The effectiveness of Google AdWords advertising campaigns: investigating the Click-Through Rate, Conversion-Rate and Return-On-Investment

(1)

The effectiveness of Google AdWords advertising

campaigns: investigating the Click-Through Rate,

Conversion-Rate and Return-On-Investment

(2)

2

Title: Team identification as a key variable to distinguish supporters of a football club fan base

Author: Michiel van Gessel Department: Marketing Qualification: Master thesis Completion date: March 24, 2011

Supervisors: prof. dr. J. C. Hoekstra and dr. J. E. M. van Nierop

Title: The effectiveness of Google AdWords advertising campaigns: investigating the Click-Through Rate, Conversion-Rate and Return On-Investment.

Author: Tom Jansen

Department: Marketing

Qualification: Master thesis

Completion date: 15/11/2012

(3)

3

Management summary

This research investigated how to optimize Google AdWords advertising campaigns, and thereby takes all important aspects of the search term marketing process into account. Only sponsored advertisements that appear above, on the right hand side, or below the organic search results of Google are included in this research. This report is not focusing on just one important search term marketing metric, but it focuses on the three most important metrics: the Click-Through Rate (CTR), the Conversion-Rate (CR) and the Return-On-Investment (ROI). The CTR is the percentage of people that click on an advertisement after having an impression of it. The CR is the percentage of people that proceed to a certain desired action after clicking on an advertisement. Finally, the ROI measures if the advertisement campaigns were worth the money or not. The CR and ROI are really connected to each other, but still it is important to investigate the influence of certain variables on both metrics separately. The advertisement data used in this research comes from Catawiki, a small-sized enterprise that created an online platform for collectors. Catawiki is using Google AdWords, but the ROI of their advertisement campaigns is really bad. Catawiki has data of several advertisements of three different product groups: comics, stamps and coins. For companies like Catawiki it is really useful to get better insights in how different variables influence the CTR, CR and ROI. Three multiple regression models were conducted to get insights in these three important metrics.

As expected, the rank of an advertisement has a significant influence on the CTR (higher rank/position means higher CTR). Also more specific advertisement campaigns have significantly higher CTRs than less specific campaigns. Specific advertisement campaigns have really targeted keywords with targeted advertisements. Another outcome of this model is that advertisements with a large amount of words score significantly worse than

advertisements with a small or average amount of words. The above described effects were kind of expected, but this model also shows some really surprising outcomes. It was expected that featuring the keyword in the advertisement would improve the CTR, but the effect of this variable on the CTR is insignificant. Action triggers are often used in advertisements in order to persuade people to click on it. This research shows, in contradiction with earlier research, that a question mark, an exclamation mark and the interaction effect between an exclamation mark and an action word all have significant and negative effects on the CTR. The negative effect of the interaction effect is relatively the most important factor that influences the CTR. The effect of these action triggers is also tested with two different language groups as

moderators. It was expected that action triggers are more effective in countries with low power-distance-belief (because these countries are more sensitive for impulsive buying behavior, so more easily to convince with action triggers). The results show that exclamation marks can indeed better be used in low power-distance-belief countries, but question marks score better in high power-distance-belief countries. This outcome could indicate that a question mark is not really an action trigger.

(4)

4

clicked on. As expected, the more incongruence between the landing page and the advertisement, the lower the CR. Another factor that influences the CR negatively is the ‘’keyword featuring’’ variable.

The influence of rank on the ROI is also a positive and linear relationship. The specificity of advertisement groups really has a positive influence on the ROI, while the influence of specificity on the CR is not that large. This outcome is due to the fact that more specific advertisements have higher CTRs, and thereby lower CPCs. These lower CPCs result in lower total costs, which has a positive influence on the final ROI.

(5)

5

1. Introduction and problem statement

This report investigates how Google AdWords can be effectively used by companies that want to engage in the paid search term marketing tool of Google. Google AdWords is the tool of Google that gives companies the possibility to create and publish advertisements between the organic search results of users. Companies that want to engage in the paid search term marketing tool of Google make a bid for a certain word, or for a group of words. This word or group of words set by the advertiser is called the keyword. People that use the search engine type in a search query, which can match with the keywords a company has made a bid on. The results of the search queries are shown on top, on the right hand side, or below the non-sponsored results. Ghose and Yang (2009) indicate that these kind of advertisements are, because they are based on consumers’ own queries, far less intrusive than online banner ads or pop-up ads, which is a major advantage above other kind of advertisements. The search engine users are actively searching for products and/or services. The “keywords” in response to which the ads are displayed are often chosen based on user-generated content in online product reviews, social networks, and blogs where users have posted their opinions about a firms’ products, often highlighting the product features they value the most (Dhar and Ghose, 2009). In many ways, the increased ability of users to interact with firms in the online world has enabled a shift from “mass” advertising to more “targeted” advertising. Companies only have to pay when a customer clicks on their advertisement. The goal of most companies that participate in paid search term marketing is to convert visits to their website into desired actions like direct sales or subscriptions. There are also companies that engage in paid search term marketing to get brand awareness. The sales or subscriptions on a companies’ website must finally lead to a positive Return On Investment (ROI) for the total search term marketing campaign. Businesses spent worldwide about 8.5 billion dollar on paid search in 2004, and this amount was expected to grow to 16 billion dollar in 2009 (Jansen and Resnick, 2005). The real worldwide spending on paid advertisements in 2009 was 25.9 billion dollar, which indicates that this marketing technique is getting even more popular than was expected. According to Ghose and Yang (2009) the compound annual growth rate of the global search advertising market will be 37% in the next years. Google is worldwide, and especially in Europe, by far the largest search engine. This indicates that Google AdWords has the largest potential to reach a large amount of customers. Because of this large potential it is very interesting for companies to get insights in the effectiveness of different Google AdWords advertisements.

The first important aspect in search term marketing is the Click-Through Rate (CTR), which is the percentage of people that click on an advertisement after having an impression of it (in formula: ). Getting better insights in the CTR can be really important because there is a lot of competition for certain keywords. With high competitive keywords it is really important to design the advertisements in such a way that they

(7)

7

information. The conversion rate is an even more important metric than the CTR. Companies pay Google for every click made on an advertisement. When a company gets a lot of clicks, but no conversions, it can cost a lot without getting any revenue. At the end, the advertisement campaign should have a positive ROI (in formula:

, because otherwise the investment is larger than the revenues/benefits. In general, a higher CR means a higher ROI, because sales/subscriptions usually have a positive influence on the profit and the ROI. The previous statement is not true in all cases, so therefore it is also important to investigate the influence of certain variables on the ROI separately.

There are a lot of companies that do not have much knowledge about paid search term marketing, and lose a lot of money on it. To dig further into the effectiveness of different advertisements it is interesting to investigate which aspects of the individual advertisement have a positive- and which aspects have a negative influence on the CTR, CR and ROI. Keywords and landing pages could also have a big influence on the successfulness. A lot of companies are only focusing on one important outcome metric, instead of looking at the whole process. One of the companies that has problems with the above described search term marketing process is Catawiki, which is the specific focus of this study.

Catawiki is a company located at Assen that runs a website for collectors. The company was found in 2008 by René Schoenmakers and Marco Jansen. Catawiki is an online catalog for all objects that people collect. Catawiki also has a market place, and it is free to open a shop. Catawiki makes revenues by asking a small percentage commission costs for every sale. Catawiki has 70 product categories on their website, there are more than 10 million objects in the collections of the users, there are more than 1.5 million items in the central catalog, and there are more than 3.5 million items for sale at the moment. The users of Catawiki mostly come from the Netherlands and Belgium, but every day more and more people from

especially Germany, England and France are subscribing. Catawiki is really trying to increase sales and attract new subscribers by using Google Adwords. Catawiki started using Google AdWords to sell extra comics through the market place. The company created advertisements for every comic series and individual comic title that was present in the market place. These two advertisement groups aim for direct sales, and are purely focused on the Dutch market. Later, Catawiki also created general advertisements for people that collect comics, stamps and coins. These general campaigns have advertisements that aim for direct sales, and also

advertisements that aim for subscriptions. All these general advertisement campaigns are focused on Dutch, French, English and German speaking people. Users of Google first have to search for the keywords that Catawiki has made a bid on. Hereinafter, it is envisaged that the user pays attention to the advertisement and clicks on it. Catawiki is interested in both sales and subscriptions, because every subscriber has a certain lifetime value for the company. The Google software provides companies with very useful information about the effectiveness of different advertisements and different keywords. Catawiki also hired the company

(8)

8

categories and products. First, the company needs more insight in the effectiveness of several aspects of the advertisements. Catawiki has, especially through their cooperation with

Traffic4u, a lot of data available about the effectiveness of all different advertisements. Several useful analyses could be executed with this data. Simply stated, Catawiki wants more traffic to their website, and wants to convert this extra traffic into desired actions, so that the revenues outweigh the costs. Besides Catawiki, also other companies are really interested in getting better insights in the influence of several variables on the effectiveness of paid search term advertisements. With this information companies can establish their advertisements in a way that has a bigger chance of being profitable.

It is really important for companies like Catawiki to better understand the above described metrics. And despite several earlier research about the above discussed metrics, there are still variables not adequately studied yet. Also some researchers only investigated one important outcome metric, and were ignoring the other outcome metrics while drawing conclusions. Using Google Adwords is not about focusing on just one aspect of the total process, so further research is necessary. The whole process from getting clicks, till the final result (ROI) of advertisement campaigns should be taken into account. This research takes this whole process into account. The managerial problem of Catawiki could be used to add something to the current literature, and to give companies a direction in how to design a Google Adwords campaign effectively.

Based on the (gaps of the) current literature, the influence of several variable constructs on three dependent variables is investigated. The influence of the following variables on the CTR and CR is investigated: rank, the specificity of advertisements, the relevance of

advertisements and action triggers in advertisements. The influence of the length (in words) of the advertisement on the CTR is also investigated. Some variables are already investigated in previous research, but these variables have to be included because of their importance in the models. Leaving these variables out of the regression models would give biased results. Also the influence of these variables on both the CTR and the CR is examined in the same study now. The relevance variables are really important variables that were not previously studied. These variables investigate the influence of the relevance between the keyword and the advertisement on the CTR, and the relevance between the advertisement and the landing page on the CR. It is not just about the keywords, advertisements and landing pages itself, but the connection (relevance) between these three aspects is really important, and investigated in this research. The influence of rank and the specificity of advertisements on the CR could differ from the influence of these two aspects on the final ROI. Therefore the influence of these two variables on the ROI is also tested. Rank, specificity, relevance and action triggers are the four groups of independent variables, and the CTR, CR and ROI are the three dependent variables. Two different language groups are used as moderators. All variables will be discussed more extensively in the literature overview and data section.

(9)

9

‘’How can companies optimize several variables that influence the effectiveness of Google AdWords advertisements to get additional traffic to their website that will result in a positive ROI’’?

This research question results in the following sub-questions:

-‘’Which variables significantly influence the Click-Through Rate of Google AdWords advertisements’’?

-‘’Which variables significantly influence the Conversion-Rate of Google AdWords advertisements’’?

-‘’For which variables does the influence on the ROI differ from the influence on the CR’’? After this introduction the current literature about paid search term marketing is described. Based on this literature overview some hypotheses are drawn, which can be seen in the three conceptual models. The data section explains the data that is available for this research, and the final regression models are specified and validated in the method section. After this the models will be estimated, and the results will be published and discussed. Thereafter some conclusions are drawn, and based on these conclusions recommendations are given to

companies that want to engage in search term marketing. The report ends with limitations and recommendations for further research.

In the next chapter the relevant literature, with corresponding hypotheses, is described.

2. Literature overview

2.1 General

Nowadays, especially with the upcoming Internet possibilities, firms are able to choose between several channels and communication tools to interact with their customers. Allocating firm resources across channels and communication activities becomes an

(10)

10

and Herschel, 2008). Bowman and Narayandas (2001) found that customers that initiate contact with a company are more loyal, spend more and will engage more in positive word-of-mouth. Wiesel et al. (2011) researched three off-line marketing activities (fax, flyers, and catalog) and two online marketing activities (email and Google AdWords). From all these marketing activities only Google AdWords is a CIC. The ‘’purchase funnel’’ stages Wiesel et al. (2011) distinguished in their research were leads (info requests), quote requests and orders. The customers can move along this funnel through the online and offline channel. Wiesel et al. (2011) found several cross-channel effects, so off-line marketing activities could result in online orders and the other way around. For example 73% of the total profit impact of Google AdWords comes from off-line orders. Wiesel et al. (2011) also found that Google AdWords makes 55.72€ from every euro invested, which is 17 times higher than the estimated profit impact of faxes, which is the best scoring off-line (and FIC) activity. Firms must take the cross-channel effects and effectiveness of the CICs into account when allocating their marketing budget across different activities. Firms should especially focus on the

effectiveness of Google AdWords, because several researchers indicate that this instrument is really able to boost profits. Firms must not totally move to online marketing activities and the online sales channel, because, as is indicated above, online marketing activities could lead to off-line orders. For some companies offline marketing activities and/or offline sales channels really belong to the culture of their company. These companies can keep on using these offline marketing activities, but should reconsider the allocation of their marketing budget. As is indicated in the introduction this report will only focus on the marketing instrument Google AdWords. As is indicated above this tool is proved to be a very effective (paid search term) marketing instrument in the multi-channel marketing strategy.

(11)

11

percentage of the total number of searches. An advertiser must strive to find all keywords that are relevant for the company. A company can choose for a few keywords with high search volume, but a company could also choose for several low volume keywords. These low volume keywords can be difficult to discover, but therefore have less competition. According to Bartz et al. (2006) this lower competition results in a lower CPC (Costs-Per-Click). General keywords normally have high search volumes. Bidding on these general keywords really increases the number of impressions, but it is very likely that the advertisement is less relevant for the search query than when using more specified keywords. The competition is also higher for these general keywords, which can lead to higher costs and/or a lower position. Regelson and Fain (2006) indicate that, despite variation in advertisers, some (more specific) keywords have higher CTRs than other keywords. This has everything to do with the fact that some keywords have more relevance for a certain group of advertisers than other keywords have. Regelson and Fain (2006) also classified every keyword into a keyword cluster, and state that the CTR of a new keyword can be predicted by looking at the CTR of the keyword cluster it belongs to. Richardson et al. (2007) hypothesized that looking at advertisements with the same, or related, bid terms would be useful in predicting the CTR of a certain advertisement. The outcome of their research confirms this hypothesis. Ghose and Yang (2009) state that a lot of literature on sponsored search is done on the mechanism design of the auction concept, and not much research is done on variables that influence the CTR and CR. Thereby, Ghose and Yang (2009) state that given the shift in advertising from traditional banner advertising to search engine advertising, an understanding of the determinants of conversion rates and Click-Through Rates in search advertising is essential for both traditional and Internet retailers.

As is described above the first important thing for a company that engages in paid search term marketing is to find the most relevant keywords for their type of company/products. This keyword choice has an influence on the CTR, and also an influence on the CR. A lot of research is already done on the influence of the keyword choice and the keyword matching type. Besides these aspects the appearance of the advertisement itself also has an influence on the CTR and CR. The appearance of the advertisement is a really broad concept, because it contains a lot of aspects. The appearance of advertisements is about the title, description and visible URL, but it is not just about the words, signs or punctuations in it. An advertisement communicates a message, it gives a first impression to future customers, and creates a certain expectation to them. All aspects of the advertisement should cohere with each other, and should cohere with the landing page and company itself. The way companies fill in these advertisement appearance aspects really influences the CTR, CR and ROI. Companies should carefully analyze how to design the appearance of an advertisement. Ghose and Yang (2009) indicate that a limitation of their research is that it does not have any knowledge of

(12)

12

creates a certain expectation, which must be fulfilled on the companies’ website. Baye et al. (2009) also did an extended research on factors that influence the CTR, but they also lacked data about the appearance of the advertisement itself. The variables Baye et al. (2009)

investigated where the price of the product offered, the price of rival companies’ products, the number of rivals, and the location of the advertisement. Richardson et al. (2007) did research the influence of variables that have to do with the appearance of the advertisement on the CTR. However, this research does not indicate the direction and magnitude of the influences, which is very important information for managers. The main focus of this research paper is on the appearance of the advertisement itself, and the keyword choice is not taken into account. The different matching types are also not taken into account, the matching type of all

keywords in this research is ‘’broad matching’’. This research wants to explain which advertisement specific variables have an influence on the CTR, CR and ROI, and wants to indicate the magnitude of these variables.

2.2 Click-Through Rate

The conceptual model of the CTR model is shown below, followed by the relevant literature accoding to this model. To make this research more clear the hypotheses are shown between the relevant literature.

Figure 1: Conceptual model of the CTR.

Brooks (2004) indicates that the CTR is mostly influenced by the rank of the advertisement. Brooks (2004), Ghose and Yang (2009), Baye et al. (2009) and Agarwal et al. (2008) all indicate that the higher the location of the advertisement, the higher the CTR. Baye et al. (2009) also included a quadratic term for rank, and concluded that the negative relationship between rank and CTR increases at a decreasing rate. This finding has useful implications for managers interested in quantifying the impact of rank on CTR. Other researchers state that the relationship between rank and CTR is linear, so this quadratic term still needs further

investigation. Every company that engages in paid search term marketing should consider the extra costs and efforts that are necessary to increase in rank (Brooks, 2004). Search engines strive to improve the user experience with the search experience through quality search results. The ranking of the page has an impact on the satisfaction of the user (Witten et al, 1994). Page ranking affects search engine quality because it has been observed that in the real

(13)

13

world, most users only consume the top ranked results due to cognitive limitations, time constraints and other factors (Johnson et al., 2004). Bhargava and Feng (2005) investigated that the diminishing attention for search results has an exponential decay over rank, which confirms the importance of including the quadratic term of the rank variable when researching the CTR. Hypothesis one is based on this quadratic relationship.

H1: The positive relationship between rank and CTR is quadratic.

Based on the above described reasons a company, that wants a high CTR, must ensure that their advertisements will get a high rank in the search results. It is not that easy to get high positioned advertisements, because Google uses complicated systems in determining the rank. According to Bartz et al. (2006) there are three parties, with all different goals, involved in the paid search term marketing process. The user wants to receive the most relevant and

qualitative information on top of the results. The search engine wants to maximize their revenue from the advertisements. The revenue per search (RPS) for a search engine is: (CTR*CPC). Search engines cannot place all advertisements with high CPC on top of the results, because users can find these advertisements not relevant or of poor quality. To avoid losing customers, search engines rank advertisements not only based on the CPC, but also on a lot of other variables. The goal of most advertising companies is direct marketing, which means selling products or get subscribers. The common goal of the search engine and the advertising company is to get a high CTR, and users probably like advertisements with a high CTR, because otherwise they will not click on it. Every search engine takes a lot of quality factors into account, but it is for sure that the CTR is a very important factor in determining the rank. Google organizes an auction to determine the rank and CPC of all advertisements that have made a bid for a certain keyword. As is described above rank influences CTR, and CTR is one of the important factors that has an influence on the rank. It is also very important to have good insights in other factors than rank that have an influence on the CTR.

According to Brooks (2004) other important factors that influence the CTR are title, description, type of industry and the relevance of the ad according to the keyword that is searched on. The relevance of the advertisement can be increased by featuring the keyword in the advertisement. This is done by showing the keyword that is searched on bold in the title or description. Just as Regelson and Fain (2006), Ghose and Yang (2009) also state that there is a difference in CTR between search queries, and advertisements shown after the search query, that are specific, versus search queries (and subsequent advertisements) that are broad. This statement has nothing to do with the matching type of the keyword, just with the specificity of the search query and the subsequent advertisement. The influence of the specificity of

advertisements on the CTR is not adequately studied yet, so this research will test the influence of this variable.

H2: The more specific the keyword group and advertisements are, the higher the CTR.

(14)

14

grammar, brand recognition and, again, prominent featuring of the keyword term are

important, advertisement specific, aspects that influence the CTR. Stern (1988) indicates that the language of an advertisement is really important. An ad is approached as a literary work, similar to a dramatic monologue in form, in which imagistic language is used to create a company in persona, the embodiment of a firm’s mystique. The choice of words, images, grammatical style and rhetorical structure (length, flow of sentences, style) offers a basis of inference about the firm, product and consumer. An important part of the grammatical style of the advertisement is the length of the advertisement/number of words in the advertisement. Richardson et al. (2007) indicate that the number of words can influence the CTR, but this research does not state what the optimum amount of words is. Dyson and Haselgrove (2001) researched the influence of line length on the effectiveness of reading from a screen. They found out that a medium line length supports the effective reading from a screen. If line lengths are too long, the return sweeps to the beginning of the next line are difficult. If the lines are too short, readers cannot make use of much information in each fixation. Looking at this research it looks like advertisers have to search for an optimum length/number of words when creating an advertisement.

H3: An average amount of words has a more positive influence on the CTR than a small or large amount of words.

(15)

15

more precise because they indicate that people with high power-distance-belief (expecting and accepting power disparity) display less impulse buying. This research indicates that people with high power-distance-belief have more self-control. Hofstede (1983) researched countries according to four cultural dimensions, and one of these dimensions is power-distance-belief. Of the five countries interested for this research France and Belgium score significantly higher on power-distance-belief (68 and 65 points) than Germany, Great-Britain and the Netherlands (35,35 and 38 points). Looking at the above described researches it is expected that people in France and Belgium have more self-control than people in the other three countries. Since action-oriented advertisements can provoke impulsive behavior, it is expected that the effectiveness of action-oriented advertisements differs across these two different country groups.

H4: The presence of a question mark is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

H5: The presence of an exclamation mark is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

H6: The presence of an action word is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

Richardson et al. (2007) investigated the influence of several ad features and search terms on the CTR. This article focused on the pay-per-performance model with cost-per-click (CPC) billing, which is also the model that Google uses. The article tries to predict the CTR of new advertisements by using information of the advertisement itself, the page the ad points to, and statistics of related advertisements. Richardson et al. (2007) hypothesized some rough

advertisement quality categories that influence the CTR. This research used some

categories/constructs that are also described in the above part of the literature overview. This research clearly indicates which variables belong to which category/construct, which gives a nice overview. The relevant categories with their variables are described below.

-Words/signs: Is the advertisement aesthetically pleasing? Variables that have to do with words and signs are the following:

-Number of words in the title.

-Number of words in the description. -Long or short words in the description.

-Number of exclamation points, dollar signs, or other punctuations.

-Action capture: Does the advertisement draw the user in? Variables that have to with this category are:

- Are there action words in the title or description? Action words are words like: ‘’buy’’, ‘’subscribe’’ and ‘’join’’.

(16)

16

-Relevance: How relevant is the ad for the search term query?

-Does the bid term appear in the tile and/or description exactly?

-Do any subsets of the term appear in the title and/or description?

The result of the analysis of Richardson et al. (2007) is that advertisement quality influences the CTR significantly. Richardson et al. (2007) state that taking the above described variables into account makes the prediction of the CTR of an advertisement more accurate. A

shortcoming of this research is that only predictions of new advertisements can be made, but nothing is known about the magnitude of the influence of the variables separately.

The model that should predict the CTR in this research is based on the above described literature and the available data.

2.3 Conversion-Rate and Return-On-Investment

The conceptual models of the CR and ROI models are shown below, followed by the relevant literature accoding to these models.

Figure 2: Conceptual model of the CR.

Figure 3: Conceptual model of the ROI.

The factors that may influence the CTR of an advertisement are described now, which is only one part of the search engine marketing process. A lot of researches finish their investigation at this point, and draw conclusions about the CTR on its own. Other researchers start their investigation at this point, and only describe important factors in determining the CR. It is more effective to test the influence on both metrics in the same study, because everything

(17)

17

coheres with each other. When users of Google have clicked on an advertisement they will be directed to the companies’ website. From this moment the advertiser pays a certain amount to Google. The advertisement is only lucrative for the advertiser when the visits to their website convert into sales or subscriptions. There are also a lot of factors/variables that influence the CR. Moe et al. (2004) indicate that typical conversion rates are rarely exceeding five percent, and managers are struggling to understand conversion behavior at their website. Moe et al. (2004) stress out the importance to account for various patterns in the relationship between visiting and purchasing. This is important because customers can have different reasons for visiting a retail website. Moe et al. (2004) researched these different purchase patterns of different people, and indicate that the response of the (potential) customers can vary per person. This makes it very difficult to find out how to organize the landing page. Moe et al. (2004) also gave several recommendations for further research. The sequence of the pages viewed after a customer has clicked on an advertisement and the website design/appearance really influence the conversion rate. The credibility of the website is also an important factor for customers to buy a product online. Customers are more likely to buy from a website that has a good reputation. Companies can increase the credibility by publishing their contact details on the website, having a privacy policy, and having a high quality website. Strouchliak (2009) indicates that the average online conversion rate is around 2.3%, with the highest at approximately 9%. Strouchliak (2009) stresses the importance of the demographics of the potential customers that visit the website. Gender, income, occupation, age, education etc. all influence how customers respond to different landing pages. Also psychographics, the personalities of the customers, influence the conversion rate of a website. Furthermore

Strouchliak (2009) indicates that firms should clearly communicate their unique-selling-point to the customer. This in combination with an attractive design, a help-option, and guarantees should persuade the customer to buy or subscribe.

The variables described above all have to do with the design of the landing page. As is described earlier this research is interested in the influence of the advertisement itself. The appearance of the advertisement is not just influencing on the CTR, but also the CR and the ROI. It is expected that the specificity of the keyword and advertisement is influencing the CR. When search queries of customers are really targeted, and companies can meet the expectations of these targeted search queries, it is likely that the CR will increase.

H7: The more specific the keyword group and advertisements are, the higher the CR.

Ghose and Yang (2009) state that the relevance of the landing page with the advertisement is also an important factor in determining the CR. Ghose and Yang (2009) state that anecdotal evidence suggests that if online consumers use a search engine to direct them to a product, but do not see it addressed adequately on the landing page, they are likely to abandon their search and purchase process. People have an expectation when they click on an advertisement, and therefore it is very important that the landing page fits with the advertisement. This statement of Ghose and Yang (2009) still needs to be translated into hypotheses that will test the

(18)

18 H8: More relevance between the advertisement and the landing page by using the same title on both has a positive influence on the CR.

H9: The more incongruence between the advertisement and the landing page, the lower the CR.

It can be the case that some search engine users are impulsively clicking on a certain

advertisement because they were persuaded by an action trigger (for example an action word or exclamation mark), or a featured keyword, while they are not at all interested in the website they are landing on. For this reason the action triggers and keyword featuring can have a positive influence on the CTR, but a negative influence on the CR, and finally on the ROI. It is really important to test these assumptions, because other researchers only investigated the influence of these variables on the CTR, and published positive results about these variables.

H10: The presence of a question mark in the advertisement has a negative influence on the CR.

H11: The presence of an exclamation mark in the advertisement has a negative influence on the CR.

H12: The presence of an action word in the advertisement has a negative influence on the CR. H13: There is a negative interaction effect between an action word and an exclamation mark according to the CR.

H14: Featuring the keyword has a negative influence on the CR.

(19)

19 H15: Rank and ROI are expected to have an inverted U-shape relationship.

H16: The positive influence of a specific vs. a general keyword, and thereby specific vs. general advertisement, is more positive for the influence on the ROI, than for the influence on the CR.

A lot of literature about factors that influence the CTR, CR and ROI is described now. Now it will be made clear what this report adds to the already existing literature about search engine advertising. Also the main focus of this study is made clear now.

As is already explained earlier, the main focus of this paper is the appearance of the advertisements themselves, and the relevance of the advertisement appearance with the keyword and the landing page. In this paper the keyword a consumer is searching on does not matter in determining the CTR. Every advertisement that is tested in this research is found by keywords that belong to a certain advertisement group and all keywords have the same matching type: broad. This research is interested in the influence of the appearance of an advertisement, given a certain keyword group and keyword matching type. According to the CR this research is only focusing on properties of the advertisement, and the relationship between the advertisement and the landing page. This research does not take specific landing page properties into account. Furthermore this research investigates whether there are

differences between two country groups. According to the final goal of most Google

AdWords campaigns, this research not only tests the influence of certain variables on the CR, but also on the ROI. As is described above it is expected that some variables have a certain influence on the CR, but there influence on the ROI could differ. In some cases it is really important to investigate the influence on the CR and the ROI separately. A lot of research papers only tested the influence of several variables on the CTR of advertisements. Most of these papers conclude that rank is the most important variable for companies to take into account when engaging in paid search term marketing. However, in paid search term

marketing, it is not all about CTR, but also about CR, and finally about the profit and ROI of the campaign. This paper takes all these aspects into account.

4. Hypotheses

To make something clear: when this report talks about a higher rank, then this will also mean a higher position in the search results. This section summarizes the hypotheses stated above, which will give a clear overview of what this research is about.

4.1 Click-Through Rate

Most researchers indicate that rank is the most important factor that influences the CTR of an advertisement. Some researchers indicate that the relationship between rank and CTR is increasingly decreasing.

H1: The positive relationship between rank and CTR is quadratic.

(20)

20

The length of the text could also have an influence on the CTR. It is expected that advertisements with an average amount of words have the highest CTR.

H3: An average amount of words has a more positive influence on the CTR than a small or large amount of words.

People in France and Belgium have more power-distance-belief than people in the

Netherlands, Great-Britain and Germany. For this reason it is expected that the action oriented advertisement style has a larger positive influence on the CTR than in the last three countries. H4: The presence of a question mark is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

H5: The presence of an exclamation mark is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

H6: The presence of an action word is more efficient in the Netherlands, Great-Britain and Germany than in France and Belgium.

4.2 Conversion-Rate

It should be taken into account that the influence of the variables in the below stated

hypotheses is the same as their influence on the ROI. The variables that are expected to have a different influence on the ROI than on the CR are described in section 4.3

It is expected that specific keywords and advertisements have a higher CR than general keywords with their corresponding general advertisements.

H7: The more specific the keyword group and advertisements are, the higher the CR.

It is expected that the more relevant the landing page is according to the advertisement a user has clicked on, the higher the CR will be. A landing page could look very relevant according to the advertisement when the title of the landing page is the same as the title on the

advertisement. A landing page could look really irrelevant to the advertisement when the text of the advertisement is incongruent with the content on the landing page. The data section discusses this in more detail.

H8: Creating more relevance between the advertisement and the landing page by using the same title on both has a positive influence on the CR.

H9: The more incongruence between the advertisement and the landing page, the lower the CR.

As is discussed above, action oriented advertisements and advertisements where the keyword is featured could trigger the user to click on it, without being really interested in the

company/product. Therefore it is possible that action triggers and featuring the keyword is negatively influencing the CR.

(21)

21 H11: The presence of an exclamation mark in the advertisement has a negative influence on the CR.

H12: The presence of an action word in the advertisement has a negative influence on the CR. H13: There is a negative interaction effect between an action word and an exclamation mark according to the CR.

H14: Featuring the keyword has a negative influence on the CR.

4.3 Return-On-Investment

The variables that are expected to have a different influence on the ROI than on the CR are described now.

Rank could have a positive, and linear, influence on the CR, but have a quadratic relationship with the profit/ROI. This all has to do with higher costs that could be associated with higher ranks. Rank probably does have a positive influence on the CR, but rank and ROI are expected to have an inverted U-shape relation.

H15: Rank and ROI are expected to have an inverted U-shape relationship.

Because more targeted and specific advertisements probably have a higher CTR, and have less competition with other companies for advertisement space, it is expected that the CPC is lower than for more general advertisements. This lower CPC will result in higher profits for a certain advertisement.

H16: The positive influence of a specific vs. a general keyword, and thereby specific vs. general advertisements, is more positive for the influence on the ROI, than for the influence on the CR.

5. Data

Based on literature the influence of several concepts/variables on the CTR, CR and ROI is hypothesized. How the influence of these concepts could be tested depends on the available data, which is described below.

The advertisements that are used in this paper all started in October, and the data of these advertisements till June is used. Every row in the data set represents a single advertisement. The data of 1160 advertisements is available, and the advertisements come from five different countries: the Netherlands, Belgium, Germany, France and the United Kingdom. Of all

advertisements 437 are targeted on comic collectors, 288 on coin collectors, and 435 on stamp collectors. After removing the outliers 515 advertisements will be used for the model

(22)

22

created based on the available data. Unfortunately, Catawiki has not the data available to test the influence of all variables discussed in the literature overview. There are also some

variables that do not vary that much between the different advertisements, and could therefore not be tested as well. The hypothesized constructs and/or variables are described below.

5.1 CTR

Language

As is described in the literature overview the five languages are divided into two different language groups. The Netherlands, Great-Britain and Germany belong to the same group, as well as Belgium and France.

Specificity

All advertisements belong to a certain group. Advertisements in the same group will be found with the same group of keywords. There are general advertisement groups, where

advertisements are found when people are searching for general keywords like ‘’comics’’, ‘’coins’’ or ‘’stamps’’, and there are also other groups of advertisements that are more specific. These groups are more specific because the groups are about specific sorts of comics. These groups of advertisements are found when people are searching for more specific keywords like ‘’graphic novels’’, or ‘’manga’’. Furthermore there are groups of advertisements that are even more specific. These advertisements are found when people are searching for keywords like ‘’collecting comics’’ or ‘’comic catalogue’’. These keywords consist of two or more words and are therefore more targeted than the other groups. The advertisements in these groups are also really targeted. These specific keywords and specific advertisements really suit to the service that Catawiki delivers.

Number of words

The ‘’number of words’’ variable is split into three categories: a small amount of words, an average amount of words, and a large amount of words. The advertisements that are available have a minimum of seven words, and a maximum of twelve words. Advertisements of seven or eight words are classified into the ‘’small’’ group, advertisements with nine or ten words in the ‘’average’’ group, and advertisements with eleven or twelve words in de ‘’large’’ group.

Rank

(23)

23

when using the centered variables (Aiken and West, 1991). First the rank variable is centered, and rank2 is calculated from the centered rank variable.

Relevance (between keyword and advertisement)

The advertisement is more relevant for a user when the keyword that is searched on is featured in the title of the advertisement. This featuring is done by showing the keyword that is searched on, and making this word or small sentence bold. About half of the advertisements in the data-set have this prominent featuring of the keyword. This variable has two values: featuring of the keyword in the title vs. no keyword featuring.

Action-orientation

The variables that have to do with the aggressiveness/action-orientation of the advertisements are the number of action words in the description, the presence of an exclamation mark in the description, or a question mark in the title/description. The action words that Catawiki uses in their advertisements are: buy, sell, register, visit and check. The goal of these words is to persuade the user to take a short-term action: click on the advertisement. The question mark in the title or description means that the advertisements are asking a question to the user. The goal of these question marks is that users give affirmative answers to the questions in their mind, and therefore click on the advertisements. The exclamation mark in the description is also an action trigger, because it makes the advertisement look more important. The available advertisements have zero, one or two action words, zero or one question mark and zero or one exclamation mark in it. Also the interaction effect between an action word and an exclamation mark, and the moderating effect of language according to all these action-oriented variables, are tested on the CTR. In testing this moderating effect, the product of the centered variables is used. This is also done to reduce multicollinearity.

The variables that influence the CTR are summarized in the following table.

Construct Variable

Words/signs Number of words

Rank Rank

Rank2

Specificity Keyword/advertisement group

Relevance (between keyword and advertisement)

Featuring of keyword in the title

Action words

Exclamation mark in description Question mark in title

Action word*Exclamation mark

Moderator Language

(24)

24

5.2 CR

Rank

This is already described in the CTR section above.

Specificity

This is already described in the CTR section above.

Relevance (between advertisement and landing page)

Relevance between the advertisement and the landing page can be increased by using the title of the advertisement also as landing page title, which probably has a positive influence on the CR. This variable has two values: same title or not. This variable is called:

‘’title-congruence’’. The other variable about relevance is called ‘’in‘’title-congruence’’. This

incongruence between the advertisement and the landing page can occur in three different levels. The first level is no incongruence at all, so these advertisements really connect to the content on the landing page. There are also advertisements with some incongruence: the title of the advertisement does not clearly indicate what the landing page is about. For example, the title of some advertisements is about stamps in general, and then the future customers are directed to the marketplace section of the website. In this case the title does totally not indicate that the advertisement will lead to a marketplace. The last level of incongruence is about advertisements with a large amount of incongruence: in these advertisements the title and the description do both not indicate what the landing page will be about. The

incongruence variable is ordinal, with three different levels.

These are the same action-orientation variables described in the CTR section, so again the centered variables are used.

Relevance (between keyword and advertisement) Again the keyword featuring variable is used here.

The variables the influence the CR summarized in the table below.

Rank Rank

(25)

25

Action-orientation Action words (buy, sell, register, visit and

check)

Exclamation mark in description Question mark in title

Action word*Exclamation mark

Relevance (keyword and advertisement) Featuring of keyword in title

Table 2: variables that influence the CR.

5.3 ROI

For the ROI the rank, rank2 and specificity variables are used, which can be seen in the following table.

Rank Rank

Rank2

Table 3: variables that influence the ROI.

6. Method

6.1 Model specification

On the basis of the theoretic background and the available data three predictive models can be specified. The dependant variables are the CTR, the CR and the ROI, and all independent variables are already described above. Also several variables that are already researched by other researchers are included in this research. These variables are important factors in determining the dependent variables, so excluding them from the analyses would give biased results. Multiple Linear Regression (MLR) is used to predict the dependent variables from the independent variables. MLR is based on least squares: the model is fit such that the sum-of-squares of differences between observed and predicted values is minimized. This linear model is chosen above the multiplicative model. An important property of the multiplicative model is that it includes interaction effects between all independent variables (Wieringa, 2010). This could be really useful, but only one interaction effect is expected in this study. Therefore, in this research it is unnecessary to account for interaction effects between all variables. Multiplicative models also require a log transformation before, and an antilog transformation after estimation. These steps do not need to be done with linear models, which is more efficient. The three specified MLR models are shown below.

CTR

Y = B0 + B1X1+ B2X2 + B3 X3 + B4 X4+ B5 X5+ B6 X6 + B7 X7 + B8 X8+B9X9 + B9(X8* X9)+ E

(26)

26 X1=Rank

X2=Rank2

X3=Number of words group 2 (dummy variable, group 1 is reference group)

X4=Number of word group 3 (dummy variable)

X5=Specificity

X6=Keyword featuring (0=no, 1=yes)

X7=Question mark (0=no, 1=yes)

X8=Exclamation mark (0=no, 1=yes)

X9=Action word

X8*X9=Interaction effect between an action word and an exclamation mark

E= error CR Y = B0 + B1 X1 + B2X2+ B3X3 + B4X4 + B5X5 + B6X6 + B7(X5*X6) + B8X7 + B9X8+ E Where Y= CR X1=Rank X2=Specificity

X3=Keyword featuring (0=no, 1=yes)

X4=Question mark (0=no, 1=yes)

X5=Exclamation mark (0=no, 1=yes)

X6=Action word

X5*X6=Interaction effect between an action word and an exclamation mark

X7=Title-congruence (0=no, 1=yes)

X8=Incongruence

E= error

ROI

The normal rank variable (un-squared) is also included, because when inserting the square variable, the normal variable also need to be included.

(27)

27 Where Y= ROI X1=Rank X2=Rank2 X3=Specificity E= error

Before interpreting the results the models need validation.

6.2 Model validation

This part describes the validation of the model. Validation assesses the quality of the model (Leeflang et al, 2000). When the model is not validated some transformations need to be done to overcome the problems. Without these important transformations the estimation is of little value. Several important aspects in validating the model are described below.

First outlier check

Before running the specified model some important checks have to be done first. A lot of advertisements in the data-set are useless because these advertisements have very limited data. The CTR of some advertisements is really large because these advertisements only have a few impressions. These advertisements, outliers, were filtered out before running the model. Outliers also become clear when looking at the residuals after running the model, however in this case it is important to check for the largest outliers before running the model. Because there are so many outliers, which all influence the regression model, it could be hard to discover the real outliers from the residuals after esitmation. When looking at the CTR of all advertisements in a scatter dot it becomes clear that advertisements with a CTR of five percent or higher are outliers. The advertisements with a CTR of five percent or higher have an average of 150 impressions. This indicates that this should be the minimum amount of impressions an advertisement should have when taken into the model that predicts the CTR. The same analysis is done for the CR and ROI models. This analysis indicates that an advertisement should have 25 clicks or more when taken into the model that predicts the CR and the ROI. After deleting the advertisements that do not meet the criteria, 515

advertisements are available for further analyses. Normality and linearity

(28)

28

concluded that the residuals of all three models are far from normal, and the relationship between the variables is non-linear. To solve this problem the log transformation of the three dependent variables is done. After this transformation the model that predicts the CTR is normally distributed and linear, but the CR and ROI models are not. To solve the problems for these two variables the Box-Cox transformation is used.Besides traditional

transformations (like the log transformation), the Box-Cox transformation (Box & Cox, 1964) represents a family of power transformations that incorporates and extends the traditional options to help researchers find the optimal normalizing transformation for each variable (Osborne, 2010). As such, Box-Cox represents a potential best practice where normalizing data or equalizing variance is desired. The formula of the BoxCox transformation is: ((Y^ λ -1)/ λ), where λ means Lambda.Given that λ can take on an almost infinite number of values, the optimum λ can be found to maximize normality. Osborne (2010) suggests some standard lambdas, which stand for basic (traditional) transformations. These suggested lambdas are: 0.50 (square root transformation), 0.33 (cube root transformation), 0.25 (fourth root

transformation), -0.50 (reciprocal square root transformation) and -1 (reciprocal

transformation). The effects of all these transformations were tested for the models that predict the CR and the ROI. When looking at the normal Q-Q plots, the histograms of the residuals, and the two plots that check for linearity, it can be concluded that the square root transformation has the best result. These outcomes are still not perfect, but the plots and histograms show that the models are quite linear and normal. The Kolmogorov-Smirnov tests and the Shapiro-Wilkinson tests also confirm that both models are normally distributed now. The final plots and histograms that check for linearity and normality are shown below.

.

Figure 4: Observed vs. predicted values of CTR Figure 5: Residuals vs. predicted values CTR model

(29)

29

Figure 7: Observed vs. predicted values of CR Figure 8: Residuals vs. predicted values CR model

Figure 9: Histogram of residuals CR model

Figure 10: Observed vs. predicted values of ROI Figure 11: Residuals vs. predicted values ROI model

(30)

30 Pooling

Pooling data means estimating a single model for similar data of different groups. In this research it is tested if the data of the three different products could be taken together. It has the advantage that the number of parameters decreases (i.e. the number of degrees of freedom increases), which will give more certainty to the estimated coefficients. However, coefficients will be biased in the regression equation if coefficients differ among groups (products). Therefore pooling is only allowed when it can be assumed that there is homogeneity in the parameters across groups. Bass & Wittink (1975) state that it is important to estimate the relationships for each section separately and then test the homogeneity hypothesis. To test whether the coefficients in linear regressions on different data sets are equal the Chow-test can be conducted. The Chow-statistic (F) can be calculated as follows:



 



unpooled unpooled unpooled pooled unpooled pooled df SSR df df SSR SSR F / /    With:

SSRpooled = the sum of the squared residuals in the pooled sample

SSRunpooled = the sum of the squared residuals from the separate sections

dfpooled = n*K – L

dfunpooled = n*(K - L)

Where:

n = the number of sections

K = the number of observations per section L = the number of parameters in the model

The above stated formula is used to determine whether the data can be pooled for the three specified regression models.

CTR:



 



51 . 1 46 . 1 ) 337 , 28 ( 05 . 2 337 / 316 337 365 / 316 95 , 369       F F

(31)

31 CR:



 



67 . 1 57 . 1 ) 187 , 19 ( 45 . 1 187 / 34 , 548 187 206 / 34 , 548 309 , 628       F F

1.45 is not in the critical area, so pooling is allowed when estimating the CR model. ROI:



 



57 . 1 ) 186 , 20 ( 29 . 2 186 / 678 , 130 186 206 / 678 , 130 620 , 162      F F

2.29 lies in the critical area, so pooling is not allowed when estimating the ROI model. Before drawing conclusions about these outcomes some robustness checks are done. Robustness checks

According to the Chow-test two of the three specified regression models cannot be pooled across products. This can make this report really inefficient because seven (instead of three) models have to be validated and estimated, and this will result in a lot of analyses, tables and graphs. However, robustness checks can indicate that the pooled models are equally suitable as the un-pooled models, which means the pooled models can be used for further analyses. According to White et al. (2010) a robustness check is an exercise where the researcher examines how certain ‘’core’’ regression coefficient estimates behave when the regression specification is modified in some way. Leamer (1983) influentially advocated investigations of this sort, arguing that "fragility" of regression coefficient estimates is indicative of

specification error, and that sensitivity analyses (i.e., robustness checks) should be routinely conducted to help diagnose misspecification. A finding that the coefficients do not change much before and after the regression models are modified is taken to be evidence that these coefficients are ‘’robust’’. If the signs and magnitudes of the estimated regression coefficients are also plausible, then this is commonly taken as evidence that the estimated regression coefficients can be reliably interpreted as the true causal effects of the associated regressors, with all that this may imply for policy analysis and economic insight (White et al. 2010). According to Wieringa (2010) the parameters should have correct signs, and the outcome of the models should be plausible. This indicates that when filling in normal values in the

(32)

32

the models is plausible, it is permitted to estimate the three pooled models instead of the un-pooled models.

The outcome of the above described analyses indicates that it is allowed to continue the analyses with the three models. For the CTR models there is not any difference in the sign of the coefficients when comparing the outcome of the pooled model with the three separate models. The only difference is that some variables are significant in the pooled model, but are not significant in the separate models. The reason for this insignificance might be the lack of data when only estimating the models for one product. This shows even more that the pooled models are probably better than the un-pooled models. When comparing the ROI models there is also no difference in coefficient signs, which is a good sign. All three models have plausible outcomes when filling in some extreme values for the independent variables. Before estimating the model some other validation checks needs to be done.

Heteroskedasticity

Heteroscedasticity refers to the assumption that the dependent variable has different amounts of variance across the range of values for an independent variable. Because methods in regression analysis depend on an assumption of equal variance, the presence of

heteroscedasticity may reduce the efficiency of the parameter estimates. There are several methods of testing for the presence of heteroscedasticity. The Levene's test is used to assess the equality of variance between the three different product groups when predicting the models. The null-hypothesis of the Levene’s test states equality of variance between groups. The formula of the Levene’s test is:









_{ }







        _k i N j ij i k i i i i Z Z Z Z N k k N W 1 1 . 1 2 2 .. 1 Where:

W = the result of the test,

k = the number of different groups to which the samples belong N = the total number of samples,

Ni = the number of samples in the ith group Yij = the value of the jth sample from the ith group Zij = YijYi , withY = the mean of the ith group i Z.. = the mean of all Zij

Zi = the mean of the Zij for group i.

(33)

33

the predicted values. The plot of residuals versus time is in this study not important because all advertisements in the data-set are from the same period. Heteroscedasticity may also have the effect of giving too much weight to a small subset of the data (namely the subset where the error variance was largest) when estimating coefficients. This is not the case whenthe residuals and the predicted values are independent of each other.The plots, which can be found at the linearity and normality section (figures 5, 8 and 11), also indicate that heteroscadasticity is not influencing any of the models.

Serial correlation

When OLS is applied, the disturbance terms, or errors, should be uncorrelated over time (Leeflang et al., 2000). Although it does not bias the coefficient estimates, the errors tend to be over- or underestimated if they depend upon their predecessor. According to O’Halloran (2005) it is important to plot the residuals versus time when the data is collected over time. The residuals of t=0 should not be correlated with the residuals of t-2 etc. This study does not have time-series data, so serial correlation cannot influence the outcome of the models. In this research every data-point is an advertisement, and these advertisements were active during the same period of time.

Multicollinearity

Multicollinearity refers to the situation that some independent variables (IV) tend to be highly correlated with another IV, resulting in unreliable parameter estimates (Leeflang et al, 2000). This means that although IV's may be predictors for a dependant variable (DV), they could correlate amongst each other. Therefore these IV's partly depend on other variables and thus are not completely independent as to their effect on the DV.

Farrer & Glauber (1967) state that multicollinearity emphasizes the inability to distinguish the independent contribution to explained variance of an explanatory variable that exhibits little or no truly independent variation. Simply stated, multicollinearity increases the standard errors of the coefficient. Thereby it might turn out that certain IV's do not significantly differ from zero, where they would have when standard errors were lower. In other words,

multicollinearity misleadingly inflates the standard errors, which makes some variables statistically insignificant while they should be significant, or the other way around. Therefore might the presence of multicollinearity be a treat to effective estimation of a relationship between variables (Farrer & Glauber, 1967).

The effectiveness of Google AdWords advertising campaigns: investigating the Click-Through Rate, Conversion-Rate and Return-On-Investment