Competition in search engine marketing: the winner takes all?

Competition in search

engine marketing: the

winner takes all?

Competition in search engine marketing: the

winner takes all?

Master thesis

University of Groningen

Department of Marketing

June 15, 2010

tetteboekema@hotmail.com

Supervisor

dr. K.R.E. (Eelko) Huizingh

2nd Supervisor

Management summary

Search engine marketing (SEA) is the fastest growing form of advertising worldwide and is becoming a considerable element of the marketing mix of many companies. This study examines data of 12 companies from the package holiday industry over a one year period. The aim is to gain knowledge about the role of competition in the field of SEA. In particular, the research question is as follows:

What is the influence of competition on SEA performance at the individual keyword level, as well as at the broader level regarding a company’s size and market focus?

The results of the study show that competitive intensity mediates the link between wordographics (i.e. generic vs. advertiser specific) and cost per click. That is, competitive intensity is higher for generic keywords than for advertiser specific keywords and, subsequently, it increases cost per click. Furthermore, competitive intensity negatively moderates the positive impact of cost per click on position. In sum, the intensity of competition makes it more expensive and more difficult to perform well on the search engine results list in terms of the rank of an advertisement. The results further show that SEA performance, to a certain extent, depends on a company’s size and market focus (i.e. niche vs. mass). Company size intensifies the impact of both position on click through rate and position on conversion rate. A company’s market focus moderates the impact of wordographics on conversion rates and profit margins. The difference between generic keywords (lower) and advertiser specific keywords (higher) for these metrics is smaller for niche players. The study indicates that generic keywords drive this difference in that they show higher conversion rates and profit margins for niche players than for mass players.

This study provides insights into two contradicting ways of thinking and their link to SEA: the winner takes all theory and the long tail theory. Considering the aggregated average values of the various SEA performance metrics, as well as the moderating role of company size, on would assume the winner takes all theory best fits the field of SEA. However, the positive effects for niche players that result from the analysis concerning market focus show that the long tail theory can also be applied to SEA.

awareness. Companies with a niche market focus should be creative in the keywords they select for advertising. The keywords should truly reflect the uniqueness of the product they sell. Furthermore, related keywords that do not reflect the uniqueness of the products sold, but rather the uniqueness of the interests of the target group should also be targeted. Finally, niche players should learn what the competitive intensity of a keyword is before targeting it. Highly competitive keywords are not the ones that will increase a niche player’s SEA performance.

Table of contents

1. Introduction ... 1

2. Research Context: Search Engine Advertising ... 5

3. Literature and theoretical background ... 7

3.1 Consumer search ... 7

3.2 Internet Marketing ... 8

3.3 Search engine advertising (SEA) ... 9

4. Research Hypotheses ... 12 4.1 Competitive intensity ... 12 4.2 Company characteristics ... 14 4.2.1 Company size ... 15 4.2.2 Market focus ... 17 5. Data ... 19 6. Research Methodology ... 21 6.1 Mediation (H1) ... 22 6.2 Moderation (H2, H4 – H7) ... 23 7. Results ... 25 8. Discussion ... 31 9. Conclusion ... 35 10. Managerial implications ... 37

11. Limitations and future research ... 39

1. Introduction

The rise of the internet during the last decade has been enormous. With this exponential growth came the need to assist internet users in their quest for information. Search engines were founded and showed people the way. These days, they are extremely popular. This even resulted in the introduction of a new verb: “to Google”, referring to the leading search engine. Increasing popularity created opportunities for search engines to financially benefit from the large number of visitors. This resulted in the introduction of search engine advertising (hereafter SEA). These days, this is the major source of income for search engines and the fastest growing form of advertising worldwide. Moreover, it is now a substantial part of the marketing mix of many companies. Figures from the USA, one of the leading countries within this industry, highlight the impressive growth. According to the industry’s trade association, SEA spending in the USA was close to $15 billion in 2009. This yields a growth of nearly 10% compared to 2008, whereas total ad spending in the USA shows a downfall of nearly 13% in the same period, due to the economic crisis1. SEA spending will grow even more to an estimated $ 19,8 billion in 2011, representing a 35% growth in 3 years.2 By then, it will considerably exceed radio and newspaper advertising spending, capturing 13% of the total advertising market in the USA. This study aims to gain more insights into the role of competition in SEA.

In order to manage SEA campaigns, advertisers make use of online software programs designed by search engines (e.g. Google Adwords, Yahoo Search Engine Marketing). Through these programs, various types of metrics are made available to advertisers. Examples include number of impressions, number of clicks and number of sales for each keyword. A description of these metrics and the way SEA works will be provided in the subsequent chapter. Rich datasets from SEA programs, which comprehend keyword level data, are the source of an emerging stream of literature (e.g., Ghose & Yang 2009, Rutz & Bucklin 2007, Agarwal et al. 2008, Animesh et al. 2009). These studies describe the impact of type and length of keywords on a number of metrics, such as position, click through rates and conversion rates. Besides, the interrelation between these metrics is examined.

The contributions of this stream of research are substantial. However, since each study empirically studies data of only one company, an important limitation arises. By only examining one company at a time, differences between companies cannot be accounted for. For example, the impact of the size of a company or its market focus (i.e. mass vs. niche) on search and purchase behavior of consumers

1 _{http://www.zenithoptimedia.com/gff/pdf/Adspend%20forecasts%20July%202009.pdf}

2

has not yet been researched in literature. A second limitation is related to competition. Since each keyword in fact represents a small competition on its own, including different bidders and different bid prices, the intensity of competition for each keyword and its impact on purchase and search behavior of consumers is an important topic.

This study aims to provide insights into these gaps in literature by analyzing SEA data of 12 companies, which compete in the holiday industry. More specifically, the research question of this study is as follows:

What is the influence of competition on SEA performance at the individual keyword level, as well as at the broader level regarding a company’s size and market focus?

First, within the model of relationships that emerges from existing work, the inclusion of an additional variable is proposed: competitive intensity of a keyword. Gaining more knowledge about competition in this industry fits the call for research from several authors (e.g., Ghose & Yang 2009, Rutz & Bucklin 2007, Animesh et al. 2009). Second, differences between competing companies will be dealt with. That is, company size and market focus (i.e. niche vs. mass) are expected to have moderating roles in various relationships regarding SEA. Several authors suggest future work should discuss company characteristics in SEA (Ghose & Yang 2008, Animesh et al. 2009).

Gaining knowledge about competition in SEA is important from a scientific point of view. In SEA, the search intentions of potential customers can be derived from their search queries in great detail, so targeting can take place at the individual keyword level. This is different from traditional forms of advertising in which the focus is on groups of customers that share common characteristics, which is a much broader form of targeting. Accordingly, some interesting differences in competition may apply. By analyzing competitive intensity on the keyword level, this study aims to gain more insights into this notion.

SEA. From a scientific point of view this is important, as SEA fits the long tail theory particularly well. For example, a high degree of fit between a search query and an advertisement provides a means to offer specialized niche products to the right customers. This study aims to contribute to a growing stream of research that deals with the long tail theory by analyzing both niche players and mass players in SEA.

In field of SEA, another way of thinking contradicts the long tail theory. Through several mechanisms it is assumed that large companies that have large target groups eventually have the most advantage SEA. Rangaswamy et al. (2009) provide an interesting discussion on this topic. The authors label this way of thinking as “the winner takes all” theory.

Anderson (2006) explicitly formulates the leading search engine Google to be a typical long tail company, because most of its revenues come from the smaller advertisers. On the other hand, Rangaswamy et al. (2009) state that search engines are in the end inclined to favor large companies, as the most popular products will lead to higher turnover for the search engines. This is contrary to Anderson’s (2006) statement. This study does not deal with the performance of the search engines, but with the performance of advertisers that make use of these engines. However, a similar discussion can be applied to the advertisers in SEA. Which companies take the lead in the end and so which theory is right? Is it the larger, mass focus companies, or the smaller, niche focus companies that perform best in the field of SEA? In simultaneously examining multiple companies, this study aims to gain more insight into the answer to this question.

A better understanding of the role of competition has practical relevance as well. Regarding the competitive intensity at the keyword level, several subjects are important. Because SEA involves an auction mechanism, the outcomes are to a great extent influenced by competition. That is, the more competitors, the higher the cost of advertising. In SEA, each keyword represents its own auction. Therefore it is important to take the competitive intensity at the keyword level into account, rather than competitive intensity at the company level. For example, company A and company B compete intensively with each other, reflecting a high level of competitive intensity at the company level. However, company A sells a product with a range of corresponding keywords that is not sold by company B. In this case there is no competition regarding these keywords, despite the fact that there is competition at the company level.

expect competitive intensity of keywords to grow significantly3. The fact that entry barriers to SEA are extremely low even intensifies this growth. At the same time, the limited number of positions available to advertisers will not grow.

Finally, practitioners in the SEA business rely heavily on performance metrics such as click through rates, conversion rates and position. The inclusion of competitive intensity and its relation with these metrics within the existing models can provide helpful insights.

Practical implications regarding differences in company characteristics (i.e. size and market focus) and the impact on SEA performance arise as well. For example, SEA has very low barriers to entry, as compared to traditional forms of advertising. The pay per click mechanism allows even the smallest players in the market to compete.

Location is another relevant issue in SEA. In the offline world, large companies have the power to reside in expensive, prime locations to attract customers (e.g. city centers). Online, however, most internet users start their buying process on a search engine, which is accessible for small companies too. Increasing competition in SEA in conjunction with the limited number of advertising positions is also a relevant concern. Because of increased competition, one would expect cost of advertising to rise to high levels. This might yield different practical consequences for different types of companies. Furthermore, SEA is accessible just as easy for niche players as it is for mass players. This is different in more traditional forms of advertising, such as TV. At the same time, niche players focus on specialized products, which could yield important differences in SEA performance. It is important for practitioners to gain more knowledge about the impact of company characteristics on SEA performance

The study is structured as follows. Section 2 presents an overview of the SEA context, including definitions and an explanation of the way the auction principle is utilized in SEA. Section 3 reviews the literature in the field of SEA and other relevant streams, such as internet marketing and consumer search. In section 4 I present the data. The research hypotheses regarding competitive intensity and company characteristics are discussed in section 5. Section 6 describes the research methodology, followed by the results in section 7. Section 8 provides a discussion of the findings. Section 9 concludes the paper, followed by the managerial implications in section 10. Limitations and directions for future research are discussed in the last part, section 11.

3

2. Research Context: Search Engine Advertising

How does SEA work? A company that would like to advertise on the search engine results page, bids on keywords it considers relevant to its business. A keyword may consist of one or multiple words. For the convenience of the reader the term keyword will be used in this study, regardless of the number of words concerned. Each keyword comprehends information, such as product name, the name of the salient company or product attributes. In SEA literature, keywords are often grouped according to these types of content. The corresponding group variables are labeled “wordographics”, a term introduced by Rutz & Bucklin (2007). A company can bid a specific amount of money for each individual keyword, called bid price. It provides the search engine with an advertisement text and a link to its website for each keyword or cluster of keywords. The webpage to which the advertisement links is called the landing page. One advertiser may have multiple landing pages. Once an internet user searches for a keyword that a company is targeting, the advertisement of the company is shown alongside the normal, natural (hereafter organic) search results. The appearance of such an advertisement is called an impression. An impression is free of charge. Only when an internet user actually clicks an advertisement, known as a click or click through (in this study the term click is used), the advertiser has to pay a fee. The search engine determines the place of the advertisement on the search engine results page, which is called position. At the major search engines, such as Google, Yahoo and Bing, a maximum of three advertisements is shown above the organic results, and between five and ten advertisements are shown to the right of the organic results. Figure 1 depicts a typical search engine results page. Positions one, two and three appear above the organic results, four and further appear to the right.

(e.g. click through rate and landing page quality4). The precise algorithm and the weight of the variables is not revealed by search engines. The cost per click is always lower than the bid price, as a result of the second price auction mechanism. The percentage of impressions that generates a click is called the click through rate (CTR). This metric is extensively used by practitioners in evaluating the performance of an advertisement.

Usually, the click is not the ultimate goal of an advertiser in SEA. An advertiser might want to sell something, but it could also have the aim to merely provide information. In both cases, advertisers often evaluate their campaigns by considering the number of conversions. Examples of a conversion are a request for a brochure, an appointment with a representative, a subscription to a mailing list or the sale of a product or service. The percentage of clicks that generates a conversion is called the

conversion rate, which is an important performance indicator.

Search engines constantly monitor the degree of fit between a keyword and a related advertisement, ensuring the best possible search experience for the internet user. Here lies the strength of SEA. Compared to traditional advertising formats (e.g. TV, radio, outdoor) and other internet advertising

4 _{For example, see http://adwords.google.com/support/aw/bin/answer.py?answer=10215 for a complete list}

of Google’s quality score variables.

(e.g. banners, email), the degree of fit between the user’s goals and the advertisement is much larger in SEA.

3. Literature and theoretical background

Only recently, researchers have recognized the possibilities of using advertiser level SEA data to study new insights from the field. The foundation of the studies involved is mainly shaped by two streams of research: consumer search and online marketing. Consumer search provides the link to the search part of the topic, whereas online marketing is related to the online aspect. Both streams will be discussed subsequently. The stream of research that deals with advertiser level SEA data is most closely related to this study. It is discussed in the last paragraph of this chapter.

3.1 Consumer search

Stigler (1961) wrote the first key study regarding consumer search. The author proposed a theory called the ‘economics of information’ in which he describes a tradeoff between the costs and benefits of search by consumers. Nelson (1970, 1974) later differentiates between products that comprehend search or experience attributes, the first denotes attributes which can be obtained prior to purchase, the latter only after the actual use of the product. Darby & Karni (1973) add credence attributes as those attributes which cannot be evaluated even after purchase. Schmidt & Spreng (1996) argue consumer search not only involves product attributes, but also the source of information, external or internal (i.e. previous experience).

Why is consumer search theory important in the context of the internet and SEA? According to Klein (1998) and Klein and Ford (2003), the internet caused shifts in this paradigm in that it not only decreases search costs of search attributes, but also changes the attribute differentiation for some products. As the authors state, through ‘virtual experience’ the internet can turn offline experience attributes into search attributes in the online world. Both the notion of lower search costs and the attribute shift are arguably reasons SEA grew so rapidly, given the benefits it entails for both companies and consumers.

2009). Furthermore, by searching for a product on a search engine, an internet user can see at a glimpse which advertisers offer the salient product. Thus, internet users do not necessarily need to pay a visit to all the advertisers’ websites in order to gather information. In sum, research involving multiple competing websites contributes to our knowledge, but the neglected field of SEA unquestionably plays a role in online competition between companies. In contrast to these studies, this study explicitly focuses on SEA and in particular the competition that takes place in the field. Several authors describe differences between internet users and their search behavior. For example, Moe (2003) shows that internet users can be categorized as buying, browsing, searching or knowledge building. Each category varies in terms of purchase likelihood. Viswanathan et al. (2007) show that different clusters of internet users exist that are based on what these users are searching for: price versus product information. They grouped internet users by the infomediaries they use. Consumers focusing on price are found to pay a lower price for the same product. Related to this article, one study finds that the search engine results page can serve as a viable segmentation mechanism (Animesh et al. 2009). Specifically, the top of the list seems to attract more quality seeking internet users, whereas the bottom attracts more price seeking internet users. Although these researchers considerably added to our understanding of online segmenting of consumers, the actual keyword entries that could serve as a differentiation means are not considered. This study differs from these articles in that it uses aggregated keyword level data instead of individual level data. Although the latter might be more suitable for detailed segmentation, keyword entries of internet users do reveal information according to which advertisers can take different action. In this study, these keyword entries serve as input to reveal possible differences across internet users that could be of interest to companies.

3.2 Internet Marketing

best fits the stream of research that considers consumer behavior as it deals with consumers’ reactions to companies’ SEA efforts.

Online consumer behavior is also dealt with in literature regarding bannering, which is the other major type of online advertising. Manchanda et al. (2006) show a positive effect for banner exposure on purchase over time. The authors speculate this might be the result of an increase in brand attitude and awareness. Indeed a positive effect of bannering on brand awareness exists (Dreze & Hussherr 2003, Dahlen et al. 2003, Ifeld & Winer 2002). Although marketers can provide in a fit between the message of banners and visitors’ goals, this fit cannot be as specific as in SEA. This highlights an important difference between SEA and bannering. That is, the latter performs better in building awareness and brand attitude, whereas SEA performs better in the degree of fit between the advertisement and the user’s goal. Banners can be indirectly linked to a certain group of visitors who visit the same webpage, while an advertisement in SEA is directly linked to the individual visitor’s goal. Nevertheless, the question remains to what extent SEA can add to awareness of advertising brands. To the knowledge of the author, only one study deals with awareness and its impact on SEA performance, discussed later in this chapter (Rutz & Bucklin 2008). Although this study does not research the impact of SEA on awareness explicitly, it does consider differences in SEA outcomes between multiple companies that might be driven by awareness.

Other research has focused on email. For example, Ansari & Mela (2003) show that, by applying the right statistical methods, high levels of customization can be obtained that increase success of email advertising. For example, the order of the links presented to the internet users plays a role in the profitability of an email campaign. Although the findings are significant, they encompass multiple advertisements of one company. SEA on the other hand, involves single advertisements of multiple companies. Here, the order of advertisements also yields major consequences for the companies involved. This study deals with both the order of advertisements, as well as the existence of multiple competing companies in the list.

As discussed in a literature review on the subject of online marketing, a lot of work is to be done (Schibrowsky, 2007). This is true for SEA in particular, as it is a topic of the past couple of years. However, already some profound studies have been published, discussed next.

3.3 Search engine advertising (SEA)

In order to provide a clear description of this field of research it is sensible to use the different metrics applied by both researchers and practitioners as a guideline. These metrics can be divided into three segments. The first describes the metrics that involve direct input from the advertiser, the second those that involve the performance of a keyword within the search engine and the third those that involve the performance on the website of the advertiser. The ultimate goal of the advertiser is to enhance results in the third segment: website performance. Figure 2 clarifies this division and reflects the relationships that are found in literature.

As can be seen in this figure, the group variable wordographics (i.e. advertiser specific vs. generic) is related to both search engine performance and websites performance. More specifically, advertiser specific keywords show higher click through rates, conversion rates and profits than generic keywords. Furthermore, cost per click is lower for advertiser specific keywords than for generic keywords (Ghose & Yang 2009; Rutz & Bucklin 2008; Rutz & Bucklin 2007; Yang & Ghose 2008, Ghose & Yang 2008a). The findings of these researchers have contributed considerably to our knowledge regarding the interrelationships between various performance metrics in SEA. However, they derive their findings from datasets that belong to one single company. Although all of these articles mention competition as an important factor within the SEA mechanism, none of the studies empirically deals with it. This study differs from existing literature in that it does include competition in its examination of SEA performance.

Given the fact that generic keywords score lower on all of the important performance metrics, there does not seem to be a reason to invest in them. However, a significant amount of spillover takes place from generic keywords into advertiser specific keywords due to increased awareness of the advertiser (Rutz & Bucklin 2008). The research of these authors is related to this study in that I examine differences between competing companies and their impact on SEA metrics, possibly driven

by awareness of the advertiser. The fact that awareness plays a significant role in SEA is therefore an important finding. However, Rutz & Bucklin (2008) examine the effects of awareness in SEA by considering one company. They show that awareness is increased by generic keywords and that this awareness subsequently results in more advertiser specific keyword searches that in turn lead to more clicks and sales. This study takes a different approach regarding awareness. That is, differences between companies regarding size and market focus might be related to different levels of awareness. The question how awareness is created, is not the focus. Rather, the fact that differences exist in levels of awareness and the possible impact of it on SEA performance is the area of interest. Several relationships exist between the metrics that describe the performance of a keyword at the search engine. The position of a keyword has an impact on the click through rate. That is, the closer a keyword is to the top of the list, the more clicks it generates. (Ghose & Yang 2009; Yang & Ghose 2008; Agarwal et al. 2008; Animesh et al. 2009). Furthermore, the higher the click through rate, the higher is the cost per click. This probably reflects the attractiveness of a keyword from the advertiser point of view because it is apparently willing to pay more for keywords that generate more clicks (Rutz & Bucklin 2007). The reasoning behind these findings is quite evident and universal across multiple companies. Arguably, the strength of these relationships could be different if company characteristics are taken into consideration, an area that is neglected by these researchers. This study deals with this notion and aims to find differences related to a company’s size and market focus.

Apart from the research that has been done regarding the relationships between the metrics used in SEA, other studies take a broader perspective. Building on theories of cross selling and impulse buying, it is found that a significant amount of spill over exists between different product categories. (Ghose & Yang 2008b) That is, a certain amount of consumers that searches for a product in one category ends up buying a product from another category. This effect is most salient for advertiser specific keywords, reflecting loyalty to the advertiser. In their research, the authors used data of a large market leader. The question remains whether spillover effects also exist for smaller companies that do not have the advantage of being one of the market leaders. Spillover effects are not the focus of this study. However, the finding that large differences exist between keywords with different wordographics, do make it related to this study. The difference is that this study mainly considers differences between companies regarding wordographics and the impact it has on SEA performance. SEA results are always presented alongside organic results. Therefore, research in this area is closely related to this study. An important finding is the fact that the presence of an advertiser in the organic results list boosts the results of SEA, and vice versa (Yang & Ghose 2008). A similar effect of the influence of a third party source is reported in online marketing theory (Weathers et al. 2007). Synergies thus take place between organic search and SEA. As the authors bring forward, awareness possibly drives these synergies. In agreement with these studies, it is assumed that awareness indeed plays a role in SEA. Although the focus here is not on synergies between organic search and SEA, awareness might be the driver of differences in outcomes of several SEA performance metrics. The dataset used in this study covers multiple companies and thus allows for testing such awareness related differences.

4. Research Hypotheses

This section is subdivided into two paragraphs. First, I will provide hypotheses regarding the role of competitive intensity at the keyword level. A clarification of the role of competition on the keyword level serves as a preliminary for the subsequent paragraph. That is, it should first be established that competition on the keyword level takes place, before differences between competing companies are dealt with. These differences in company size and market focus are the subject of the second paragraph.

4.1 Competitive intensity

(Rutz & Bucklin 2008, Yang & Ghose 2008). It is debatable whether the differences between these sets of keywords can be explained by their wordographics. Cost per click is an outcome of the second price auction. It is computed by the search engine by combining the quality of the keyword and the minimum cost of the position that the advertisement is placed on (i.e. the amount of the second highest bidder). Advertiser specific keywords have less competing advertisers bidding on them than generic keywords. Therefore, competitive intensity is higher for the latter set of keywords. Competitive intensity, subsequently, has a positive impact on cost per click.

Besides this, search engines charge the highest bidder only the second highest bid, plus $0,01. Taking into account the fact that more advertisers bid on the keyword when competitive intensity is high and that advertisers do not know each other’s bid prices, one can assume that the second highest bid is closer to the maximum bid price than it is in the case of less bidders (Edelman et al. 2007). For these two reasons, it is hypothesized that competitive intensity mediates the difference in cost per click per click between advertiser specific keywords and generic keywords.

H1: The difference in cost per click between advertiser specific keywords and generic keywords is mediated by the competitive intensity of a keyword.

Ghose & Yang (2009) further show that sets of keywords with different wordographics have different average positions. That is, advertiser specific keywords are ranked higher than generic keywords. It is questionable whether the difference between these sets of keywords is directly related to wordographics. As Ghose & Yang (2009) find in their empirical analysis, cost per click has the largest role in determining the position. Although the exact magnitude of the function of cost per click is not revealed by search engines, the fact that it plays a major role is important in the light of competition. Given the fact that competitive intensity is higher for generic keywords, the number of positions has to be divided between more advertisers, which lowers the average position. But even if competitive intensity is high, if one advertiser bids a very high amount it can still be granted a high position. Differences regarding position for keywords with different wordographics are therefore not mediated by competitive intensity. Rather, when competitive intensity is high, the relationship between cost per click and position is attenuated. This leads to the second hypothesis:

Rutz & Bucklin (2007) find that keywords with a higher click through rate are more expensive. The authors speculate this might be the result of competition. A keyword with a higher click through rate is generally more attractive to advertisers. Therefore, more companies tend to bid on these keywords, which drives the cost per click. The dataset used in this study includes keyword level data of competition, allowing for testing this hypothesis:

H3: Click through rate has a positive impact on competitive intensity.

Figure 3 depicts a comparison between relationships from current literature and the revised model of this study that includes the competitive intensity variable.

4.2 Company characteristics

In the previous paragraph, competitive intensity was discussed at the keyword level. This paragraph takes a broader perspective and deals with competition by considering the differences between competing companies. Paragraph 4.2.1 discusses the role of company size. Subsequently, paragraph 4.2.2 deals with differences between niche and mass players.

Wordographics

advertiser specific vs. generic

Cost per click Click through rate

Position

Cost per click Click through rate

Position

H1 _{Competitive intensity}

+ |H3

+ |H1 - |H2

Revised model in this paper Current literature

Figure 3. Comparison between relationships from current literature and the proposed hypotheses.

+

+ Wordographics

4.2.1 Company size

One of the major implications of the mechanism of SEA is the fact that even the smallest companies can compete in the auction, due to the low cost per individual click and the absence of a minimum purchase amount. This is a disadvantage for larger companies, as they are facing more competition compared to other forms of advertising (e.g. TV, radio), that are not affordable for small companies. When a small company bids on the same keyword for the same amount of money as a large company does (not accounting for quality score), it has the same chance of being positioned high. It can thus reach the same prime position as the large company does. But does this mean that large, more established companies lose their advantage over small companies?

Rangaswamy et al. (2009) provide interesting insights in this discussion. On the one hand, one can expect consumers to search more specifically for the product they want. That is, the search engine allows searching for multiple product attributes at the same time, possibly resulting in a better fit between the consumer and the product offer. This is somewhat similar to what is called the long tail theory in recent literature (Anderson 2006). This theory speculates that, as a consequence of the possibilities of the internet, an increasing number of consumers have access to (and buys) less popular products that appear in the long tail of the sales distribution. This might add to the decreasing advantage of larger companies, as these companies generally sell popular products in large quantities. On the other hand, it is likely that search engines can contribute to the winner takes all theory. As Elberse (2008) shows in the case of music and home videos, a concentration of sales exist in the head of the distribution, rather than in the tail. Search engines are inclined to show the most popular products in the top of the list to satisfy the majority of users. These theories can be linked to SEA by considering the role of position in the search engine results list.

are placed further down the list. Therefore, the positive effect of position on click through rate is stronger in the case of large companies compared to small companies. This leads to the fourth hypothesis:

H4. The positive impact of position on click through rate is intensified by the size of a company.

Similarly to the relationship between position and click through rate, a positive relationship exists between position and conversion rate (Ghose & Yang 2009). That is, the higher the position of an advertisement on which a consumer clicks, the higher the likelihood of purchase. This is quite an interesting finding as it not only shows that higher positions result in more clicks, but also in relatively more sales after a click takes place. Apparently, there is some kind of carry-over effect of position into conversions. This effect has been empirically shown in the field of shop-bots by Brooks (2004), Baye et al. (2009) and Brynjolfsson et al. (2004). Assuming that the awareness of large companies is higher than the awareness of small companies; the mere exposure effect might result in a difference for this effect between large and small companies. That is, the carry-over effect is stronger for large companies, since internet users are more likely to buy because of the mere exposure effect (Bornstein 1989; Janiszewski 1993). Therefore, the positive impact of position on conversion rate is stronger for large companies than for small companies. This leads to hypothesis 5:

H5. The positive impact of position on conversion rate is intensified by the size of a company.

Figure 4 visually depicts the hypotheses regarding the role of company size. Profit Company size Position

Click through rate + | H4

+ | H5

Conversion rate +

+

4.2.2 Market focus

The use of the ecological paradigm in the field of marketing has a long history. Alderson (1957) first introduced the concept of market niches. Lambkin & Day (1989) build on this and show similarities between ecological niches and market niches. Recently, the ecology paradigm has also been applied to internet marketing (e.g. Debruyne & Reibstein 2005), as is the case in this study. To the knowledge of the authors, this is the first study to specifically link the ecological paradigm to SEA performance. I propose that the distinction between niche and mass players is related to profitability of keywords in SEA. The profit margin per keyword (i.e. profit per keyword divided by revenue per keyword) has been shown to be different for keywords with different wordographics. That is, advertiser specific keywords are more profitable than generic keywords (Ghose & Yang 2008). It is expected that this difference is larger for mass players than for niche players because of the profit margins of generic keywords. Niche players can charge higher prices because they add value by focusing on specialized products (Kotler 1991; Dalgic & Leeuw 1994). Therefore, revenue is higher for generic keywords of niche players, compared to generic keywords of mass players. Furthermore, one can assume that competition of generic keywords of niche players is less intense, since the number of advertisers in a niche market is lower. This decreases the costs of advertising for generic keywords. In sum, the revenue of generic keywords is expected to be higher, whereas costs are expected to be lower for niche players compared to mass players. Therefore, the profit margin for generic keywords is higher for niche players. Regarding advertiser specific keywords, no substantial difference is expected in profit margins. That is, for both niche and mass players, revenues for advertiser specific keywords are high because of high conversion rates (Ghose & Yang 2009, Rutz & Bucklin 2007, Rutz & Bucklin 2008, Yang & Ghose 2008, Ghose & Yang 2008a), whereas cost for advertiser specific keywords are very low because of low levels of competitive intensity (i.e. only the advertiser itself is interested in the keyword). The profit margin of these keywords is therefore expected to be very high for both types of companies and not statistically different. In sum, this reasoning suggests that the difference in profit margin found in literature (i.e. advertiser specific keywords are more profitable than generic keywords) is smaller for niche players, compared to mass players, because of higher profit margins for generic keywords. Therefore, hypothesis 6 is as follows:

H6: The difference in profit margin between advertiser specific and generic keywords is smaller for niche players than for mass players.

Ghose & Yang 2008a). More specifically, advertiser specific keywords show higher conversion rates than generic keywords. I expect to find differences between mass and niche players for this finding, driven by generic keywords.

A niche player only sells a specialized (range of) product(s). Therefore, one can assume generic keywords for a niche player to reflect a more advanced stage in the decision process than generic keywords for a mass player. In the latter case, a consumer might consider more advertisers, as mass players have more competitors and a more homogenous common product than niche players (Dalgic & Leeuw 1994; Kotler 1991, 2003). Furthermore, because a niche market typically only has a limited number of competitors, a search for a generic keyword of a niche player is more likely to be followed by an actual sale, compared to the same situation for a mass player. Therefore, the conversion rate for generic keywords is expected to be higher in the case of niche players. No difference between niche and mass players for the advertiser specific keywords in terms of conversion rates is expected. In sum, advertiser specific keywords have substantially larger conversion rates than generic keywords for both niche and mass players. However, generic keywords of niche players are expected to show higher conversion rates compared to those of mass players. Therefore, the difference between advertiser specific and generic keywords is smaller for niche players than for mass players:

H7: The difference in conversion rates between advertiser specific and generic keywords is smaller for niche players than for mass players.

Figure 5 visually depicts the hypotheses regarding the role of market focus. To put all hypotheses in a broader perspective and link them to existing research, figure 6 is added. This figure reflects all relationships found in literature which also appear in figure 2, enriched with the hypotheses from this study. As can be seen, competition and the moderating roles of company characteristics are added as different segments in the model.

Profit Market focus Profit margin H7 H6 Conversion rate

Figure 5. Visual representation of the hypothesized moderating roles of market focus.

Wordographics

5. Data

Data from SEA differ considerably from data collected in traditional ways of advertising. In SEA, search engines provide data on the basis of which advertisers monitor, evaluate and adjust their campaigns. It is aggregated on the basis of keywords and should therefore not be confused with click stream data which is on the basis of the individual internet user. The data used in this research is provided by one of the leading search engine marketing consultancy firms in Europe. It encompasses SEA data of 12 companies from the package holiday industry in a European country, all of which advertise at the search engine Google.

The data is collected from both Google’s SEA program, called Adwords, and from Google’s site statistics software, called Google Analytics. These sources are closely linked to each other. That is, an advertiser that uses both can chose to integrate its Adwords statistics in its Analytics reports, which enhances the data and takes it to a broader perspective. The data span all keyword advertisements over a period of 1 year, specifically from January 1st to December 31st 2009. From the Google Analytics accounts of the 12 companies, data is collected including number of impressions, number of clicks, number of conversions (i.e. an actual sale of at least one package holiday), cost per click and total revenue of all conversions for each keyword. Google Adwords provides the average position of a given keyword on the search engine results page. Note that in this study, a position with a lower number is referred to as a higher, more positive position, since numbers increase if one goes down the list. Based on these data, I computed additional metrics as follows. Click through rate is defined as the total number of clicks divided by the total number of impressions per keyword. The conversion rate is equal to the total number of conversions divided by the total number of clicks. The product of cost per click and the total number of clicks provides the total cost of a keyword. The total revenue

Figure 6. Relationships of the various SEA metrics reported in literature, enriched with the ones hypothesized

that results from searches for a given keyword subtracted by the total cost of all clicks for that keyword is the total profit. The profit divided by the revenue for a given keyword is the profit margin. A small number of popular keywords generate the majority of impressions, clicks and conversions. The majority of keywords generates only a few impressions and often no, or very few clicks and conversions. For this reason, and due to practical constraints, only keywords that generate the top 70 percent of clicks are included in the dataset. A similar reasoning and corresponding threshold was used by Animesh et al. (2009). After deleting keywords that generate to little clicks, a total number of 2046 unique keywords remain for the 12 companies.

Besides the data that is directly available through these programs, two other variables at the keyword level are part of the analyses in this study. First, I define the competitive intensity of the keywords as the average price to be paid to reach position one, two or three on the search engine results page for a given keyword. These data is provided by Google’s Traffic Estimator, a tool that can be used by Adwords’ advertisers.5 Google provides a monetary interval to estimate the amount that should be paid to reach these top positions. In this study, the middle of this interval is used as the metric for competitive intensity. The higher Google’s estimate, the higher is the competitive intensity of a keyword. Since SEA involves an auction, this metric is driven by both the money that advertisers are willing to pay and the number of advertisers, which makes it a prosperous proxy for competitive intensity.

Second, the data is enhanced by introducing a group variable that reflects the content of a keyword, labeled wordographics. In literature, wordographics are used as a group variable in several forms. For example, the presence of retailer specific, brand specific or generic information or the presence of city, state and brand name serve as values for the variable wordographics (Ghose & Yang 2009, Rutz & Bucklin 2008). In this study, I added a dummy variable for each keyword, based on whether the keyword comprehends the name of the advertiser or not. In particular, keywords that comprehend the name of the advertiser are referred to as advertiser specific keywords with the corresponding value 1. Keywords that do not comprehend advertiser specific information are identified as generic keywords corresponding to the value 0. This breakdown is universal and can be used for any industry. In addition to the keyword level data, variables that describe the differences at the level of the company are measured as follows. First, the size of the company is measured by its total online turnover. The data is provided by the Google Analytics software. The second company characteristic is market focus, for which a dummy variable is added to the data. There is no widely accepted single definition that encompasses the difference between a niche and a mass player. However, several

5

authors have endeavored to define niche marketing. For example, Kotler (2003) defines niche marketing as ”focusing on customers with a distinct set of needs who will pay a premium to the firm that best satisfies their needs, where the niche is not likely to attract many other competitors and where the niche firm gains certain economies through specialization”. Dalgic & Leeuw (1994) classify niche marketing as “positioning into small, profitable homogenous market segments which have been ignored or neglected by others”, furthermore the authors stress that a niche is different from a segment in that it fulfills a specific need in contrast to just representing a manageable part of the market. Finally, Stanton et al. (1991) define niche marketing as “a method to meet customer needs through tailoring goods and services for small markets”. Although the various aspects of the definitions are not completely identical, they all point out to two major aspects. That is, niche marketing involves specializing a company’s product offer and focusing on a specific group of customers’ needs. This is also stressed by Kotler (2003) who highlights specialization as the most important thought in niche marketing. It can exist on multiple levels, like geographical specialization, product specialization, service specialization, and more.

This study deals with companies from the package holiday industry. Based on the two aspects of niche marketing from literature, specialization and fulfilling a specific group of customers’ needs, I propose to differentiate between niche and mass players within this industry as follows. A mass player is a company that sells package holidays while not explicitly limiting its offerings in the sense of geographical area, type of holidays (e.g. winter sports, adventure), type of customers or specific customers’ needs. A niche player, however, does limit itself to one of these criteria and explicitly communicates this specialization on its website. Through this interpretation, the dataset comprehends three advertisers that are classified as mass players and seven that are classified as niche players. Considering the niche players, two companies focus on winter sports, one focuses on holidays outside of Europe, one focuses on sustainable holidays, one focuses on tailored holidays, one focuses on families, and the last focuses on religious people and corresponding holidays. Companies with a niche market focus are assigned dummy label 1, whereas companies with a mass market focus are assigned dummy label 0.

6. Research Methodology

with the remaining hypotheses (i.e. 2, 4-7) is described. These hypotheses are identical in terms of expected type of interrelationships and degree of measurement of the variables. This is why I describe the research method communally.

6.1 Mediation (H1)

According to MacKinnon et al. (2002) mediation analysis can be performed by either multiple regression or structural equation modeling (SEM). The first method is most commonly used in literature and is especially suitable if the sample size is over 500 (MacKinnon et al. 2002), which is the case in this study. The method involves a four step approach developed by Kenny and others (Baron & Kenny 1986; Judd & Kenny 1981). Figure 7 represents a visual representation of the steps involved. The basic idea of mediation analysis is to prove that the mediator seizes (some of) the explaining value of the independent variable in its relation to the dependent variable.

Applying the method to hypothesis 1 yields the following steps. First, the difference between the two groups (i.e. advertiser specific vs. generic keywords) of the independent variable wordographics regarding cost per click should be significant, which reflects the direct effect (path C). The first step is arbitral in the case of mediators that cause a counter effect (MacKinnon et al. 2002; Shrout & Bolger 2002). In this study, however, there is no reason to expect such effects. The second step is to show a significant difference between the two values of the independent variable wordographics regarding the mediator competitive intensity (path A). The third step is to show that the mediator competitive intensity is related to the dependent variable cost per click (path B), while controlling for the independent group variable wordographics (the link between wordographics and cost per click is in this case labeled path C’). The final step involves a comparison between the different paths. That is, the strength of the relation between wordographics and cost per click should be significantly and

Profit

Cost per click Path C

Figure 7. Visual representation of the mediation analysis involved in hypothesis 1.

Wordographics

advertiser specific vs. generic

Wordographics

advertiser specific vs. generic Competitive intensity Cost per click

Path C ’

substantially reduced (partial mediation) or absent (complete mediation) when the mediator is included in the model (path C vs. path C’). In addition to the original method to test whether this is indeed the case (Baron & Kenny 1986), researchers have proposed several alternatives. According to MacKinnon et al. (2002) the most commonly used and also suitable for this study are both the analysis proposed by Sobel (1982) and an increasingly popular approach called bootstrapping (Shrout & Bolger 2002). Bootstrapping is suitable when the sample size is particularly small, which is not the case in this study, or when the data is not normally distributed (Sobel 1982; Stone & Sobel 1990; Shrout & Bolger 2002; Frazier et al. 2004). An assessment of the normality of the distribution will point out whether the Sobel test or the bootstrapping method is preferred. The steps proposed by Preacher & Hayes (2004) serve as a guideline in conducting these tests.

Several issues regarding the power of mediation tests apply. To start with, mediation tests have generally low power (MacKinnon et al. 2002). A second issue is related to multicollinearity, which is inherent to mediation analysis (Kenny et al. 1998). That is, when more variance in competitive intensity is explained by wordographics, less variance of competitive intensity remains to contribute to cost per click. A large sample size can accommodate this effect (Kenny et al. 1998). Given the large sample size of 2046 cases, multicollinearity is not considered a problem in this study. Third, the effect size of path A should be equal to, or slightly larger than the effect size of path B (Kenny et al. 1998; Hoyle & Kenny 1999) to have optimal power in a mediation analysis. Fourth, the reliability of measure of the mediator can significantly influence the power of the test (Baron & Kenny 1986; Kenny et al. 1998). That is, lower reliability results in overestimation of path C’ and underestimation of path B. In the case of this study, reliability of competitive intensity is considered high, as the values are directly derived from actual behavior. Therefore, low reliability is not considered to be a problem in the analysis. What is finally reported in literature is the problem of reverse causal effects between the mediator and the dependent variable (Baron & Kenny 1986). However, in the case of this study, there is no reason to expect reverse causal effects between competitive intensity and cost per click. Therefore, this potential problem is not applicable to this particular research. Finally, covariates are added to the model since the hypotheses are dealt with in isolation.

6.2 Moderation (H2, H4 – H7)

and 7 (set B) deal with the moderating role of a company’s market focus, which is a dichotomy. Furthermore, the independent variable in the set B hypotheses is a dichotomy as well. Set A requires a hierarchical multiple regression approach (Baron & Kenny 1986; Aiken & West 1991). Set B can be tested by either using analysis of variance (ANOVA) or hierarchical multiple regression analysis. Although not very common in literature, I will use the latter option for set B as well, given the flexibility in options it provides for coding the dichotomous variables (Cohen et al. 2003; West et al. 1996).

Two general findings from literature point to the need to interpret the results of the moderation analyses with caution. First, as this study deals with empirical data, the power for detecting moderation effects is expected to be low (McClelland & Judd 1993). Second, the effect size for the interaction in a regression equation is generally quite low (Chaplin 1992). Considering the scale of measurement of the variables, both set A and B entail some specific considerations regarding methodology. Regarding set A, the reliability of measures of the independent and moderator variables is important given the fact it dramatically influences the interaction term, causing lower power of the test. As this study deals with empirical data, no problems are expected regarding reliability. Furthermore, these variables need to be normally distributed, as is a general assumption for any regression analysis. The power of the analyses in set B hypotheses is attenuated in the case of unequal sample sizes across the dichotomous variables (Aguinis 1995; Aguinis & Stone-Romero 1997). What is further reported regarding analyses involving categorical variables is the error variance across groups, which can lead to both overestimation and underestimation of the power of tests (Aguinis & Pierce 1998). Both these aspects are dealt with in the analysis. According to Jaccard & Wan (1995), the power of tests can be enhanced by adding covariates to the model, as will be done in this study. The fact that hypotheses are tested in isolation in this study also highlights this need.

scheme for the different values of the categorical variables is needed. West et al. (1996) provide insights into the several options available to researchers in coding categorical variables. Given their recommendations, both effect and dummy coding are suitable for the purpose of this study. The first scheme allows for testing for a significant difference between the slopes of the values of the moderator variable. The second scheme allows for testing whether or not these slopes statistically differ from zero. Additionally, dummy coding allows for easy interpretation of the coefficients of the predictor variable for different values of the moderator. Finally, a product term is created that represents the interaction between the moderator and the independent variable by simply multiplying these newly coded and standardized variables.

After standardization and coding is finished and an interaction variable is created, structuring of the regression equation can take place. This is a stepwise process (Baron & Kenny 1986; Aiken & West 1991; West et al. 1996). In the first step, the independent and the moderator variable serve as predictors of the dependent variable. In the second step, the interaction term is added to the equation. A moderating effect can be confirmed when the interaction effect in step 2 is significant and substantial and when explained variance is enhanced compared to step 1 (Aiken & West 1991;

Cohen et al. 2003; West et al. 1996).

7. Results

Table 1. Multiple regression results concerning mediation analysis (hypothesis 1)

The different steps in the mediation analysis (n = 2046): B SE B 95% CI β p

First step: path C

outcome: cost per click

predictor: wordographics (generic vs. advertiser specific) a

- 0.47 0.04 - 0.53, - 0.37 - 0.41 < 0.01 Second step: path A

outcome: competitive intensity

predictor: wordographics(generic vs. advertiser specific) a - 0.48 0.10 - 0.69, - 0,29 - 0.16 < 0.01 Third step: path B and C ‘

outcome: cost per click

mediator: competitive intensity (path B) predictor: wordographics 0.11 - 0.41 0.01 0.03 0.09, 0.13 - 0.48, - 0.35 0.30 - 0.37 < 0.01 < 0.01

Note. CI= confidence interval; a 0 = generic, 1 = advertiser specific; B = unstandardized coefficients; β = standardized coefficients

The coefficient concerned with the relation between wordographics and competitive intensity, which is path C’, is lower in the final regression analysis (B = - 0.41), but still significant (p < 0.01). Therefore, no support is found for complete mediation of competitive intensity in the relationship between type of wordographics and cost per click. A test needs to point out whether the drop in the coefficient for wordographics is significant (path C vs. C‘). Assessment of the normality of the distribution points out that there is no need to apply the bootstrapping method to this hypothesis. Consequently, the Sobel test is appropriate for testing partial mediation (Sobel 1982; Stone & Sobel 1990; Shrout & Bolger 2002). The drop in the unstandardized coefficients from - 0.47 to - 0.41, which is equal to the indirect effect (B = - 0.055), is significant (p < 0.01). Therefore, the results of the analysis are partially in support of hypothesis 1. Mediation takes place, but the direct effect remains significant. The fact that no support is found for full mediation is not surprising, since this is difficult for two reasons (Mackinnon et al. 2002). First, the use of empirical data makes it particularly difficult to find support for full mediation. Second, the effect size of path A (B = - 0.48) is substantially larger than the effect size of path B (B = 0.11), which makes it harder to find support for full mediation.

The results of the steps involved to test hypothesis 2 are depicted in table 2. Assessing the probability distribution points out that the requirement of normality is met. The moderator competitive intensity and the independent variable cost per click were standardized prior to the hierarchical multiple regression analysis, as is required for moderation analysis (Aiken & West 1991). As expected, the standardized variable cost per click has a negative impact on position (B = - 0.17, p < 0.05). The coefficient is negative because position improves with a decrease in the corresponding values (i.e. best position is 1, second best is 2, etc.). Furthermore, the interaction between competitive intensity and cost per click (B = - 0.30), which reflects the moderation effect, is significant (p < 0.01) and negative. The fact that the effect size is substantial is noteworthy, as interaction effects are usually very small when empirical data is used to test moderation (Chaplin 1992). The direct effects of competitive intensity and cost per click do not substantially change in the model that includes interaction. The increase of explained variance in the model that includes the moderation effect is 2%, on top of the 7% explained variance of the simple effects (R² change = 0.02,

p < 0.01). Thus, the analysis provides evidence for a negative moderating role of competitive

intensity in the relationship between cost per click and position, in support of hypothesis 2. The increase in explained variance does not seem to be very large. However, previous research points out that this is usually the case in examining moderator effects using empirical data (McClelland & Judd 1993).

Table 2. Hierarchical multiple regression results for the moderating effect of competitive intensity (H2).

Steps in the moderation analysis (n = 2046): B SE B 95% CI β p R² p

First step: regression without interaction moderator: comp. intensity (standardized) independent: cost per click (standardized)

- 0.32 - 0.17 0.07 0.07 - 0.44, - 0.19 - 0.30, - 0.03 - 0.22 - 0.11 < 0.01 < 0.05 0.07 < 0.01 Second step: regression with interaction

competitive intensity X cost per click

- 0.30 0.09 - 0.48, - 0.11 - 0.14 < 0.01 0.09 < 0.01

Note. dependent variable: position; CI= confidence interval; B = unstandardized coefficients; β = standardized

coefficients

The moderating role of company size is discussed next. It involves the positive impact on the relationship between position and click through rate (H4) and between position and conversion rate (H5). Necessarily, the variables position and company size are standardized prior to the hierarchical multiple regression analyses. Furthermore, assessing the probability distribution points out that the requirement of normality is met. The results of the analysis are depicted in table 3.

Table 3. Hierarchical multiple regression results for the moderating effects of company size (H4 & H5).

Steps in the moderation analysis (n = 2046): B SE B 95% CI β p R² p

Hypothesis 4 a

First step: regression without interaction moderator: company size (standardized) independent: position (standardized)

0.54 - 4.59 0.26 0.27 0.00, 1.04 - 5.12, - 4.05 0.06 - 0.51 0.09 < 0.01 0.25 < 0.01 Second step: regression with interaction

company size X position

0.74

0.33 0.08, 1.39 0.08 < 0.05 0.27 < 0.05

Hypothesis 5 b

First step: regression without interaction moderator: company size (standardized) independent: position (standardized)

- 0.23 - 0.21 0.19 0.03 - 0.37, 0.05 - 0.27, - 0.15 - 0.24 - 0.21 0.18 < 0.01 0.12 < 0.01 Second step: regression with interaction

company size X position

0.10 0.04 0.03, 0.18 0.09 < 0.01 0.15 < 0.01

Note. a dependent variable: click through rate; b dependent variable: conversion rate; CI= confidence interval; B = unstandardized coefficients; β = standardized coefficients

As expected, the results show a significant negative simple effect of position on click through rate (B = - 4.59, p < 0.01). The coefficient is negative because position improves with a decrease in the corresponding values (i.e. best position is 1, second best is 2, etc.). The regression coefficient that reflects the moderation effect is positive and significant (B = 0.74, p < 0.05). The direct effects of company size and position do not substantially change in the model that includes interaction. In addition, the explained variance of the model that includes the moderation term is increased by 2% (R² change = 0.02, p < 0.05).Therefore, H4 is accepted.