moorman (2003) argue that our knowledge about the effects of alliances on consumers has been quite limited. A key issue is how the impressions of one brand are transferred to or affected by the impressions of other brands to which it is strategically linked (Levin and Levin, 2000). Conversely, brand alliances do not always reinforce a brand’s image (Lans and Bergh, 2014), they may result in image impairment (Geylani, Innman, and Hofstede, 2008). Desai et al. (2001) illustrated this by concluding that while manufacturing costs always decline with the use of commonality, the firm’s overall profits may decline because of reduced differentiation. Moreover, relatively little is known about how consumers react to cooperative brand marketing and, critically, if and how exposure to such brand alliances affects consumer evaluations of the partner brands (Simonin and Ruth, 1998).

- 6 - development, there is little evidence that the potential threat to the equity of the brand is considered in calculating the overall financial implications of the practice (Olson, 2008).

Combining these issues with the emergence of new customer touch points and the need for car manufacturers to distinguish their vehicles in a traffic jam of similar products has given rise to the adoption of new channels and strategies (Waxer, 2012). A new marketing strategy, hence, could be to set up a true brand alliance, in which joint-marketing activities are presented to the public. Regardless of Delorenzo’s (2013) statement concerning the effectiveness of a common message, the underlying reasoning that precedes his statement is most important. A brand alliance with joint-marketing activities can only be effective when the car brands involved have reciprocal spillover effects in terms of quality, preference, and advertising.

Extensive academic literature (e.g. Balachander and Ghose, 2003; Bhat and Reddy, 2001; Dens and Pelsmacker, 2010) has studied this phenomenon of spill-over effects in the field of brand extensions, where they concluded that successful extensions positively impact choice of the parent brand and other extensions (Swaminathan, 2003). Additionally, consumers are more likely to judge the extension based on their knowledge of the more well-known parent brand (Bhat and Reddy, 2001). However, brand extensions are significantly different from car brands alliances. Whereas brand extensions share not only (partly) similar brands but also back- and front-office activities, car brand alliances merely share components or have a financial interest in each other. This reasoning leads to the question whether similar results can be found within the context of car brands. In addition to the brand extension literature, a growing stream of literature supports the idea that an alliance with a well-known reputable brand can improve consumer evaluations of perceived product quality as well as attitude toward the brand (Voss and Gammoh, 2004). Both the brand extension and brand alliance literature confirm that in the correct circumstances related brands affect each other, and spill-over quality and attitudes from one brand to another.

2.3 Associated Car Brands & Country-of-Origin

- 7 - between car brands. Whereas sharing components brands have an obvious direct link, the car brands which share a similar origin are indirectly related through their country-of-origin. As empirical evidence suggests that the country-of-origin affects consumers’ product evaluations (Okechuku, 1994), a subsequent reasoning would be that car brands with similar country-of-origins can affect each other.

These country-of-origin effects are generally defined as the impact which generalizations and perceptions about a country have on a person’s evaluations of the country’s products and/or brands (Bluemelhuber, Carter, and Lambe, 2007). The globalization of business enterprises has reached a point where it is sometimes difficult for consumers to determine with certainty the country-of-origin of a product (Okechuku, 1994). However, in a cross-national investigation Häubl (1996) finds that both the brand name and the country-of-origin were found to have a significant impact on consumers’ attitudes towards a new automobile. Consequently, consumers use the country-of-origin cue to evaluate foreign products when they are not familiar with the product’s intrinsic qualities (Lawrence, Marr, and Prendergast, 1992). This demonstrates that car brands with certain qualities and associations, for instance Peugeot and Renault, can spill-over qualities and associations to other country-related car brands. Although the country-of-origin would be the main effect affecting car brands’ evaluations, associated car brands can indirectly affect one another. To illustrate this statement, visualize when for some reason BMW has to return previously wholesaled cars. Such an event can indirectly affect other car brands with similar country-of-origins. These associations are caused by specific cognitive mental processes, which are explained in the subsequent section.

2.4 Information Integration and Signalling Theory

- 8 - integrate stimulus information with existing beliefs or attitudes (Simonin and Ruth, 1998). Stimulus information come in various forms, such as the notion that car brands are related by means of similar country-of-origin or sharing components. Thus, regarding brand alliances which share components or two car brands with similar country-of-origin, one brand certainly is presented in the context of the other and vice versa. Indicating that judgments about the brand alliance and the two individual brands are likely to be affected by prior attitudes toward each brand (Simonin and Ruth, 1998). Here, an important note is that these stimulus affecting brand alliances can be both positive and negative, resulting in an upward- or downward spiral of associations.

Additionally, signalling theory has been used extensively as a framework in branding research, because consumers use well-known brands as heuristic signal of product quality. Consumers use this signalling theory particularly in cases where it may be difficult to directly evaluate quality before purchase (Erdem, 1998; Erdem and Swait, 1998; Rao et al., 1999), which is applicable to the automotive industry. For example, the quality and status signalled by a higher-class car brand (e.g. Hyundai Excel) may become associated with its lower-class (e.g. Mitsubishi Precis) “twin” through the process of information integrations (Simonin and Ruth, 1998). Although signalling theory is common within branding, it has not been applied in assessing associated car brands. Therefore, this thesis will study whether not only information integration, but also signalling theory is a significant driver of spill-over effects between car brands.

2.5 Effects of Preference on Associated Brands

- 9 - To clarify this new antecedent imagine BMW and Mini, which are related through component sharing and having a financial interest in each other, and question whether these car brands affect each other. Thus, will one’s preference for BMW affect his/her preference for Mini, and vice versa? In a vertical line extension study, Sullivan (1990) showed that problems with one model (Audi 5000) in a line may influence the sales of another model (Audi 4000). Whereas Sullivan (1990) studied vertical line extensions, this thesis will research horizontal brand alliances.

There are few substantial results considering this subject. Verhoef, Pauwels, and Tuk (2012) suggest that component sharing may affect market shares of the involved brands, such that: higher-end brands sharing with lower-positioned brands may lose market share, while lower-end brands may gain market share when they share with higher-positioned brands. This is in line with the findings of Aaker and Keller (1990) and Bottomley and Holden (2001), who show that quality of the parent brand has been commonly identified as a key driver of extension success. Here, brand extensions originating from high-quality brands tend to be evaluated higher. In addition, although the degree of component sharing differs for different types of cars (Wells, 2001), Desai et al (2001) argue that even the best-camouflaged common components will lower perceived valuation. Especially when attributes that consumers value are involved. In contrast, studies have found that consumers are willing to pay a price premium to receive the unique features they associate with their favourite brand, even when those differences have disappeared through platform sharing (Sprott and Shimp, 2008; Sullivan, 1998).

- 10 - 2.6 Effect of Advertising on Associated Brands

Up to 90 percent of the carmakers’ spending is devoted to the first stage of the purchase process (in the form of mass advertising) or to the last (incentives and rebates) stage (Chatterjee et al. 2002). In general, brand-oriented advertising (e.g. nonprice advertising) strengthens brand image, causes greater awareness, differentiates products, and builds brand equity (Aaker, 1991; Keller, 1993). This finding is rather conflicting, because in contrast to fast moving consumer goods, in which equity is created substantially through advertising, automotive brand perceptions change primarily through consistent and sustained changes in the underlying product portfolio (Hirsh, Hedlund, and Schweizer, 2003). Using fast moving consumer goods (yoghurt and powdered detergents), Balachander and Ghose (2003) find that the advertising of brand extensions produces significant reciprocal spill-over that favourably affects the choice of the parent brand.

Besides the field of brand extensions, Erdem and Sun (2002) present evidence for advertising spill-over effects for umbrella brands, which results in advertising efficiencies (Tauber, 1981). Although reciprocal spillover effects within the fast moving consumer goods industry are empirically embedded, these results cannot be generalized in the context of durables. However, consumers generally seek out information when making complex decisions, specifically selection of an automobile (Akdeniz, Calantone & Voorhees, 2014). This suggests that advertising car brands has the primary purpose to publicly disseminate information. Hirsh, Hedlund, and Schweizer (2003) confirm this by arguing that consumers are well informed, and their opinions accurately reflect the accumulated performance of the products that are the

physical embodiment of those car brands

.

Due to these inconclusive results, it would be of interest to see what impact comparative advertising would have in pointing out a competing upscale brand’s “commoner” origins. Hence, this paper will study whether reciprocal advertising spillover effects occur between associated car brands. For instance, to which extent has the advertisement of solely BMW an effect on the preference of Mini.

Similar to preferences, this thesis will introduce two moderators affecting the relationship between, for example, brand advertising of BMW and the preference for Mini. This study argues that the level of the dependent car brand awareness and the car brand category of the two associate car brands will affect their relationship.

- 11 - 3. CONCEPTUAL MODEL AND HYPOTHESES

As already stated, this study will add to the current literature by introducing associated brands, through sharing components or having similar country-of-origins, as a new factor affecting car brand preference. Accordingly, to be more conclusive this study will also empirically study whether car brand preferences affect car brand sales. As this thesis is majorly two-folded, multiple combinations of associated car brands are possible, which will be used when analysing the results.

3.1 Sharing Components and Country-of-Origin

This study will formulate separate hypotheses for sharing components and country-of-origin as country-of-origin will have a different impact than sharing components on paired brands. For example, BMW, as a German brand, is related to Mini, a well-established British icon. However, a single combination fitting both associations is also possible, such as Hyundai and Kia. Both are situated in South-Korea and share components. In addition, this study will also discriminate between parent- and individual brand, since both will have dissimilar influence on each other. In doing so, this study is able to draw conclusions on specific car brand combinations, based on country-of-origin or sharing components as illustrated by the conceptual model in figure 1. The conceptual model is alienated in two similar parts, based on the division of country-of-origin and sharing components, and integrates the variables advertising, preference, and sales in both parts.

3.2 Advertising, Preference, and Sales

Advertising In recent Opel’s commercials, a well-known German celebrity ends each commercial with the slogan: “It’s a German”. Furthermore, Mercedes is the main sponsor of the German soccer association and uses the German national soccer team to endorse their commercial, and Renault uses a French “C’est si bon” song in their latest commercial. All three examples exploit the strongly supported notion in their advertisements that country-of-origin information serves as an indicator of quality (Bluemelhuber, Carter, and Lambe, 2007). Linking these findings with the positive reciprocal spill-over effects of advertising within the brand extension industry (Balachander and Ghose, 2003) implies the following reasoning.

- 12 -

Car brand A Car brand B Car brand A

H1A

Car brand B

H1B

H2A H2B

H4 H4

= random associated car brands /

H3 H3

= individual- and parent brand interaction Figure 1. Conceptual Model

Germany” brand. Consequently, all country-related car brands are associated in their branding and advertising. Hence, this study issues the following hypothesis:

Hypothesis 1A. Advertising of car brand A with similar country-of-origin as car brand B will

affect the preference for car brand B positively.

Platform-sharing brands were generally viewed as less unique and less honest (Olson, 2008), indicating a loss of quality for the involved brands. However, subjects who are aware of one brand in a choice set tend to choose the known brand even when it is lower in quality than other brands (Hoyer and Brown, 1990). To extend this concept, advertising is found in multiple industries (e.g. Balanchander and Ghose, 2003; Erdem and Sun, 2002) to generate communal advertising spill-over effects. Accordingly, this thesis reasons that advertising of car brands, associated by sharing components, impact each other positively.

Hypothesis 1B. Advertising of car brand A, which shares components with car brand B, will

affect the preference for car brand B positively.

Preference Even after controlling for prior brand attitudes, significant spillover effects of brand alliances on the partner brands were observed (Simonin and Ruth, 1998). Here, an

Sales Preference Advertising

Preference Sales

Moderators

 Dependent car brand awareness (H6)

 Car brand category (H5)

Country-of-origin

Preference

Sharing Components Advertising

- 13 - important note is that these spillover effects can be positive as well as negative. During the literature review this study already mentioned the process by which the country-of-origin and brand name information influence consumers’ evaluation of a new automobile (Häubl, 1996). By combining the country-of-origin influence with the construct of spillover effects this study argues that preference of a particular car brand in a country will assimilate to other car brands within this specific country. For example, one’s preference for BMW will assimilate to other car brands originating from Germany.

Hypothesis 2A. A high (or low) level of preference for car brand A with similar

country-of-origin as car brand B will lead to a high (or low) level of preference for car brand B.

Inter-brand platform sharing has generally been found to reduce the attractiveness and/or value of the brand (Olson, 2008). However, Olson (2009) finds that platform sharing within a single parent brand is not seen as a negative event for either the platform sharing products or the parent brand. This suggests that whether brands lose value through platform sharing depends on the level of sharing. Since car brands usually share components at an inter-brand level, a logical conclusion would be that car brands will affect each other negatively. However, Verhoef, Pauwels, and Tuk (2012) conclude in their main results that component sharing may affect, positively and negatively, market shares of the involved brands depending on the combination of car brands. This indicates that not only depend car brands on the level of sharing, but also on the specific combinations of car brands. Applying the signalling theory to these associated brands implies that consumers use heuristic signals from one brand to infer quality of the associated brand. These signals can be positive as well as negative and, therefore, the following hypothesis is constructed:

Hypothesis 2B. A high (or low) level of preference for car brand A, which shares components

with car brand B, will lead to a high (or low) level of preference for car brand B.

- 14 - that the same principle holds for preferences, such that a parent car brand’s preference influences an individual car brand more than the other way around.

Hypothesis 3. An associated individual car brand’s advertising and preference will affect a parent car brand preference more positively than a parent car brand’s advertising and preference will affect the individual brand preference.

Sales Since the image of an automobile brand is expected to affect demand (Sullivan, 1998), this study is able to use actual sales. This way this study can not only examine the changes in preferences for car brands, but also the actual alterations in terms of sales while studying the effects of a single car brand’s preference on its own sales. Hence, the following hypothesis is constructed:

Hypothesis 4. The preference of car brand B will positively influence the sales of car brand B in both the sharing of components and country-of-origin context.

3.3 Car Brand Category

Lans and Bergh (2014) question whether partner similarity drives alliance success (due to brand fit) or whether alliance success drives partner similarity (due to spill-over effects). This conclusion insinuates that this study should account for brand fit as it studies the spill-over effects of associated brand alliances. Brand fit refers to the consumer’s perception of brand image cohesiveness and associative consistency between the brands of the marketing alliance (Simonin and Ruth, 1998). In this study, consistency between brands refers to whether both brands are in the same product category. Verhoef, Pauwels, and Tuk (2012) argue, by using three categories: luxury, volume, and economy brands, that component sharing harms customer evaluation of the higher-end brand, while it benefits the lower-end brand. Partially based on the classification of Verhoef, Pauwels, and Tuk (2012), this study will classify the car brands in a similar manner.

- 15 - 1991; MacInnis and Nakamoto, 1991). Finally, Randall, Ulrich, and Reibstein (1998) argue that when a brand attempts to extend the offerings into similar categories, brand associations carry over to the extensions, as opposed to different categories where association do not carry over to the extensions. This study argues that this concept (i.e. brands within the same category affect each other more than when they are dissimilar categories) holds also for associated car brands.

Hypothesis 5. Associated car brands’ advertising and preferences, through having similar country-of-origin or sharing components, within the same category will have greater spill-over effects than associated car brands’ advertising and preferences in different categories.

3.4 Awareness of the dependent car brand

Bluemelhuber, Carter, and Lambe (2007) conclude that when consumers are familiar with the brands, the brand effect of fit is stronger than that of country-of-origin fit. However, when consumers are unfamiliar with a brand the effect of country-of-origin fit is stronger than that of brand fit. Lovett, Peres, and Shachar (2014) observed in their seminal brand alliance study the differences between the levels of consumer familiarity with car brands. The authors asked a single question: “To what extent are you familiar with the following brands?”, by using a 1-5 Likert scale. The results show that on the one end are Audi, Mini, and Ford with a high level of familiarity, and on the low-end are Volkswagen, Volvo, and Toyota.

- 16 -

Hypothesis 6. As a customer’s level of brand awareness of the dependent car brand (car brand B) increases, the effect of advertising and preference of the independent car brand (car brand A) on the preference of the dependent car brand increases positively.

3.5 Covariates

Country-of-origin effect has a significant impact on evaluation of cross-border alliances (Bluemelhuber, Carter, and Lambe, 2007) and should, therefore, be included in the main model as a covariate. When taking country-of-origin into account, this study is able to increase the validity of the conclusions not only between brands which share components, but also across brands which have similar country-of-origin. The second covariate is the business cycle which is a key factor influencing price and advertising. That is, compared with expansions, during contractions, consumers are less responsive to advertising and at the same time, they react more strongly to price reductions (Van Heerde et al., 2013). Additionally, the variable lagged sales is implemented to substantiate this study’s results. All three covariates are included to decrease uncertainty and accordingly increase the correctness of this study.

4. METHODOLOGY

Over more than a four-year period PanelWizard in cooperation with Kien tracked weekly 100 Dutch consumers planning to buy a new car within three years1_{. This contemporary dataset,}

which started in December 2010 and ended in December 2014, contains 45.812 consumer evaluations over 212 weeks. Since the initial dataset contained merely cross-sectional data points this study, with the aid of fellow students, recoded the dataset into time-series using the 212 weeks as reference points. The dataset encompasses, among others, questions regarding the spontaneous and assisted familiarity of 26 car brands, general demographics, car brand preferences, and mind-set metrics.

In table 1 are the distribution of demographics from 2011 to 2014 in the dataset and according to Statistics Netherlands presented. Here, one can observe that although the dataset has certainly enough variance, it misrepresents in some degree the actual demographics of the Netherlands. For instance, some age groups (i.e. < 30 and > 60) are underrepresented and the remaining age groups are overrepresented. This indicates that the dataset does not fully grasps the overall variance and distribution among the demographics. However, most variables

- 17 - Table 1. Distribution of Demographics (%)

Kien Dataset Statistics Netherlands

Gender Age Gender Age

Male < 30 30-39 40-49 50-59 >60 Male < 30 30-39 40-49 50-59 >60

52.7 13.8 21.1 24.3 23.8 17 49.51 37.7 11.17 15.5 14.17 21.4

Education Employment Education Employment

Low Medium High FT PT None Low Medium High FT PT None

24.3 43.7 32 44.4 25.3 30.3 31.8 39.6 27.6 35.7 33.2 31.1

and sub-variables contain more than enough data and are consistent with Statistics Netherlands. Hence, although the dataset deviates in some degree from Statistics Netherlands, it is at least in part consistent with the actual demographics of the Dutch people.

However, week 52 (in 2012, 2013) and week 1 (in 2013, 2014) provided no data, which were deleted and not analysed in this study. In addition, to simplify measurements across weeks, this thesis implemented week 1 of year 2011 as starting point, resulting in a total of 204 weeks across 4 years.

4.1 Data Collection

- 18 - Fifth, car brand preferences were measured by asking the participants two questions: “if you had to choose one brand, which car brand would you prefer when buying a new car?” and “And if you had to choose another brand, which car brand would be your second preference?” Obviously, participants had to select a number ranging from 1 to 28 corresponding to a car brand or to “another brand” and “I do not have a preference”. To be able to analyse these figures, these statistics were converted into relative preferences per week for all car brands. That is, all observations within a particular week were collected, summarized, and converted into relative numbers. For example, in one particular week 12 observations of a total of 100 observations preferred Volvo, implying that in that specific week Volvo had a preference of 12%. In addition, this study differentiated between the two questions, which were one’s first preference and his/hers second preference. To discriminate between the preferences this study has weighted the results, such that one’s first preference accounts for 60% and one’s second preference accounts for 40% of one’s overall preference. Accordingly, this study assumes that there is a difference between a first and second preference in terms of car brands. More specifically, this study believed that it was more important to not overestimate the difference between one’s first and second preference than to underestimate the difference. For these reasons, a subtle but realistic weight of 60/40 instead of 70/30 was chosen.

In figure 2 and 3 are the level of preference of four car brands and the advertising expenditures of four car brands in 2011 presented. The average preference per month in 2011 varies over time, however, when reviewing the long-term values preference seems to be rather stable over time. There are, though, differences between months in terms of preference, indicating factors affecting short-term preference variation. Figure 3 shows that Audi and Kia invest exponentially in advertising in specific months, whereas Ford and especially Chevrolet have a rather stable advertising expenditure across the months.

Figure 2. Level of Preference of Four Car Brands in 2011

0 2 4 6 8 10

Jan Feb Mar Apr May Jun Jul Aug Sep Okt Nov Dec

L ev el o f p ref er en ce

- 19 - Figure 3. Advertising Expenditure of Four Car Brands in 2011 (x €1000)

4.2 Combinations of Sharing Components and Country-of-Origin

The main focus of this study is to test various combinations based on sharing components and/or country-of-origin and review whether these specific combinations affect each other in terms of advertising and preference. By using various websites and information, ranging from car brands’ homepages to fan pages of car brands, and the dataset at hand, this thesis has constructed in total 47 car brand combinations. However, five combinations fall in both the sharing of components and country-of-origin category, as can be noted in appendix C. Additionally, in table 2 are the number of associations per brand presented. There are in total 74 similar country-of-origins and 30 sharing of components associations, which vary significantly across different car brands. Mini has merely 1 sharing of components association, whereas Nissan has in total 8 associations. All combinations were analysed in both ways, that is, the effect of brand A on brand B as well as the effect of brand B on brand A. Hence, 94 (47 x 2) car brand combinations were analysed.

Table 2. Number of Associations per Brand through Country-of-origin or Sharing of components Car brand COO Sharing Car brand COO Sharing Car brand COO Sharing

Alfa Romeo 1 1 Mitsubishi 6 … Renault 2 2

Audi 4 3 Mini … 1 Seat … 3

BMW 4 1 Mercedes 4 … Skoda … 3

Chevrolet 1 1 Honda 6 … Opel 4 1

Citroën 2 1 Hyundai 1 1 Ford 1 …

Dacia … 2 Kia 1 1 Suzuki 6 …

Daihatsu 6 1 Nissan 6 2 Toyota 6 1

Fiat 1 1 Peugeot 2 1 Volkswagen 4 3

Mazda 6 … 0 500 1000 1500 2000 2500 3000 3500 4000

Jan Feb Mar Apr May Jun Jul Aug Sep Okt Nov Dec

Ad ve rtis in g ex p en d itu re in €

- 20 - 4.3 Model specification

Initially, one overall model was constructed not only because it was more efficient, but also because it provides this thesis with the ability to compare combinations and eventually generalize findings. More specifically, by combining all the individual results, this study tried to establish whether main effects were present. However, due to the limitations of time-series data, major multicollinearity issues, and to generalize findings across brands, this thesis was forced to analyse the data in four consecutive stages

Additional difficulties surrounding this thesis were the major differences between car brands in general. For instance, advertising budgets among car brands are significantly different from each other. To account for these differences this thesis log-transformed all variables, including the dependent variables. By log-transforming the variables this thesis was not only able to compare all car brands with each other, but also to test whether differences between specific car brand combinations existed. All values of the log-transformed variables were non-zero or non-negative, which is a prerequisite of log-transforming values. Consequently, the interpretation of the outcomes is in terms of elasticity. That is, as X increases by one percent, Y will increase/decrease by β percent.

4.3.1 Dependent Variables

This study has two consecutive dependent variables in the order of preference (PREF) and sales (SALE) of brand B in time t. The former was measured using relative preferences per week. That way this study used numerical values, instead of the initial dataset’s nominal values. Additionally, since there were two questions related to one’s preference, this study constructed one combined preference measure by using the following equation: (Relative preference1 in

time t for brand x * 60) + (Relative preference2 in time t for brand x * 40). The latter numerical

variable was implemented by equally distributing monthly sales figures across the relevant weeks, after which they were log-transformed.

4.3.2 Independent Variables

- 21 - using data retrieved from various sites, however, the business cycle contained majorly negative figures with on the lower-end boundary a value of -44. To be able to log-transform these figures a constant of 45 was added to the business cycle across all weeks. Additionally, lagged sales effects were constructed using the Koyck transformation (Leeflang et al., 2015).

Lagged effects of marketing activities have long been recognized as one the complicating features of market measurement (Leeflang et al., 2015). Therefore, this study has included lagged sales effect as a covariate. The direct effect of sales is represented by β11 while

the retention rate λ measures the decline of sales over time. To avoid a great loss in degrees of freedom this study has implemented two lagged terms, leading to: 𝛽11𝜆1ℓ𝐿𝐴𝐺𝐺𝑡−ℓ = 𝛽11𝜆11𝐿𝐴𝐺𝐺𝑡−1+ 𝛽11𝜆12𝐿𝐴𝐺𝐺𝑡−2. To determine the retention rate λ of sales, this study has lagged

sales of each individual brand with one term and treated it as an independent variable solely affecting sales of car brand B in time t. All lagged sales variables proved to be highly significant (p < .00), indicating that lagged sales is an important predictor of future sales. After obtaining the β for each brand (table 3), the lagged sales in period t-1 and t-2 were multiplied with their respective retention rate λ and λ2_.

Table 3. Summary of Retention Rates per Brand

Brand β Brand β Brand β Brand β

Alfa Romeo .895 Kia .809 Mini .727 Opel .840

Fiat .909 Hyundai .883 Audi .764 Chevrolet .836

BMW .710 Mercedes .765 Seat .839 Citroën .831

Volkswagen .778 Ford .832 Skoda .874 Peugeot .816

Renault .804 Suzuki .865 Dacia .882 Toyota .871

Nissan .822 Mitsubishi .858 Mazda .734 Honda .921

4.3.3 Moderator Variables

Two moderators: dependent car brand awareness (AWAR) of car brand B in time t and car brand category (CATE), were included in the model. To measure the dependent car brand awareness this study used the question: “To which extent are you familiar with car brand X?”. Respondents had to fill in a number ranging from 1 to 4, where 1 suggested respondents did not know the car brand and 4 represented they knew not only the brand but also different models and designs. To restructure the data in weekly points, this thesis used the mean of each weak to indicate how familiar the respondents in a specific week were with a particular brand.

- 22 - International Council on Clean Transportation (ICCT) website. This article, named European Vehicle Market Statistics and written by P. Mock (2014), encompasses data on various car brands. Following Verhoef, Pauwels, and Tuk (2012) this study classified the car brands in: economy-, luxury-, or volume brand. However, not all car brands were represented in Mock’s (2014) article or could be classified in one distinct category. Therefore, a fourth category named “Other” was included. For these car brands no representative information was available (i.e. according to Mock’s [2014] article) or were too ambiguous to categorize. The car brands are divided in table 4. The car brands were classified according to their average price, where the luxury class ranged from €42,095 to € 37,318, the volume class from €25,613 to €21,188, and the economy class from €20,228 to €12,647. A visual representation of the classification is provided in appendix B. Based on this classification, the combinations of car brands were implemented as in similar categories (dummy = 1) or in non-similar categories (dummy = 0).

Table 4. Classification of Car Brands (€ x 100)

Economy Luxury Volume Other

Brand Price Brand Price Brand Price Brand Price Brand Price Brand

Chevrolet 18,62 _{Kia 20,2 Mercedes 42,1 VW*} 25,6 _Toyota* 21,8 _{Alfa R.}

Renault 20,3 Seat 19,7 BMW 41,7 Ford 21,8 Opel 21,8 Honda Dacia 12,6 Skoda* 20,1 Audi* 37,3 Mini 23,4 Citroën 21,5 Mazda

Fiat 16,1 Suzuki* … Peugeot 21,1 Nissan 22,9 Mitsubishi

Hyundai 19,4 Daihatsu

* based on Verhoef, Pauwels, and Tuk (2012)

4.3.4 Empirical Model

All variables mentioned in the preceding sections were implemented in four consecutive stages. However, one overall empirical model was established to visualize the analysis of the combinations of car brands, their interaction effects, and the moderation effects. This empirical model is depicted in equation (1). Since preference as the dependent variable is the main focus of this study, sales as a dependent variable has been excluded in the underlying empirical model.

- 23 - (1) 𝑃𝑅𝐸𝐹_𝐵𝑡 = 𝛼 + 𝛽1𝐴𝐷𝑉𝐸𝐴𝑡+ 𝛽2𝑃𝑅𝐸𝐹𝐴𝑡+ 𝛽3𝐶𝐴𝑇𝐸𝑡+ 𝛽4𝐴𝑊𝐴𝑅𝐵𝑡+

𝛽5(𝐴𝐷𝑉𝐸𝑅𝐴𝑡𝐶𝐴𝑇𝐸𝑡) + 𝛽6(𝐴𝐷𝑉𝐸𝑅𝐴𝑡𝐴𝑊𝐴𝑅𝐵𝑡) + 𝛽7(𝑃𝑅𝐸𝐹𝐴𝑡𝐶𝐴𝑇𝐸𝑡) + 𝛽8(𝑃𝑅𝐸𝐹𝐴𝑡𝐴𝑊𝐴𝑅𝐵𝑡) + 𝛽9𝐵𝑈𝑆𝐼𝑡+ 𝛽10𝐶𝑂𝑂𝐵𝑡+ 𝛽11𝜆1ℓ𝐿𝐴𝐺𝐺𝑡−ℓ+ 𝜀𝑡

where

α = constant ADVE = advertising

A = independent car brand PREF = preference

B = dependent car brand CATE = car brand category (0/1)

t = week 1-204 AWAR= dependent car brand awareness

λ = retention rate BUSI = business cycle

ℓ = amount of weeks COO = country-of-origin (0/1)

ɛ = error term LAGG = lagged sales

4.4 Plan of Analysis

In the first stage, this study performed multiple regressions to establish which car brands affect each other in terms of advertising and preference. These effects were accompanied by two covariates: consumer confidence index (i.e. business cycle) and lagged sales of the dependent car brand. Accordingly, in the results section this study will present all possible combinations with their associated interaction effects and significance values. Since the first stage is concerned with time-series data this study used EViews 8.0, which is commonly viewed as more accurate and reliable when analysing time-series data.

- 24 - selection was that the effects of stage one were similar to path A in a mediation analysis. Hence, path A should be at least partly significant when reviewing mediation variables.

During the third stage this study used the added Z method, following Gijsenberg (2014), to generalize findings across all car brands. Specifically the overall effects of advertising and preference were calculated and reviewed. For both advertising and preference a new Z-score and a new weighted average parameter was calculated, indicating whether or not an overall advertising or preference effect occurred.

In the fourth stage country-of-origin and car brand category were implemented as dummy variables (similar country-of-origin or similar car brand category = 1). In addition, dependent car brand awareness was added to the model. This variable contained the average car brand awareness across all weeks, which was constructed by summing and dividing average car brand awareness over all weeks. Since the moderators and covariate caused major multicollinearity issues, this study used weighted least squares. This method weighs the dependent variable (i.e. preference or advertising) according to its associated variance. Weighted least squares is a two-step approach, this study first estimated the parameters for advertising and preference, and second related these parameters to the moderators and covariate. By using weighted least squares these dummy variables were reviewed across all combinations of brands in a single regression analysis. More specifically, by using the parameters of advertising and preference of the multiple regressions in the first analysis, this study was able to examine whether the moderators and the covariate affected the effects of preference or advertising.

Table 5. Summary of Applied Methods

Stage Independent variables Dependent variable Method Program 1st PREF; ADVE; LAGG;

BUSI PREF

Ordinary Least

Squares EViews 8.0 2nd PREF; ADVE; LAGG;

BUSI SALES, mediated by PREF

Mediation Analysis

Mediate macro / SPSS 3th _{An overall Z-score and weighted response parameter Added Z method IBM SPSS 22}

4th _{COO; CATE; AWAR} Beta of PREF and ADVE,

weighted by their individual variance

Weighted Least

- 25 - 5. RESULTS

In the results section this study will present the findings across all four stages. To organize the findings and interpret them correctly, this study has used the four stages as a consecutive structure. Hence, the findings will be presented in a sequence as indicated in table 5. However, this study starts by reviewing the specification issues, which came to light after the initial analyses.

5.1 Specification Issues

First, due to the severe multicollinearity issues, this study has already improved the used methods. That is, this study applied four consecutive methods instead of the more preferred three-steps method in which the moderators were included in the first ordinary least squares. This alteration of methods has reduced the associated VIF scores to a minimum. Second, the large number of observations ensures this study with common normality standards. This is illustrated by Mordkoff (2011) who notes that as long as each sample contains a very large number of observations, the sampling distribution of the mean must be normal.

- 26 - 5.2 The first stage – Ordinary Least Squares

In this stage, this study will first present the results of the covariates in the following order: lagged sales, consumer confidence index, and lagged preference. Thereafter, the individual effects of advertising and preference on their associated car brands are presented.

5.2.1 Covariates

In table 6 are the lagged sales effects on individual car brands’ preferences presented. One can observe that certain car brands are in particular susceptible to lagged sales effect. These brands are: Alfa Romeo, Fiat, Volkswagen, Audi, Toyota, Suzuki, partially Mitsubishi, and Nissan. There are, however, major differences between the effects of lagged sales on the preference of a certain brand. That is, as the sales of Fiat will increase in the previous periods, Fiat’s preference will increase (.036) positively in the next period. In contrast, as Volkswagen’s sales increases in the preceding periods, its preference will decline in the next period. Similar results hold for Audi, and partially Suzuki. However, the other significant brands experience a positive influence of lagged sales effects. Interestingly, the magnitude of the effect and even the sign of the effect depends on the car brand combination.

The second covariate was the consumer confidence index, which significantly (p < 0.05) affected the brands Fiat, Dacia, and Opel and marginally (p < 0.10) affected Mitsubishi and Nissan (table 7). However, most car brands were not affected by the consumer confidence index. Here, an important note is that this study used car brand preference as dependent variable, which possibly is rather stable over time. Whereas Dacia, Nissan, and Mitsubishi are positively affected by the consumer confidence index, Fiat and Opel are negatively affected by the index. In simplistic words: as the consumer confidence index increases (i.e. people have more confidence in the economy, and as a result are willing to spend more money), people will prefer the brands Dacia, Nissan, and Mitsubishi more. To the contrary, as people’s confidence in the economic market increases, their preference for Fiat and Opel will decrease accordingly.

- 27 - Table 6. Lagged Sales Effects on Individual Car Brands

* p < 0.1, ** p < 0.05. Impact of lagged sales on Daihatsu’s preference is missing due to the lack of Daihatsu’s sales data. Table 7. Significant Consumer Confidence Index Values

* p < 0.1, ** p < 0.05

(β = .160) were marginally significant, Honda’s (β = .190) lagged preference had a positive significant effect on Honda’s preference. These results indicate the impassivity of the automotive market, meaning that people tend to hold similar preferences for car brands over time.

To review the first three hypotheses this study will use the significant and non-significant results provided by the multiple regressions. These results are displayed in table 8, the non-significant results are provided in appendix C. Since merely 17 out of 94 car brand combinations were significant in terms of advertising and preference, the hypotheses will be reviewed across the significant car brand combinations. Hence, when a car brand combination is significant, do these values support or reject the associated hypothesis? The following section will first review the advertising hypotheses, followed by the preference hypotheses, and lastly the individual versus parent brand hypothesis is reviewed.

Brand combination Dependent brand Brand combination Dependent brand Brand A Brand B Brand A Brand B Brand A Brand B Brand A Brand B

Alfa Romeo FIAT ,239** ,036** BMW Volkswagen -,022 -,130**

Volkswagen Audi -,111** -,071* Mercedes Volkswagen ,049 -,112**

Volkswagen Seat -,146** ,010 Opel Volkswagen -,034 -,119**

Volkswagen Skoda -,121** ,004 Toyota Mitsubishi ,190** ,109

Audi Seat -,096** -,023 Toyota Nissan ,114* ,117

Audi Skoda -,106** -,021 Daihatsu Suzuki … -,370**

Dacia Nissan -,067 ,197* Mazda Nissan -,008 ,141*

Audi BMW -,095** .023 Mitsubishi Suzuki ,155* ,180**

Audi Mercedes -,103** .066 Nissan Suzuki ,114 ,131*

Audi Opel -,103** -,031

Brand combination Dependent brand Brand combination Dependent brand Brand A Brand B Brand A Brand B Brand A Brand B Brand A Brand B

Alfa Romeo FIAT -,058 -,134** Mercedes Opel ,012 -,058**

Dacia Renault ,160** ,009 Opel Volkswagen -,049** -,027

Dacia Nissan ,184** ,069 Mazda Mitsubishi -,018 ,111*

Renault Nissan ,015 ,101* Mitsubishi Nissan -,041 ,114*

- 28 -

5.1.2 Advertising Effectiveness

The car brands with similar country-of-origins have mixed significant effects on each other (table 8). Whereas BMW has a positive advertising effect (β = .010, p < .05) on Volkswagen’s preference, Honda has a negative effect (β = -.073, p < .05) on Mitsubishi’s preference. Both a negative and a positive effect are plausible. When BMW increases their advertising spending with one percent, the preference for Volkswagen increases accordingly with 10%. However, these mixed results suggest merely partial support for hypothesis 1A, which stated that

advertising of a car brand will affect an associated car brand’s preference positively. Similar results hold for hypothesis 1B as some car brands affect other car brands positively, and again

other car brands are negatively affected. Consistent with hypothesis 1A, both positive and

negative effects are present across the car brands which share components. Although hypothesis 1B is also partially confirmed based on the significant values, one should keep in mind that most

advertising effects were non-significant.

5.1.3 Preference Effectiveness

All significant car brand combinations support both hypothesis 2A and 2B, which stated that

positive (or negative) preference for a car brand will affect an associated car brand’s preference positively (or negatively). Accordingly, in both types of associations, two car brands affect each other positively or negatively. For instance, Seat’s preference was found to affect Skoda’s preference positively (β = .193, p < .05). Similar results hold for the other way around (β = .241, p < .05). To substantiate the support for both hypotheses, all car brand combinations affect each other at least marginally (p < .10) with similar signs. When car brand A affects car brand B positively (negatively) in terms of preference, car brand B will also affect car brand A positively (negatively).

5.1.4 Individual versus Parent Brand

- 29 - Table 8. Individual Preference and Advertising Effects

* p < 0.1, ** p < 0.05. Note, the crossed figures were significant on Breusch-Godfrey tests, indicating autocorrelation. However, Volkswagen as a parent brand issued mixed results. Whereas advertising of Skoda affects Volkswagen more positively than the other way around, Volkswagen’s preference affects Audi more than the other way around. Hence, merely these results imply the influence of other variables affecting the relationship between car brands. Hypothesis 3 is, therefore, based on two different car brand relationships partially confirmed.

5.3 The second stage – Mediation Analysis

To reduce the number of mediation analyses this study used merely the effects that were significant (p < 0.10) in the first stage. Moreover, since most main effects were not significant in affecting the preference of an associated car brand, the likelihood of affecting that similar car brand’s sales is low. However, in a follow-up study, one could focus on the effects of advertising and/or preference on sales. By using a mediate macro, based on Baron and Kenny (1986), this study performed 28 mediation analyses. Each mediation was constructed as follows: Y = dependent car brand sales, X = preference or advertising of the independent car brand, M = preference of the dependent car brand, and three covariates: consumer confidence index, lagged sales in t -1, and lagged preference in t -1 of the dependent car brand. Overall, the results (table 9) showed no mediation in any car brand combination due to the non-significant B paths. That is, in no single combination was sales mediated by its preference.

Car brand combination Brand A > Brand B Brand B > Brand A Type

Brand A Brand B Preference Advertising Preference Advertising COO Sharing

Alfa Romeo FIAT -,019 -,051 -,095 -,127** X X

- 30 - Accordingly, this study can profoundly conclude that sales of a car brand is not affected nor mediated by its preference. As a consequence, hypothesis 4 is entirely rejected.

These results were confirmed by using bootstrapping, in which all mediation analyses enclosed a zero in the confidence intervals. However, similar to stage one, all (except Audi on Skoda and Mitsubishi on Honda) paths A were found to be significant. More interestingly, although no mediation occurred, 6 out of 28 indirect effects were significant (p < 0.10). Additionally, these significant indirect effects were merely advertising effects of the independent car brand on the dependent car brands’ sales. Accordingly, there seem to be significant effects of advertising of an associated car brand on another car brand’s sales.

Table 9. Summary of Mediation Analysis

Car brand combination Variable Path Bootstrap

IV DV (sales) C’ A B C LLCI ULCI

Fiat Alfa Romeo ADVER .0679** -.1183* .0060 .0672** -.009 .0072

Mini BMW ADVER .0123 .0601* .0368 .0146 -.0038 .0096

Volkswagen Audi PREF -.0457 .1837** .0101 -.0439 -.0282 .0336

Volkswagen Skoda PREF -.0517 -.3088** -.0330 -.0415 -.0080 .0345

Audi Volkswagen PREF .0060 .1114* -.0723 -.0020 -.0291 .0077

Audi Skoda ADVER -.0144 -.0444 -.0294 -.0131 -.0014 .0054

Skoda Audi ADVER .0842** .0453** -.0083 .0838** -.0076 .0065

Seat Skoda PREF .0222 .2049** -.0371 .0146 -.0222 .0048

Skoda Seat PREF -.0062 .2439** .0349 .0023 -.0052 .0250

Chevrolet Opel PREF -.0360 .0890** .0267 -.0336 -.0078 .0135

Chevrolet Opel ADVER -.0255 -.0597** .0063 -.0259 -.0105 .0093

Opel Chevrolet PREF .0681 .5641** .0250 .0822 -.0567 .0858

Renault Citroën ADVER .0327** -.0605* .0312 .0308** -.0064 .0012

Citroen Peugeot PREF .0338 .0826* .0488 .0378 -.0022 .0132

Audi BMW PREF .0464 .2306** -.0337 .0386 -.0316 .0125

BMW Audi PREF -.0354 .1222** -.0741 -.0444 -.0363 .0139

Opel Volkswagen ADVER .0708 -.0936* -.0585 .0762 -.0078 .0228

Volkswagen Opel ADVER .0082 -.0382** .0313 .0070 -.0060 .0028

Opel Audi PREF .0805 -.2819** .0597 .0637 -.0610 .0247

Opel Audi ADVER .1379* -.1338** .0618 .1296* -.0323 .0105

BMW Volkswagen ADVER .0041 .0093** -.0814 .0033 -.0025 .0005

Volkswagen BMW ADVER -.0272 .0475* -.0198 -.0281 -.0061 .0034

Honda Mitsubishi ADVER -.0343** -.0677** -.0029 -.0341** -.0053 .0059

Mitsubishi Honda ADVER -.0203* .0474 -.0063 -.0206* -.0041 .0032

Mazda Mitsubishi ADVER -.0213 -.0757** -.0171 -.0200 -.0046 .0079

Suzuki Mitsubishi PREF .0047 .1517* -.0140 .0026 -.0145 .0089

Suzuki Mitsubishi ADVER .0073 -.1194** -.0119 .0087 -.0048 .0082

Mitsubishi Suzuki PREF -.0634 .1589** -.0082 -.0647* -.0132 .0099

* p < 0.1, ** p < 0.05. Significance of the bootstrap is determined by using the 2.5th _{and 97.5}th _percentiles

- 31 - 5.4 The third stage - The Added Z method

In the previous section this study analysed the individual effects of specific brand combinations on each other. Here this thesis established that some car brands affect each other in terms of preference and advertising. To generalize these findings across all car brands, this thesis used the added Z method. Effects both across all brands and across specific subsets of brands can be evaluated by means of the added Z method (Rosenthal, 1991). This method was used by, among others, Gijsenberg (2014) and Kremer et al., (2008) to generalize individual results across all brands.

Following Gijsenberg (2014) this study calculated both a new Z-score and a weighted average response parameter. The former score was calculated by summing the associated t-values and dividing the sum by the square root of 263 (i.e. number of car brands). This calculation was performed for both preference and advertising. In general, the overall effect of preference of a car brand on associated car brands is not significant (p = 0.146193), as well as the effect of advertising on other car brands is not significant (p = 0.083332) at p < 0.05. Note that there are, as established in the first analysis, significant individual effects between specific car brand combinations. However, advertising is significant at p < 0.10, indicating a marginal overall effect of advertising on associated car brands. To confirm these statistics, this study calculated the weighted average response parameter for both preference and advertising.

Specifically, the following equation was used:

𝑃

= [

𝑆𝐷𝑖 𝑃𝑖 ∑ 1

𝑆𝐷𝑖

]

,

parameter, SDi is the standard deviation for each parameter and Pi is the parameter for each

model i. The results (PREF = 0.004368, ADVE = -0.00591) partially confirm the non-significant effects. However, based on p > 0.10, advertising has a marginal effect of -.006. This indicates that in general advertising of a car brand has a marginal negative effect on other associated car brands.

5.5 The fourth stage – Weighted Least Squares

The last stage was used to review the effect of two moderators and one important covariate, which also can be viewed as a third moderator, on the effects of advertising and preference. That is, to what extent moderate country-of-origin, car brand category, and dependent car brand awareness the effects of advertising and preference. In the first analysis this study obtained for