EXPLORING THE EFFECT OF GOOGLE DISPLAY ADVERTISING ON BRAND EQUITY

(1)

Master Thesis Marketing Intelligence & Marketing Management

Eline Nieboer

EXPLORING THE EFFECT OF

GOOGLE DISPLAY ADVERTISING

ON BRAND EQUITY

(2)

Exploring the effect of Google display advertising on brand equity

The role of product involvement

Eline Nieboer

University of Groningen

Faculty of Economics and Business Economics

MSc Marketing Intelligence & Marketing Management

Master Thesis 26-06-2017 Eline Nieboer Sabangplein 15 9715 CW Groningen +31634797790 eline_nieboer@hotmail.com 1st supervisor: dr. P.S. (Peter) van Eck

p.s.van.eck@rug.nl

2nd supervisor:

dr. ir. M.J. (Maarten) Gijsenberg m.j.gijsenberg@rug.nl

University of Groningen Faculty of Economics and Business

(3)

MANAGEMENT SUMMARY

Companies are continuously looking for opportunities to promote their brands and products. Especially the Internet offers effective ways of advertising these days as about 3488 million individuals all over the world use it on a daily basis (ITU 2016). For this reason, online advertising has long been of marketing researchers’ interests. For example, research generally found that display advertising is very effective in terms of sales (e.g., Ghose and Todri-Adamopoulos 2016; Srinivasan, Rutz, and Pauwels 2016). However, current literature still reveals a gap as the effects of display advertising on the well-known online search engine Google has not yet been investigated. Google display advertising might directly boost sales, but an indirect effect on sales is more likely to exist (Hanssens, Pauwels, Srinivasan, Vanhuele, and Yildirim 2014). Therefore, the research question of this paper is:

What is the effect of Google display advertising on brand equity?

Literature concerning the path to purchase or customer journey is applied in this study to find explanations regarding the effect of Google display ads on brand equity. In this case, brand equity refers to spontaneous brand awareness and the number of branded search terms. The effects of Google display advertising on both spontaneous brand awareness and the number of branded search terms are included in the study to account for an indirect effect. These variables are expected to have a different effect across different product categories. The investigated product categories differ in terms of product involvement, i.e., consumers’ purchase motivation and/or purchase ability regarding the specific product category (Keller 2012).

Based on data concerning a dairy brand and an energy provider, which were collected and offered by GfK, several models were tested to discover the branding effects of Google display advertising. In these models, the moderating effect of brand awareness was included, and lagged effects were accounted for. In order to test whether the proposed indirect exist as depicted in the conceptual model involving two serial mediators, the serial multiple mediator model proposed by Hayes (2013) was used.

(4)

effect of Google display advertising only exist regarding the dairy brand, which is a low-involvement product.

Brand awareness and branded search were found to play different roles in the data sets. Regarding the dairy brand, spontaneous brand awareness was found to positively moderate the effect of Google display advertising on sales. Even though Google display advertising positively affects the number of branded search terms, brand awareness is often sufficient for consumers to buy the dairy product.

In contrast, in case of high involvement, the moderating effect of spontaneous brand awareness does not exist. For the energy provider, the number of branded search terms seems to be a good measure for sales, which can be explained by the fact that consumers often follow a path to purchase before actually buying a product. Google display advertising is an effective medium in order to generate more branded search terms, which will in turn generate more sales.

Keywords: Google display advertising, (spontaneous) brand awareness, branded search

(5)

PREFACE

Almost six years ago, in September 2011, I started my bachelor studies in Business Administration at the Rijksuniversiteit Groningen. Although I always worked very hard in order to achieve good grades, I lost my motivation and interest after a while. Therefore, after completing my bachelor studies, I decided to take some extra time before starting my master studies. Mainly by attending several MARUG activities, I realized marketing was something that really interested me. Soon after starting the marketing master program in Groningen, I realized I was even more interested in the analyses that generate insights than actual marketing practice such as branding and customer management. However, as an analysist, it is important to be able to translate findings into clear insights understandable for marketing practitioners. Therefore I decided to study Marketing Intelligence as well as Marketing Management. Here I am, almost two years after starting my master studies, presenting my thesis, which is the final but above all the most prominent achievement of my time as being a student. Overall, it has been an intensive process. Especially in the very beginning, when I was extremely busy organizing the MARUG Conference, I experienced hard times actually starting the project. However, with the help and support I received from my supervisor, I got on track quite fast. I just realize by now that, apart from the struggles, I actually really enjoyed analyzing the data sets and working on this paper.

Obviously, I have not been able to do this all by myself. Therefore, first of all, I would like to thank my supervisor, Peter van Eck, for being of such great help during the entire process. He gave detailed, usable feedback and hinted at some analyses during our meetings. However, what I valued most was the fact that he let me discover the data and analyses and figure it all out myself. Obviously, this has led to some struggles, but ultimately that is how I learned most of it.

Besides, my family has always supported me during the entire process. Though they understand barely anything regarding the analyses I performed, they have always helped me to achieve my goals, supported my decisions, and showed their faith in me. This did not only apply to this thesis in particular, but has been true for my entire studies in Groningen. They helped me to become the person I am today, for which I am very grateful.

(6)

1. INTRODUCTION

Google dominates the online search market. In 2016, the search engine enjoyed a market share of 85.82 percent of all searches in the U.S. (Southern 2016). However, Google is no longer just a search engine, rather, it is another online media channel for brands to advertise their products. And it is found to be an effective one; customers acquired through Google search advertising reveal a higher transaction rate than customers acquired through other media channels (Chan, Qu, and Xie 2011). Because of the many advantages and opportunities in terms of advertising, companies promote their products and brands on the search engine. In terms of advertising, Google offers no longer solely Google Search Advertising (also known as SEA). Nowadays, Google Shopping and Google display advertising are also very-well known among brands and are used frequently.

Display advertising in general has been a popular topic in marketing research. For example, Ghose and Todri-Adamopoulos (2016) focused on the impact of display advertising on online consumer behavior. They found that, after consumers are exposed to display advertising, they engage in passive search as well as in active search. Hoban and Bucklin (2015) concluded that display advertising results in more consumers visiting the firm’s website. Overall, display advertising has repeatedly been proved to increase consumers’ propensity to make a purchase (e.g., Ghose and Todri-Adamopoulos 2016; Srinivasan, Rutz, and Pauwels 2016).

The type of website on which a banner ad is shown, affects the effectiveness of the ad. For example, Auschaitrakul and Mukherjee (2017) focused on the differences between commercial websites like Amazon and social websites such as Facebook. They found that online display advertising on commercial websites are more effective in terms of attitudes towards the ad and brand, compared to comparable advertisements on social websites. However, the effects of Google display advertising in particular remain unclear. Therefore, it is interesting and valuable to investigate the effectiveness of Google display advertising. This paper focuses on the direct effect of Google display advertising in terms of sales, but also on the indirect effects, i.e., the effects related to brand equity.

(9)

2 else about the brand in a category)” (Keller 2012, p. 74). Literature regarding the effectiveness of online advertising among different product categories is contradicting. For example, Yoon and Kim (2001) found that the Internet is very well-suited for products with high involvement. However, Lecinski (2011) states that consumers perform online research for any kind of product, and thus low-involvement products also involve online consumer research. Because of the contradicting findings, it is valuable to investigate the effects of Google display advertising for both low-involvement products and high-involvement products and compare these effects.

The main research question that will be addressed in this paper is:

What is the effect of Google display advertising on brand equity?

In this question, brand equity refers to both brand awareness and brand preference expressed by branded search terms. In order to examine the effect of Google display advertising, its direct effect on sales needs to be discussed. However, as (Google) display advertising has repeatedly been proved to have a positive effect on sales (e.g., Ghose and Todri-Adamopoulos 2016; Srinivasan, Rutz, and Pauwels 2016), no sub question or hypothesis regarding this effect will be included in this paper. More interestingly, according to literature, an indirect branding effect is more likely to exist. Therefore, the branding process that follows after exposure to Google display advertising will be presented and discussed extensively. Brand awareness might serve as a mediator in the branding process as well as a moderator with respect to the direct effect of Google display advertising on sales. Also branded search might play a role in the process and will be included as a mediator as well. Therefore, the following sub questions will also be addressed:

What is the effect of Google display advertising on brand awareness? What is the effect of Google display advertising on branded search?

What is the effect of brand awareness on branded search?

What is the moderating effect of brand awareness regarding the direct effect of Google display advertising on sales?

(10)

3

How does the branding effect of Google display advertising differ according to product involvement?

In order to answer the main research question and the sub questions, a linear additive model was generated as this type of model allows to test for specific moderation effects, in this case the interaction between Google display advertising and brand awareness. In order to investigate the branding effect through brand awareness and branded search, a serial multiple mediator model with two mediators (Hayes 2013) was composed.

By revealing the effectiveness of Google display advertising on brand equity and sales, and its difference between product categories, a gap in current literature will be filled. The managerial contribution of this paper includes enabling managers to use Google display advertising as media channel for the right purpose and adjust the content of the banner advertisement accordingly.

As stated before, the effects of Google display advertising are expected to differ between product categories. Therefore, in order to test Google display advertising’s effectiveness, two product categories were investigated. Several models were tested and the results were compared in order to draw conclusions regarding the role of product involvement. The data of a dairy brand and an energy provider was collected by GfK in the period 2009-2011.

(11)

4

2. LITERATURE REVIEW

In this chapter, current literature regarding the main research question and the sub questions is presented and reviewed. Based on these insights, hypotheses will be formulated. The first three sections of this chapter focus on the path to purchase, or the process after consumers are being exposed to Google display advertising. After this path/process is discussed, these effects regarding two different brands (a dairy product and an energy provider) will be discussed based on the level of product involvement, and hypotheses regarding expectations about the differences will be formulated. Lastly, a conceptual model will be generated in order to graphically present the hypotheses.

2.1 The effect of Google display advertising on sales

In this paper, Google display advertising is defined as “display ads that Google visitors see alongside other content in the Google search engine. These display ads are graphic images that can vary in size, shape, animation, duration, and more” (Goldstein, Suri, McAfee, Ekstrand-Abueg, and Diaz 2014, p. 742). In Figure 1, several examples of Google display advertisements are presented. These ads are shown next to the results of the search terms entered in the online search engine and can have different positions and sizes.

Figure 1:Google display advertisements (http://adhance.nl/display.php).

(12)

5 Transactions route

2.2 The role of brand awareness

Brand awareness is expected to serve different functions. First of all, it might mediate the effect of Google display advertising on sales. However, it might also moderate the direct effect of Google display advertising. Each of these functions are discussed in the following subsections.

2.2.1 Brand awareness as mediator

A direct effect of marketing actions on brand sales, also labeled as “transactions route” is expected to exist (see Figure 2). However, consumers are more likely to follow a “mind-set route” (Hanssens et al. 2014). In this case, consumers follow a certain path to purchase, also known as customer journey, before actually purchasing a product. First, consumers need to be aware of the existence of a brand before even considering buying the product. This stage is also known as cognition or learning, and results from marketing actions such as advertising and promotions. Brand awareness is a measure that can be used in this stage. “Brand awareness is related to the strength of the brand node or trace in memory, which we can measure as the consumer’s ability to identify the brand under different conditions” (Keller 2012, p. 72). By increasing brand awareness, a brand might become a member of the consideration set. This set exists of a handful of brands that will receive serious consideration for purchase (Keller 2012). In the next stage, affect or feeling, consumers will start to like the brand and prefer it over other brands. Lastly, affect might result in consumer behavior in the form of a purchase, which is labeled as conation (Srinivasan, Vanhuele, and Pauwels 2010). Several researchers found that sales can be explained by mapping the customer journey or path to purchase. This journey starts with awareness and knowledge-building, flowing to consideration and liking or preference which will ultimately result in a product purchase (Hanssens et al. 2014; Srinivasan, Vanhuele, and Pauwels 2010).

Figure 2: Mind-set route versus transaction route (Hanssens et al. 2014, p. 537).

(13)

6 Marketing channels can affect consumers in cognitive, affective, and conative ways. Advertisements are most effective in terms of cognition as it informs consumers about a brand, what it offers and why it is special. In turn, this will increase consumers’ brand awareness (Bolton and Saxena-Iyer 2009). More specifically, Rutz and Bucklin (2011) found that paid search, which is comparable but not similar to Google display advertising, will confront consumers with information that is related to the brand. In turn, this exposure will lead to (greater) awareness of relevance of the specific brand. Awareness of relevance holds that the consumer will connect the brand to his or her problem, and ultimately the searcher believes that the brand provides a solution to the problem. Even though Google display advertising is not completely similar to paid search advertising, it shows comparable characteristics since both Google display advertisements and paid search advertisements are shown to consumers who are entering specific search terms in an online search engine. Therefore, it is interesting to investigate whether these effects also hold for Google display advertising.

According to Song (2001), click-through rates should not be the marketing metric for managers to focus on as he found only little correlation between click-through rates and sales conversion rates. His study revealed that consumers who saw online ads generated 10 percent more sales compared to consumers who did not see any ads on the Internet. Furthermore, this study shows that this effect is not a direct effect. Rather, an “awareness conversion” effect exists which is greater than the conversion after click-throughs (also known as direct-response conversions). This “awareness conversion” effect holds that sales is indirectly affected by increasing brand awareness. Lastly, the study reveals that online banners are very effective during the branding process, since online banner advertising increases brand awareness by 6 percent.

(14)

7 As can be concluded form the previous paragraphs, Google display advertising is expected to increase brand awareness. In turn, brand awareness might positively affect sales. In some cases, brand awareness is sufficient in order to directly generate a favorable consumer response or behavior, for example the purchase of a product. This is especially the case in low-involvement decisions, “when consumers lack either purchase motivation or purchase ability” (Keller 2012, p. 74). In these instances, consumers base their choices on familiarity (Keller 2012). Even though sales conversion is found to be higher in later stages of the path to purchase, awareness does have a positive effect on sales. By increasing advertising awareness by 10 percent, sales conversion increases by 3 percent (Srinivasan, Vanhuele, and Pauwels 2010). As advertising awareness slightly differs from brand awareness, it is interesting to test whether this effect also exists for brand awareness.

To conclude, the effect of Google display advertising on sales is expected to be, at least partially, mediated by brand awareness. As previous research mainly focused on effects of general online display advertisements, the specific effect of Google display advertising is interesting and relevant to investigate. Based on current literature, the following hypothesis can be formulated:

Hypothesis 1: Brand awareness partially mediates the effect of Google display advertising on sales.

2.2.2 Brand awareness as moderator

(15)

8 visitors purchasing a product (Baye, Santos, and Wildenbeest 2015). Therefore, it might be expected that the higher brand awareness, the greater the chance a consumer will click on the Google display advertisement and ultimately purchases a product.

In order to test this expected positive moderator effect of brand awareness, the following hypothesis is formulated:

Hypothesis 2: Brand awareness positively moderates the effect of Google display advertising on sales.

2.3 The role of branded search

Brand awareness may be sufficient in order to generate a favorable consumer response or behavior, but this happens to be true in only a few instances. Rather, in most cases, brand associations are crucial and thus more important in determining responses or behavior (Keller 2012). In this paper, entering branded search terms is a representation of consumers revealing brand liking or preference. Google display advertising is expected to increase branded search in two different ways; indirectly via brand awareness, and directly after exposure to the display advertisement.

2.3.1 The mediating effect of branded search in the effect of brand awareness on sales

After brand awareness is increased by means of Google display advertising, consumers will most likely move to the second stage of the path to purchase in order to actually buy a product. In this stage, consumers will start to like the brand and prefer it over other brands (Srinivasan, Rutz, and Pauwels 2010). Consumers might search more specifically for the branded product (possibly at a later point in time). This progress in the path to purchase can occur by entering the brand name as search term in an online search engine such as Google. One way to measure this is by means of branded search, which can be defined as “users entering brand-related keyword searches in order to research the brand’s offering” (Joo, Wilbur, and Zhu 2016; Rutz and Bucklin 2011). By means of entering branded search terms, consumers show preference for a certain brand.

(16)

9 her intention to research the brand and its offerings more closely, and ultimately may or may not choose to purchase the brand’s product. As entering branded search terms might be a signal that a consumer seriously considers purchasing a brand’s product, the probability to actually purchase the product increases. Furthermore, the authors show that branded search flows from generic search, and not the other way around (i.e., generic search occurs first, followed by branded search) (Rutz and Bucklin 2011).

Sales conversion increases by 6 percent if brand liking is increased by 10 percent (Srinivasan et al. 2010). Since branded search is a possible indicator of brand liking as the second stage in the path to purchase – affect or feeling – which is the final stage before a product purchase, the following hypothesis can be formulated:

Hypothesis 3: Branded search partially mediates the effect of brand awareness on sales.

2.3.2 The mediating effect of branded search in the effect of Google display advertising on sales

As stated before, marketing channels can affect consumers in cognitive, affective, and conative ways. Even though advertisements are most effective in cognitive terms by increasing brand awareness, advertising can also have affective results by directly improving brand liking and preference. By informing consumers about a brand, and why it is special, it enhances brand preference (Bolton and Saxena-Iyer 2009). Furthermore, Park, Roth, and Jacques (1988) found that advertising stimulates formulation of consumers’ product preferences and positively affects customers’ evaluation of a brand. Besides, Ghose and Todri-Adamopoulus (2016) found that consumers engage in active search after being confronted with display advertising; they are gathering information by means of search engines. One possible way to gather information is by entering branded search terms. As display advertising has a lot of variations, it is interesting to investigate whether this effect also exists for Google display advertising.

In turn, as stated before in Section 2.3.1, brand liking increases sales conversion (Srinivasan, Rutz, and Pauwels. 2010). Therefore, the effect of Google display advertising on sales is expected to be mediated by branded search:

(17)

10 2.4 The role of product involvement

A great number of researchers has focused on the multichannel environment, cross-channel effects and different functions of available channels. According to Verhoef, Neslin, and Vroomen (2007), the Internet is often used by consumers as a mean to search for brands, their offerings and other relevant information. However, consumers often visit the offline store in order to actually purchase the product. Therefore, it is possible that consumers might click on an Google display advertisement, then visit the retailer’s website (either via a direct entry of the URL, entering a generic or branded search term in a search engine, or clicking on an online advertisement), and ultimately visit the offline store in order to purchase the advertised product.

When comparing the roles of channels among different product categories, literature is contradicting. For example, Yoon and Kim (2001) found that factors affecting Internet as channel are more closely associated with two dimensions of consumer characteristics, namely involvement and affective/rational orientation. “Low involvement results when consumers lack either purchase motivation (i.e., they don’t care about the product or service) or purchase ability (i.e., they don’t know anything else about the brand in a category)” (Keller 2012, p. 74). The Internet is very well-suited for highly involved products and it is a channel that performs very well for rationally oriented consumers as it is able to fulfill their information needs, whereas TV is a channel that is more useful for low-involvement products. Therefore, one could expect Google display advertising (as a form of communication via the Internet) to be more effective for energy providers. On the other hand, Lecinski (2011) states that consumers perform online research for any kind of product category, ranging from houses and healthcare to Band-Aids and ballpoint pens. This can be explained by the fact that consumers want to learn about all products they use, regardless the purchase price of the product (Lecinski 2011). This finding would imply that Google display advertising might be effective for both product categories, and that consumers will enter branded search terms in online search engines for both dairy products and energy providers.

(18)

11 As purchasing a service of an energy provider is more costly and thus increases consumer purchase motivation, this product category involves higher product involvement. Therefore, consumers are less likely to make irrational, affective decisions at the moment of purchase. Rather, they would extensively research offerings of the available brands in the product category. In case of high product involvement, the relative weight of attitudes and intentions becomes more important in the path to purchase (Kokkinaki 1999). If product involvement is high, a brand has to positively change consumers’ attitudes before a product can be sold, and attitudinal metrics are affecting sales conversion.

On the other hand, in case of low product involvement, a well-formed attitude may not even be necessary for a purchase to occur and consumers are less likely to spend effort in forming their intentions. In these instances, consumers’ purchase decisions are based on heuristics such as the brand being available or being promoted. In these settings, only a minimum of brand awareness may be sufficient in order for a consumer to buy a certain brand (Keller 1993; Keller 2012). Therefore, brands do not need to change consumers’ attitudes in order to sell low-involvement products (Hanssens et al. 2014). To conclude, consumers are expected to follow the “transaction route” in case of low product involvement, whereas a “mind-set route” will be followed if a product involves high involvement (see also Section 2.2.1 and Figure 2).

Due to the fact that consumers have higher purchase motivation regarding energy providers and attitudinal metrics are required before sales conversion in this product category, the mediating role of both brand awareness and branded search are expected to be stronger for high-involvement products compared to low-involvement products. The following hypotheses can be formulated:

Hypothesis 5: The mediating effect of brand awareness in the relationship of Google display advertising and sales is stronger for high-involvement products.

Hypothesis 6: The mediating effect of branded search in the relationship of Google display advertising and sales is stronger for high-involvement products.

(19)

12 Google display advertising H1 H2 (+) H3 H1 H3, H4 H4

Brand recognition in the store is especially important when the level of in-store decision-making in the product category is high (Keller 2010). Besides, consumers are more likely to use a heuristic such as “I choose the brand I know” and buy familiar, well-established brands in case the product involvement is low in order to minimize the decision-making process (Keller 1993; Keller 2012). Therefore, the moderating effect of brand awareness on the effect of Google display advertising and sales is most likely to exist for the diary product category and is expected to be higher than the moderating effect of brand awareness for the energy product. The following hypothesis can be formulated:

Hypothesis 8: The moderating effect of brand awareness in the relationship of Google display advertising and sales is stronger for low-involvement products.

2.5 Conceptual model

The three-staged path to purchase is represented in the conceptual model by three variables, which results in a serial multiple mediator model with two mediators (see Figure 3). The first stage, cognition, will be measured by means of brand awareness. Next, affect or liking as the second stage in the path to purchase, is represented by the number of branded search terms. Lastly, after completing the first two stages, consumers will show some kind or conation or behavior, in this case by means of a purchase (sales volume). However, there is a possibility that consumers skip both stages in the path to purchase. Therefore, the direct effect of Google display advertising on sales is accounted for. For these instances, brand awareness might affect the strength of the direct effect and is thus included as a moderator in the model. Furthermore, Google display advertising might positively affect brand liking and preference. Therefore, branded search is included as a mediator between Google display advertising and sales. However, branded search might be skipped implying a partially mediating effect of branded search and thus a direct effect of brand awareness on sales, which is most likely to exist in case of low-involvement products.

Figure 3: Conceptual Model.

Sales Branded search

(20)

13

3. RESEARCH DESIGN

In this chapter, the methodology of the research will be discussed. First, the data offered by GfK will be elaborated upon. Besides, the research type and the variables included in the model will be described. The model choice will be discussed, and the model will be specified by including the interaction effect and lagged effects. Next, the serial multiple mediator model proposed by Hayes will be explained. This chapter will conclude with a plan of analysis.

3.1 Description of the data

In order to empirically examine the hypotheses proposed in the literature review, two data sets will be used. The first data set concerns data from a dairy product, whereas the other data set contains data from an energy provider. Both data sets contain the same variables which are measured in an equal way, allowing comparison between the two brands. The data is collected and offered by GfK, which collects market and consumer information in order to deliver relevant insights with regard to two complementary sectors, namely consumer choices and consumer experiences (GfK, 2017).

The data in the data sets are quantitative, aggregate demand data, “which refers to the demand across a sample of customers or households that can be measured at levels such as store, chain and market demand” (Leeflang, Wieringa, Bijmolt, and Pauwels 2015, p. 223). More specifically, the data is aggregated on week level, and the periods of the two data sets overlap each other for a great period. The data set of the dairy brand covers data collected over a period of two years and is collected from the first week of 2009 until the last week of 2010. Therefore, this data set contains 105 weeks (2009 had 53 weeks, whereas 2010 had 52). The data set from the energy provider contains more weeks, namely 115 weeks, as data is collected from week 29 in 2009 until week 38 in 2011.

(21)

14 The main goal of this study is descriptive, i.e., to describe the relationships between the variables included in the models, rather than predicting the volume of sales. The following variables are included in the model: week, sales (volume), spontaneous brand awareness, the number of brand search terms, and Google display (cost of ad). Spontaneous brand awareness is “a measure of saliency, or share of mind when cued by the product” (Kapferer 2005, p. 17). In this case, “consumers are asked, without any prompting, to name the brand they know, even if only by name, in the product category” (Laurent, Kapferer, and Roussel 1995, p. 170). Spontaneous brand awareness can be considered as a form of brand recall, which can be defined as “consumers’ ability to retrieve the brand when given the product category, the needs fulfilled by the category, or some other type of probe as a cue” (Keller 1993, p. 3). Spontaneous brand awareness differs from brand recognition, which relates to “consumers’ ability to confirm prior exposure to the brand when given the brand as a cue” (Keller 1993, p. 3). Next to the existing variables, several new variables will need to be computed. First of all, an interaction term of spontaneous brand awareness and Google display advertising will be created in order to test the proposed moderation effect of spontaneous brand awareness (see Section 3.3.1). Besides, the lagged effects of Google display, spontaneous brand awareness, the number of branded search terms, and the moderation effect of spontaneous brand awareness on sales will be included. These effects will be discussed more extensively in Section 3.3.2.

3.2 Model choice

In order to test the hypotheses, a direct brand sales model is used. In such a model, “sales of brand j are explained as a function of marketing variables of brand j, marketing variables of competing brands, and environmental variables” (Leeflang et al. 2015, p. 230). The model composed is linear additive, rather than multiplicative. As the moderating effect of spontaneous brand awareness on the effect of Google display advertising on sales is of particular interest, an additive model is preferred. A multiplicative model would include interaction effects of all variables (Leeflang et al. 2015), which would make it difficult to interpret the specific moderation effect of spontaneous brand awareness. Therefore, in order to test whether the moderating effect of spontaneous brand awareness exists, the linear additive model will be extended by adding the product of the variables Google display advertising and spontaneous brand awareness.

(22)

15 sets will need to be composed. Based on these models, conclusions will be drawn regarding the moderation and mediation effects, i.e., hypotheses 1 to 4 will be tested. Then, the effects of the two different models will be compared in order to draw conclusions regarding hypotheses 5 to 8.

3.3 Model specification

The basic model includes the independent variables cost of Google display advertising, spontaneous brand awareness, and the number of branded search terms. Model 1 can be specified as follows:

(1) 𝑆𝑆𝑡𝑡= 𝛽𝛽1𝐺𝐺𝐺𝐺𝑡𝑡+ 𝛽𝛽2𝐵𝐵𝐵𝐵𝑡𝑡+ 𝛽𝛽3𝐵𝐵𝑆𝑆𝑡𝑡+ 𝜀𝜀𝑡𝑡

Where,

𝑆𝑆𝑡𝑡 = volume of sales in week t;

𝐺𝐺𝐺𝐺𝑡𝑡 = Google display cost of ad in week t;

𝐵𝐵𝐵𝐵𝑡𝑡 = spontaneous brand awareness in week t;

𝐵𝐵𝑆𝑆𝑡𝑡 = number of brand search terms in week t; 𝜀𝜀𝑡𝑡 = the error term.

3.3.1 Testing moderation

A moderator is a variable which has an impact on the direction and/or strength of the relation between an independent variable and a dependent variable (Baron and Kenny 1986). In case a moderation effect is present, an interaction effect between the independent variable and the moderator exists. By including an interaction effect between Google display advertising and brand awareness, the moderation effect is included in the second model. This interaction term will be created by computing a new variable which is a multiplication of the variables cost of display ad and spontaneous brand awareness for each case. In order to test whether the moderation effect is significant, this model will be compared to the basic model on the basis of model fit. Model 2 can be specified as follows:

(2) 𝑆𝑆𝑡𝑡 = 𝛽𝛽1𝐺𝐺𝐺𝐺𝑡𝑡+ 𝛽𝛽2𝐵𝐵𝐵𝐵𝑡𝑡+ 𝛽𝛽3𝐵𝐵𝑆𝑆𝑡𝑡+ 𝛽𝛽4(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡+ 𝜀𝜀𝑡𝑡

where

(23)

16 𝐺𝐺𝐺𝐺𝑡𝑡 = Google display cost of ad in week t;

𝐵𝐵𝑆𝑆𝑡𝑡 = number of brand search terms in week t;

(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡 = the interaction (moderating) effect of Google display advertising and brand

awareness in week t; 𝜀𝜀𝑡𝑡 = the error term. 3.3.2 Lagged effects

Most marketing models contain lagged effects. Such effects can have two causes. First of all, delayed response effects may be present, which can be classified into four sources; execution delays, noting delays, purchase delays, and recording delays. In this case, especially purchase delays are relevant, where a consumer is motivated by an ad but is not in a purchase situation until a considerable time afterward (Kotler 1971). On the other hand, lagged effects can be caused by customer holdover effects, which occur “when the marketing mix change creates new customers or increases the purchase rate of current customers and some of this increase lingers through subsequent periods” (Doyle and Saunders 1985, p. 54). Both causes might affect the volume of sales for any of the brand being studied.

As lagged effects are likely to be present in the data sets used in this study, these effects need to be accounted for in the model. In order to so, a finite distributed lag model will be composed. However, there is no indication of the lasting effect of the variables. Therefore, two different models are composed in order to test how many weeks the effects last. Model 3 contains the lagged effects of one week, whereas Model 4 contains lagged effects of two weeks as well. Model 3 can be specified as follows:

(3) 𝑆𝑆𝑡𝑡 = 𝛽𝛽1𝐺𝐺𝐺𝐺𝑡𝑡+ 𝛽𝛽2𝐵𝐵𝐵𝐵𝑡𝑡+ 𝛽𝛽3𝐵𝐵𝑆𝑆𝑡𝑡+ 𝛽𝛽4(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡+ 𝛽𝛽5𝐺𝐺𝐺𝐺𝑡𝑡−1+ 𝛽𝛽6𝐵𝐵𝐵𝐵𝑡𝑡−1+ 𝛽𝛽7𝐵𝐵𝑆𝑆𝑡𝑡−1

+ 𝛽𝛽8(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡−1+ 𝜀𝜀𝑡𝑡

And Model 4 can be specified as follows:

(4) 𝑆𝑆𝑡𝑡 = 𝛽𝛽1𝐺𝐺𝐺𝐺𝑡𝑡+ 𝛽𝛽2𝐵𝐵𝐵𝐵𝑡𝑡+ 𝛽𝛽3𝐵𝐵𝑆𝑆𝑡𝑡+ 𝛽𝛽4(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡+ 𝛽𝛽5𝐺𝐺𝐺𝐺𝑡𝑡−1+ 𝛽𝛽6𝐵𝐵𝐵𝐵𝑡𝑡−1+ 𝛽𝛽7𝐵𝐵𝑆𝑆𝑡𝑡−1

+ 𝛽𝛽8(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡−1+ 𝛽𝛽9𝐺𝐺𝐺𝐺𝑡𝑡−2+ 𝛽𝛽10𝐵𝐵𝐵𝐵𝑡𝑡−2+ 𝛽𝛽11𝐵𝐵𝑆𝑆𝑡𝑡−2+ 𝛽𝛽12(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡−2

(24)

17 where

𝑆𝑆𝑡𝑡 = volume of sales in week t;

𝐺𝐺𝐺𝐺𝑡𝑡 = Google display cost of ad in week t;

𝐵𝐵𝑆𝑆𝑡𝑡 = number of brand search terms in week t;

(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡 = the interaction (moderating) effect of Google display advertising and brand

awareness in week t;

𝐺𝐺𝐺𝐺𝑡𝑡−1 = Google display cost of ad per week in week t-1;

𝐵𝐵𝐵𝐵𝑡𝑡−1 = spontaneous brand awareness per week in week t-1;

𝐵𝐵𝑆𝑆𝑡𝑡−1 = number of branded search terms per week in week t-1;

(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡−1 = the interaction (moderating) effect of Google display advertising and brand

awareness in week t-1;

𝐺𝐺𝐺𝐺𝑡𝑡−2 = Google display cost of ad per week in week t-2

𝐵𝐵𝐵𝐵𝑡𝑡−2 = spontaneous brand awareness per week in week t-2;

𝐵𝐵𝑆𝑆𝑡𝑡−2 = number of branded search terms per week in week t-2;

(𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)𝑡𝑡−2 = the interaction (moderating) effect of Google display advertising and brand

awareness in week t-2; 𝜀𝜀𝑡𝑡 = the error term. 3.4 Testing mediation

After modeling the direct effects of Google display advertising, spontaneous brand awareness and the number of branded search, and including the moderation effect of spontaneous brand awareness, an additional model needs to be tested to investigate the expected mediation effects of spontaneous brand awareness and the number of branded search terms.

(25)

18

Figure 4: Serial multiple mediator model with two mediators (Hayes 2013, p. 145).

A mediation effect in general needs to be addressed in several steps. First, the direct effect of Google display advertising on sales needs to be addressed. If this effect is significant, a relationship between the variables is present. In case the effect is not significant, one can conclude no relationship exists and testing for mediation would be useless. If the direct effect is present, one should continue by examining the effect of Google display advertising on brand awareness (𝑎𝑎₁). Lastly, the effect of brand awareness on sales (𝑏𝑏₁) needs to be tested. In case both 𝑎𝑎₁ and 𝑏𝑏₁ are significant, a mediating effect of brand awareness is present. The original, direct effect between X and Y (𝑐𝑐) consists of the direct and indirect effect (𝑐𝑐 = 𝑐𝑐′+ 𝑎𝑎1𝑏𝑏1). In this equation, 𝑐𝑐′ captures the remaining direct effect that is not accounted for by the

mediation effect. This means that if a mediation effect is present, 𝑐𝑐′ has a lower value than 𝑐𝑐. In case 𝑐𝑐′ has any value greater than zero, partial mediation exists. However, it is also possible that a full mediation effect exists. In this case, the value of 𝑐𝑐′ becomes zero.

The same holds for the two possible mediating effects of branded search. A mediating effect of branded search in the effect of Google display advertising on sales is present if both 𝑎𝑎2 and 𝑏𝑏2 are significant, and conclusions regarding the strength of the mediator should be

(26)

19 In case two mediators are correlated, even after adjusting for X, the serial multiple mediator model with two mediators is appropriate. This model assumes that one mediator affects the other mediator (Hayes 2013). In this case, X affects Y through four pathways, which include three specific indirect effects and one direct effect, namely:

1. Indirectly from X to Y through M1;

2. Indirectly from X to Y through M2;

3. Indirectly from X to Y through both M1 and M2;

4. Directly from X to Y, without passing through either M1 or M2.

This results in three equations:

𝑀𝑀1 = 𝑖𝑖𝑀𝑀1+ 𝑎𝑎1𝑋𝑋 + 𝑒𝑒𝑀𝑀1

𝑀𝑀2 = 𝑖𝑖𝑀𝑀2 + 𝑎𝑎2𝑋𝑋 + 𝑑𝑑21𝑀𝑀1+ 𝑒𝑒𝑀𝑀2

𝑌𝑌 = 𝑖𝑖𝑌𝑌+ 𝑐𝑐′𝑋𝑋 + 𝑏𝑏1𝑀𝑀1+ 𝑏𝑏2𝑀𝑀2+ 𝑒𝑒𝑌𝑌

The total effect of X exists of the indirect effect of X on Y through M1 (𝑎𝑎1𝑏𝑏1), the indirect

effect through M2 (𝑎𝑎2𝑏𝑏2), the indirect effect of X on Y through both M1 and M2 (𝑎𝑎1𝑑𝑑21𝑏𝑏2), and

the direct effect of X (c’). This results in 𝑐𝑐, the total effect of X:

𝑐𝑐 = 𝑐𝑐′_{+ 𝑎𝑎}

1𝑏𝑏1+ 𝑎𝑎2𝑏𝑏2+ 𝑎𝑎1𝑑𝑑21𝑏𝑏2

The total indirect effect of X on Y is the difference between the total effect of X on Y and direct effect of X on Y:

𝑐𝑐 − 𝑐𝑐′_{= 𝑎𝑎}

1𝑏𝑏1+ 𝑎𝑎2𝑏𝑏2+ 𝑎𝑎1𝑑𝑑21𝑏𝑏2

The standard error of the indirect effect of 𝑎𝑎₁𝑑𝑑₂₁𝑏𝑏₂is equal to:

𝑠𝑠𝑒𝑒𝑎𝑎1𝑑𝑑21𝑏𝑏2 = �𝑎𝑎21𝑑𝑑212 𝑠𝑠𝑒𝑒𝑏𝑏22 + 𝑎𝑎12𝑏𝑏22𝑠𝑠𝑒𝑒𝑑𝑑221+ 𝑑𝑑212 𝑏𝑏22𝑠𝑠𝑒𝑒𝑎𝑎21

(27)

20 symmetrical. This occurs especially in small samples. Ultimately, this will yield underpowered tests of mediation. Because of Sobel’s assumption of a normal distribution, an alternative approach is necessary to test for mediation. Therefore, bootstrap will be conducted in order to test for mediation (Zhao, Lynch Jr., and Chen 2010). Bootstrap confidence intervals are calculated by repeatedly resampling from the data with replacement, estimating the model in each bootstrap sample, calculating the indirect effects, and deriving the endpoint of a confidence interval for each (Hayes 2013).

3.5 Plan of analysis

The first step in the data analysis is data cleaning, which will be performed by means of consistency checks and treatment of missing responses. Missing values will be addressed by applying the method of Expectation Maximization. “Consistency checks identify data that are out of range, logically inconsistent, or have extreme values” (Malhotra 2009, p. 461). Outliers will be detected, discussed and, if required, deleted, adjusted, or accounted for by introducing a dummy variable.

After the data sets are prepared for analysis, the data will be described by means of frequency and descriptive analyses. Graphs will be used in order to graphically present the data to ease interpretation.

Next, the four different models as described in Section 3.3 will be composed for each of the data sets. In order to compose these models, several new variables need to be created. First of all, the interaction of Google display advertising and spontaneous brand awareness will be computed by multiplying these two variables. Furthermore, the lagged effects of the independent variables Google display, spontaneous brand awareness, the number of branded search terms will be created. Also for the interaction effect, lagged variables will be created. For each of these variables, two different lagged effects will be created, which are the lagged effect of one week and the lagged effect of two weeks. By means of model comparison based on the Adjusted R-Square, a conclusion will be drawn regarding the best-performing model. The model outcomes will be presented, and significance and estimates of the parameters will be discussed.

(28)

21 After the models are composed, the models need to be validated. Here, tests for nonzero expectation, autocorrelation, heteroscedasticity, nonnormality, and multicollinearity will be performed. If necessary, i.e., in case of one or more violations of these assumptions, the causes will be addressed and the model will be re-estimated.

(29)

22

4. RESULTS

In this section, the results of the analyses will be presented. First of all, both data sets will be prepared for analyses by performing checks regarding missing values, outliers and other oddities. Next, an overview of the descriptive statistics of the variables of both data sets will be presented. Afterwards, the models as described in Section 3.3 will be composed and the model fit will be discussed. The best performing models will be estimated and the outcomes will be presented. The models need to be validated by testing violation of the assumptions regarding OLS. If needed, i.e., in case of violation of one or more assumptions, the models will have to be re-estimated. Afterwards, the final model outcomes will be presented in order to generate conclusions regarding the parameter effects and their levels of significance. This section will be concluded by testing the hypotheses presented in the Literature Review.

4.1 Data preparation

4.1.1 Diary data set

(30)

23 Besides, week 52 of 2010 in the dairy data set was presented twice in the data set. As both lines of this observation contain the exact same values and the year 2010 existed of 52 weeks (and not 53), one can assume that this might have happened accidentally. Therefore, one of these lines was deleted from the data set.

After the missing value in the data set was imputed by means of the Expectation Maximization method, outliers had to be detected and addressed. An outlier is “a unusual observation that does not seem to belong to the pattern of variability produced by the other observations” (Johnson and Wichern 1992, p. 187). A value is an outlier as soon as the case has a value that is either below (Q1 − 1.5 ∗ ICR) or above (Q3 + 1.5 ∗ ICR). The dairy data set reveals seven extreme high peaks in sales (see Figure 5). These peaks can have several explanations. For example, during these weeks, the product might have enjoyed price promotions, or the brand invested heavily in Google display advertising or any other marketing activities (either offline or online). After closely investigating the data, most of the weeks that contain outliers in sales also reveal high values for both Google display advertising (either in the current week or in week t-1) and spontaneous brand awareness. The value of these observations might therefore actually help explain the hypothesized positive effect of Google display advertising on sales, as well as the positive effect of spontaneous brand awareness on sales, and therefore these outliers are valuable values. For this reason, these values were not deleted from the data set, nor adjusted. However, two outliers seem to be caused by an external event not included in the model such as price promotion as these weeks did not reveal high cost of Google display advertising in either the current week, week t-1 or week t-2. Therefore, a dummy variable was included in de model in order to account for the external event causing these values.

Figure 5: Sales of dairy brand.

(31)

24 Furthermore, the variable spontaneous brand awareness reveals 4 extreme low values from week 10 until week 13 in 2009 (see Figure 6). The outliers represent only 3.81 percent of all values within the variable spontaneous brand awareness and did not affect model performance nor the parameter estimates and its significance. Therefore, these outliers are not treated with a dummy variable.

Figure 6: Spontaneous brand awareness of dairy brand.

Lastly, the variable the number of branded search terms contains a high number of outliers, namely 12 extreme high values (see Figure 7) which represent 11.43 percent of the total values. Therefore, it is required to investigate these outliers more closely. The extreme values for branded search terms do not help explain the relationship between cost of display and branded search terms (see Figure 8). However, as can be seen in Figure 9, they seem to help explain the hypothesized positive effect of the number branded search terms on sales volume. Therefore, these outliers are actually very useful and no dummy variable was included.

Remarkable is that in the earlier weeks of data collection, the number of branded search terms was substantially higher compared to the rest of the data-collection period (from week 23 in 2009 onwards), as can also be seen in Figure 7. However, these high numbers of branded search terms were accompanied by high levels of investment in Google display advertising until week 26 (see also Figure 13). Therefore, these high values for the number of branded search terms might help explain the effect of Google display advertising on branded search terms. For this reason, the outliers are not adjusted, nor a dummy variable was created.

0 5 10 15 20 25 30 35 20 0901 20 0906 20 0911 20 0916 20 0921 20 0926 20 0931 20 0936 20 0941 20 0946 20 0951 20 1003 20 1008 20 1013 20 1018 20 1023 20 1028 20 1033 20 1038 20 1043 20 1048

(32)

25 Figure 7: The number of branded search terms of dairy brand.

Figure 8: Curve cost of display vs branded search terms. Figure 9: Curve branded search terms vs. sales. 4.1.2 Energy data set

In contrast to the dairy data set, the energy data set did not contain any missing values. However, this data set also reveals some extreme values or outliers. More specifically, it contains one extreme low value in sales in week 25 in 2010 (see Figure 10). In this week, no substantial changes in Google display advertising, spontaneous brand awareness or the number of branded search terms were discovered. Therefore, this low peak can only be explained by means of external events, for example by competitors’ behavior, such as price promotions or advertising, or negative news of the energy provider itself in the media. As such factors are not accounted for in the model, a dummy variable had to be created to account for such external effects.

0 0,050,1 0,150,2 0,250,3 0,350,4 20 0901 20 0906 20 0911 20 0916 20 0921 20 0926 20 0931 20 0936 20 0941 20 0946 20 0951 20 1003 20 1008 20 1013 20 1018 20 1023 20 1028 20 1033 20 1038 20 1043 20 1048

(33)

26 Figure 10: Sales of energy provider.

Besides, this data set contains one extreme high value for spontaneous brand awareness (see Figure 11) that cannot be explained by an increase in Google display advertising, spontaneous brand awareness or the number of branded search terms. Therefore, this value is likely to be explained by an external event such as advertising using any media channel but Google display. However, including a dummy variable did not lead to different parameter estimates and parameter significance, and the model performance even decreased. Therefore, no dummy variable accounting for the external events explaining the outliers of this variable was included in the final model.

Figure 11: Spontaneous brand awareness of energy provider.

Lastly, the variable number of branded search terms contains three extreme high values (see Figure 12). Two of these are accompanied by high levels of Google display advertising and help explain the hypothesized effect of Google display advertising. Therefore, these values were not deleted from the data set, nor adjusted. However, the third outlier seems to be caused by an external event, for example positive online reviews or advertising using any media channel but Google display. Again, including a dummy variable that includes this

0 1000 2000 3000 4000 5000 6000 20 0929 20 0935 20 0941 20 0947 20 0953 20 1006 20 1012 20 1018 20 1024 20 1030 20 1036 20 1042 20 1048 20 1102 20 1108 20 1114 20 1120 20 1126 20 1132 20 1138

Sales (volume)

0 10 20 30 40 50 60 70 80 20 0929 20 0935 20 0941 20 0947 20 0953 20 1006 20 1012 20 1018 20 1024 20 1030 20 1036 20 1042 20 1048 20 1102 20 1108 20 1114 20 1120 20 1126 20 1132 20 1138

(34)

27 effect did not change the parameter estimates nor parameter significance, and the model performance even decreased. Therefore, no dummy variable capturing the external events explaining the outliers of the number of branded search terms was included in the model.

Figure 12: The number of branded search terms of energy provider.

4.2 Descriptive statistics

4.2.1 Dairy brand

As described before, the sales of the dairy brand reveal seven extreme high peaks, for which several explanations exist. The mean sales volume per week is 301756.06, with a standard deviation of 64503.826.

The investment in Google display advertising by this brand is highly fluctuating. Overall, the Google display advertising expenses are not too high (on average only 50.82 euros, and the maximum is 152 euros per week). As can be seen in Figure 13, Google display advertising is not a continuous investment. Rather, it seems like there are some campaigns for which Google display advertisements are used as a media channel. This occurs for example from week 7 in 2009 until week 26 in 2009, as well as from week 24 in 2010 until week 32 in 2010.

Spontaneous brand awareness of the dairy brand has a mean of 23.23, which seems to enjoy a stepwise increase over time (see Figure 6), and has a standard deviation of 4.397.

The number of branded search terms is quite low as the mean is only .12. A complete overview of the descriptive statistics of this dataset can be found in Appendix 1, Table 11.

0 0,5 1 1,5 2 20 0929 20 0935 20 0941 20 0947 20 0953 20 1006 20 1012 20 1018 20 1024 20 1030 20 1036 20 1042 20 1048 20 1102 20 1108 20 1114 20 1120 20 1126 20 1132 20 1138

(35)

28 Figure 13: Google display (cost of ad) of dairy brand.

4.2.2 Energy provider

The sales of the energy provider are more fluctuant compared to the sales of the dairy brand (mean = 2896.3913, standard deviation = 950.34219), as can also be concluded from Figures 5 and 10.

The Google display investments of the energy provider seem to follow some kind of random pattern (see Figure 14). The energy provider invests more money in Google display advertising; the maximum is 550.75 euros per week, which is in great contrast to only 152 euros per week by the dairy brand. The average spending on Google display advertising of 178.65 euros by the energy provider is even higher than the maximum spending of the dairy brand. One possible explanation for this might be that consumers often make their purchase decision regarding low-involvement products when they are in the store. Therefore, the dairy brands might possibly spend a great percentage of its marketing budget on advertising in retail stores.

Spontaneous brand awareness seems pretty consistent over time (see Figure 11) and is quite large (mean = 53.4435) compared to both the dairy brand (mean = 23.23) as well as competitors (mean = 30.8283). This might imply that the energy provider has a strong brand equity.

The average number of branded search terms per week (1.0848) is about nine times larger than the average number of branded search terms of the dairy brand. This difference might be explained by means of the difference in level of product involvement, resulting in more research in case consumers experience high product involvement. Besides, the number of branded search terms seems to slightly increase over time (see Figure 12). A complete overview of the descriptive statistics of this dataset can be found in Appendix 1, Table 12.

0 20 40 60 80 100 120 140 160 20 0901 20 0906 20 0911 20 0916 20 0921 20 0926 20 0931 20 0936 20 0941 20 0946 20 0951 20 1003 20 1008 20 1013 20 1018 20 1023 20 1028 20 1033 20 1038 20 1043 20 1048

(36)

29 Figure 14: Google display (cost of ad) of energy provider.

4.3 Exploratory analysis

Before creating and comparing models for the different data sets, the correlations within each data set are explored. The correlation coefficients of the dairy data set can be found in Table 1, whereas the correlation coefficients of the energy data set are presented in Table 2.

Regarding the dairy brand, the cost of Google display advertising is positively correlated with the number of branded search terms, which also holds for the two lagged effects of each of these variables. Besides, Google display advertising has an effect on the number of branded search terms that remains quite strong over time, as the correlations of the current Google display advertising, its lagged effect of one week and the lagged effect of two weeks reveal values close to each other (.456, .439, and .413 respectively).

Unexpectedly, spontaneous brand awareness and the number of branded search terms are negatively correlated (-.537). However, this finding can be explained by the observation that brand awareness is often sufficient for consumers to buy low-involvement products. Therefore, when consumers are familiar with a brand, they might buy it when recognizing the brand in-store. In this case, no further research regarding the brand is required (Hanssens et al. 2014; Keller 1993; Keller 2012). This negative effect becomes stronger over time, i.e., the lagged effects of spontaneous brand awareness have more negative correlations with the current number of branded search terms (-.584 for 𝐵𝐵𝐵𝐵_𝑡𝑡−1 and -.617 for 𝐵𝐵𝐵𝐵_𝑡𝑡−2).

Spontaneous brand awareness does not have a correlation with sales, but its lagged effects are, surprisingly, negatively correlated with sales. One possible explanation for this finding might be that the dairy brand has a negative, or at least a neutral, brand image. Positive brand awareness is required in order for a brand to be in consumers’ consideration set

0 100 200 300 400 500 600 20 0929 20 0935 20 0941 20 0947 20 0953 20 1006 20 1012 20 1018 20 1024 20 1030 20 1036 20 1042 20 1048 20 1102 20 1108 20 1114 20 1120 20 1126 20 1132 20 1138

(37)

30 Table 1: Pearson Correlation Coefficients (Diary)

* indicates significant correlation at the 0.05 level (2-tailed). ** indicates significant correlation at the 0.01 level (2-tailed).

(38)

31 Table 2: Pearson Correlation Coefficients (Energy)

* indicates significant correlation at the 0.05 level (2-tailed). ** indicates significant correlation at the 0.01 level (2-tailed).

(39)

32 when they consider purchasing a product from a certain product category (Keller 2012). Only when included in the consideration set, the brand might eventually be bought. In case the brand image is not positive, it is not included in the consideration set and therefore not purchased.

Obviously, the variables are highly positively correlated to their lagged effects, and the interaction effect of Google display advertising and spontaneous brand awareness is strongly correlated to its separate variables. The Pearson correlation coefficients of this data set are presented in Table 1.

4.3.2 Energy data set

Regarding the energy provider, cost of Google display advertising and the number of branded search terms are found to positively correlate (.401). This effect remains significant over time but decreases substantially (.336 for one week and .268 for two weeks after investing in Google display advertising), in contrast to the dairy brand.

Again, the variables are highly positively correlated to their lagged effects, and the interaction effect of Google display advertising and spontaneous brand awareness is correlated to its separate variables. The Pearson correlation coefficients of this data set are presented in Table 2.

4.4 Model fit

Four different models were created for each of the data sets by means of multiple linear regression (see also Section 3.3). In Model 1, cost of Google display ad, spontaneous brand awareness, and branded search terms were included as independent variables. In Model 2, the interaction effect of cost of Google display ad and spontaneous brand awareness was added by the creation of a new variable labeled (𝐺𝐺𝐺𝐺 ∗ 𝐵𝐵𝐵𝐵)_𝑡𝑡. In Model 3, the lagged effects of one week of Google display advertising, spontaneous brand awareness, the number of branded search terms, and the interaction effect were added. Lastly, Model 4 also contained the lagged effects of two weeks of Google display advertising, spontaneous brand awareness, the number of branded search terms, and the interaction effect.

(40)

33 in this table, all four models explained only a little percentage of the variability in the volume of sales of the dairy brand. Though Model 4 resulted in the highest R Square (R² = .322), Model 3 is chosen as best-performing model as this model resulted in the highest Adjusted R Square (Adjusted R² = .223). The Adjusted R Square punishes a model for containing too many variables and is therefore used as most important and guiding performance indicator. As a model should be as simple as possible (Leeflang et al. 2015), Model 3 is chosen as best model.

Model 1 Model 2 Model 3 Model 4

R Square .159 .193 .287 .322

Adjusted R Square .126 .152 .223 .219

Table 3: Performance measures of the four models (Dairy) 4.4.2 Energy data set

Regarding the energy provider, the ANOVA F-statistic of Model 1 was not significant (p-value = .055). This also holds for Model 2 (p-(p-value = .078), Model 3 (p-(p-value = .110) and Model 4 (p-value = .253). In Table 4, the model performance indicators of the four models for the energy provider are presented. As can be seen in this table, all models performed poorly. As the goal of this study to describe the relationship between the variables included in the model, rather than predicting the volume of sales, model fit should be taken into account but is not of such a great concern compared to predictive models. However, poor model fit indicates that the variables included in the model explain only a small portion of the variance in sales. For this reason, the model outcomes have to be interpreted carefully. As Model 3 performed best among the three models (Adjusted R² = .050), this model will be used in the next sections.

Model 1 Model 2 Model 3 Model 4

R Square .080 .086 .125 .142

Adjusted R Square .047 .044 .050 .029

Table 4: Performance measures of the four models (Energy)

4.5 Model estimation

After comparing the four different models on the basis of model fit and identifying the best-performing models, the model outcomes for each of the best models are presented.

(41)

34 the lagged effect of the interaction effect were found to have a significant effect. The dummy variable is highly significant, which captures the effects of external events on two sales peaks.

B Std. Error Beta t Sig. VIF

(Constant) 392091.198 70638.766 5.551 .000 GDt -3166.004 1072.205 -2.451 -2.953 .004** 90.813 BAt 1636.012 5251.420 .112 .312 .756 16.930 BSt 64665.582 183693.710 .067 .352 .726 4.744 GD*BAt 145.902 47.772 2.577 3.054 .003** 93.905 GDt-1 2052.017 1108.015 1.590 1.852 .047* 97.136 BAt-1 -6357.543 5446.578 -.427 -1.167 .246 17.668 BSt-1 96202.229 172155.695 .100 .559 .578 4.194 GD*BAt-1 -99.580 49.219 -1.761 -2.023 .046* 99.846 Dummy_S 154986.251 42162.030 .330 3.676 .000** 1.065

Table 5: Estimation Model 3 (Diary).

* indicates significant correlation at the 0.05 level (2-tailed). ** indicates significant correlation at the 0.01 level (2-tailed). 4.5.2 Energy data set

Also for the energy provider, Model 3 performed best. In this model, only the dummy variable for sales was found to significantly affect sales volume (see Table 6).

B Std. Error Beta t Sig. VIF

(Constant) 710.109 2087.399 .340 .734 GDt -6.146 7.484 -.805 -.821 .413 114.350 BAt -5.913 29.530 -.037 -.200 .842 4.047 BSt -718.715 635.181 -.194 -1.132 .260 3.489 GD*BAt .112 .136 .800 .825 .411 111.664 GDt-1 9.383 7.886 1.209 1.190 .237 122.741 BAt-1 42.818 30.080 .267 1.423 .158 4.168 BSt-1 986.667 653.091 .263 1.511 .134 3.600 GD*BAt-1 -.178 .142 -1.258 -1.254 .213 119.651 Dummy_S -2600.421 962.161 -.255 -2.703 .008** 1.061

Table 6: Estimation Model 3 (Energy).

** indicates significant correlation at the 0.01 level (2-tailed).

4.6 Testing mediation

EXPLORING THE EFFECT OF GOOGLE DISPLAY ADVERTISING ON BRAND EQUITY

Master Thesis Marketing Intelligence & Marketing Management

Eline Nieboer

EXPLORING THE EFFECT OF

GOOGLE DISPLAY ADVERTISING

ON BRAND EQUITY

Exploring the effect of Google display advertising on brand equity

The role of product involvement

MANAGEMENT SUMMARY

PREFACE

TABLE OF CONTENTS

1. INTRODUCTION

2. LITERATURE REVIEW

3. RESEARCH DESIGN

4. RESULTS

Sales (volume)