Effect of TV advertising on search engine marketing in the Digital Era

MSc in Business Administration

Specialisation: Digital Business

Effect of TV advertising on search engine marketing in the

Digital Era.

Master’s Thesis Jèva Koop 10809791

Supervisor: Shan Chen Amsterdam: June 22nd, 2018

2

Statement of originality

This document is written by Jèva Koop who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

3

Abstract

Advertising expenditures account for one of the biggest expenses organisations have nowadays. Therefore, it should work to its full potential and result in a competitive advantage for companies. Marketeers and researchers see that advertising has moved to a different direction due to digitalisation, online advertising, and an increase in search engine usage. Due to these changes organisations struggle to find the best possible marketing strategy to maximise their sales. However, it is proven that TV advertising is still one of the most effective marketing channels. Hence, this study believes that traditional marketing has a strong significant effect on search engine marketing (SEM) and that these two channels complement each other. A debate in literature has been recognised on the effectiveness of TV advertising, which led to a recognition of a literature gap. The effect of TV advertising on search engine advertising (SEA) has not been studied thoroughly. Therefore, this study wished to examine this relationship. There is reason to believe that there are other factors that influence this relationship such as: TV channel type; and the difference between branded and generic paths. In cooperation with a Dutch company, called Pricewise, a sample of 172 days was collected and used to find significant relationships between the two marketing channels. Many correlations have been statistically proven, showing strong effects between variables. Furthermore, comparisons were emphasised between: (1) commercial and government-owned channel types; (2) volume and rate; (3) branded and generic paths. The general conclusions of this study show that: (1) tv advertising has a strong effect on branded SEM; (2) volume is a better measure of success; and (3) channel type has an effect on SEA. Furthermore, this study shows direct effects and provides a starting point for future research.

Keywords: SEA, SEM, Branded, Generic, TV advertising, Marketing, CTR, CR, Conversion,

4

Table of Content

1.2 GROSS RATING POINT 10

2. SEARCH ENGINE ADVERTISING (SEA) 10

2.1 SEMCLICKS 12

2.2 SEMCONVERSION 13

2.3 CLICK-THROUGH RATE 14

2.4 CONVERSION RATE 15

METHODOLOGY 18

1. PROCEDURE 18

2. SAMPLE AND DATA COLLECTION 18

3. VARIABLES AND MEASURES 18

3.1 INDEPENDENT VARIABLES 19 3.2 DEPENDENT VARIABLES 19 3.3 CONTROL VARIABLE 21 RESULTS 22 1. STATISTICAL PROCEDURE 22 2. RESULTS 22 2.1 MODEL 1 22 2.2 MODEL 2 26 2.3 MODEL 3 30 2.4 MODEL 4 34 DISCUSSION 38 1. DISCUSSION OF FINDINGS 38 1.1 HYPOTHESES FINDINGS 38 1.2 MODEL FINDINGS 44 2. PRACTICAL IMPLICATIONS 46

3. FUTURE RESEARCH AND ACADEMICAL IMPLICATIONS 47

4. LIMITATIONS 49

5. CONCLUSION 50

5

Introduction

In 2013, the worldwide advertising expenditure amounted to $516.5 billion and it is expected to grow up to $667.65 billion by 2018 (Statista, 2014). In the 21st century one of the most researched topics has been digitalisation. It has also been one of the most important news topics throughout the last ten years. However, even in the booming online environment it is hard to forget the importance of traditional marketing channels, such as television. On average the ratio between televisions and people in an American household is almost one, meaning there are as many televisions as people in the United States (Nielsen, 2011). At this time, news and information comes from many different sources including: newspapers, social media, and television. The use of online marketing channels is increasingly expanding. But it has been proven that in many countries TV remains the most trusted source (Danaher & Rossiter, 2011). Despite this trust in traditional marketing, digital marketing tools are becoming increasingly essential sources of competitive advantage (Zanetti, Bijmolt, Leeflang & Klapper, 2014). For instance, online searches have drastically increased. On average, users search Google at least twice a day. This is a relatively large amount due to the fact that Google did not exist a couple of decades ago (ComScore, 2012).

There are many companies that experiment with using congeneric messages across multi-channel media such as broadcasting television commercials, traditional print advertisements, radio advertisements and videos on the company’s website. But, most of the time ignoring individual consumer preferences is a factor that helps in cross-media communication planning (Lin et al. 2013). Therefore, it is of high importance that companies maximise their use of online marketing channels. Nevertheless, traditional marketing tools, such as TV advertising, are still important to reach the targeted audience. Hence, a marketing mix of online and traditional marketing can aid companies to sustain a competitive advantage.

6 These two marketing sources address different points in the customer journey and target various types of consumers. TV advertising is mainly used as a mass marketing tool to raise awareness about the brand and to find new potential customers whereas search engine advertising targets already interested customers in the consideration stage (Lemon & Verhoef, 2016). Hence, these two sources complement each other and are used to retarget customers and to enhance the experience with the brand. However, McKenzie and Royne (2009) mention that point planning is still not widely used; many have trouble combining different touch-points in their marketing communication. This paper argues that the touch-point analysis is an important tool and that the two marketing tools address different points. To see this effect this study will be done in cooperation with Pricewise, a Dutch company that invests more than four million euros in traditional advertising and search engine advertising.

Even though the trust in traditional advertising is still substantial and often used by marketeers, its effectiveness is an ongoing debate in academia (Ewing, 2013). Some authors say that it is becoming more and more irrelevant (Kotler, Kartajaya, & Setiawan, 2016). Others say that companies still have to create awareness using traditional advertising (Pfeiffer & Zinnbauer, 2010). Another opinion is that cross-media campaigns are more effective than single-medium campaigns so a combination of different advertising channels is required to maximise returns (Freeman, 1999; McMains & Morrissey, 2009). However, others have proven it not to be the case (Walkolbinger, Denk & Oberecker, 2009; Kitchen, Kim & Schultz, 2008). These studies show variable results and that academia has not yet come to a clear conclusion on the relationship between traditional and online marketing.

This study will try to get more clarity on this subject and fill the gap in academia. It is important to understand how companies in this digital era can increase advertising effectiveness. The effect of combining traditional marketing channels, such as TV advertisements and online channels is not yet clear. Nowadays, SEA is one of the leading

7 marketing channels and this study will focus on SEA over other marketing channels. Furthermore, after conducting a thorough literature review the following research question was chosen:

What effect does TV advertising have on search engine advertising?

There are two close studies to this present research (Joo, Wilbur, Cowgill & Zhu, 2013;

Joo, Wilbur & Zhu 2016). Joo and colleagues (2016) investigated whether TV advertising influences online searches. They concentrated on the distinction between branded and non-branded keywords and found that TV advertising has a small effect on non-branded keywords, however, non-branded search was seen to increase. This research aims to contribute to both academia and marketing management. Firstly, this paper will fill in the literature gap of the impact that TV advertising has on search engine marketing. The paper will go in depth on the relationship between the two channels and will show significant results to answer the research question. Secondly, this paper will be useful for marketeers and management because the spending on marketing and the impact of marketing on conversions are substantial. As mentioned above, Pricewise invests substantially on marketing, hence, gaining new insights will be financially significant. Also, getting the right combination of channels can lead to greater conversions. In particular, the outcomes of the research will be useful for the marketing officer in Pricewise in order to make the company more effective and efficient.

8

Theoretical framework

This section will bring clarity to the main variables that are examined in this research. The views from academia and the debates between authors on the subject of these variables will be stated and the explanations will follow. Firstly, traditional TV advertising will be explained together with its variables: (1) Gross Rating Point and (2) channel type. Secondly, search engine advertising will be described along with its measures starting with the number of clicks and conversions made and followed by the conversion and click-through rates. It is important to understand these terms in order to analyse their effect. Furthermore, each section will end with the hypotheses used in this research.

1. Traditional TV Advertising

As mentioned in the introduction, one of the most used traditional marketing channels is TV advertisement. This channel is widely used all over the world; however, the attitudes towards TV ads vary between countries. For example, the study done by Ewing (2013) mentions that Chinese consumers give advertising poor ratings on factors such as pleasantness, honesty and intelligence. However, in other countries, such as Australia, the attitudes towards this marketing source are not getting worse. This shows that there is a difference in culture and that Dutch attitudes might work as an advantage for a company such as Pricewise. This study will use gross rating point (GRP) to operationalize TV advertising. Also this research will look at the difference between government-owned channels and commercially-owned channels. To grasp the essence of this study, these terms have to be explained and the relationships have to be elaborated. The next sections will focus on these variables.

9

1.1 Channel type

The organization that provided the data for this research uses two different channels to broadcast their TV advertisements. They use a commercial channel, called RTL the Netherlands, and a government-owned channel, called Ster. There are two big differences between the two. First, the advertisements are shown at different time slots. Commercial channels are allowed to broadcast advertisements during a program, whereas government-owned channels are only allowed to show the ads before or after a program. The difference between prime and off-peak times that the advertisement was shown will be measured (Danaher, 1995). The effect of timing is questionable because in some cases the audience does not watch the commercial and uses the break time to talk, read or change channels to avoid advertising (Tse & Lee, 2001). Secondly, the demographics of the audience is proven to be different for the two channels (Mediamonitor, 2017). Hence, the audience has different characteristics, such as age, personal interest, and gender. For example, statistics show that the preferred channel of people above 65 years old is NPO 1 and second preferred is NPO 2, which both are government-owned channel. The most chosen channel for the 50 to 64 year old age range is also NPO 1, but second most watched channel is RTL, which is the commercial channel (Mediamonitor, 2017). Furthermore, figures show that the commercial channel is also preferred across other age groups, whereas the government-owned channel does not show such multiplicity. This can be explained by the type of programs shown in the two channels. The Dutch media asserts that RTL has a very diverse range of programs and movies that are shown throughout the day. There is not much research done on the difference between commercial and government-owned channels so this research will try to elaborate on this subjects based on these two factors presented above. This study seeks to see whether there is a difference between the two channels, what that difference is, and how it can be explained.

10

1.2 Gross rating point

GRP is one of the most commonly used measures within and across the traditional media platforms such as television, radio, and print to asses advertising success and potential target reach of customers. Schultz (2016) explains that GRP assumed one could estimate the total size of the audience watching the television and then deconstruct that total audience to determine what one percentage point of that audience would be. Many authors argue that this measure is questionable, meaning that its accuracy is uncertain (Fulgoni, 2015). These authors reason this point by saying that the GRPs grew out of the traditional over-the-air and cable TV planning, meaning it is outdated and cannot asses the new platforms correctly (De Pelsmacker, Geuens & Van Den Bergh, 2010). However, it is still widely used. There are more compelling arguments for the continued use of GRP (Aleksandrovs, Goos, Dens, & De Pelsmacker, 2015). Batra et al. (1995) used gross rating points in order to measure the effectiveness of advertising spending. Even for cross-platform campaigns the GRP metric remains a particularly important measure to understand what was delivered across platforms and weather it has been successful in cumulatively reaching its intended targets. In layman’s terms it is assumed that the higher the GRP, the better the reach. This research uses GRP to assess traditional TV advertising, which academia still considers a good measure for this particular advertising platform. Hence, this is assumed to also have a valuable outcome in the analysis.

2. Search Engine Advertising (SEA)

There are many different types of online marketing channels, such as: paid advertising, organic searches, direct type-in, and email. However, as mentioned in the introduction section this paper focuses on SEA. According to Zilincan (2015) SEA is a component of SEM. Rutz and Trusov (2011) state that search engine advertising is the leading customer acquisition tool of Internet marketers. Paid searches account for one of the biggest expenses in online marketing

11 nowadays and is part of outbound marketing (Zilincan, 2015). For example, Pricewise invests over three million in their SEA activity.

Once consumers search a specific keyword, they are likely to be directed to a specific product or service they are looking for. The reason behind this is that companies pay for specific keywords and create text ads that pop up as soon as the consumer search for that specific keyword (Rutz & Trusov, 2011). For instance, a Google or Bing advertisement appears on the search engine once the consumer types a keyword has been typed into the search bar. This channel has been proven to be most profitable because it is a direct response to a consumer’s search query in which consumers are likely to get exactly what they are searching for (Dahlen, 2001).However, not all the banners are equally important and in most of the cases they can be disturbing to consumers and impact the conversion rate to firms(Cho & Cheon, 2004; Dahlen, 2001). A number of approaches are proposed to model the difference and similarities across keywords, which can also be used to forecast the performance of each keyword (Yang & Ghose, 2010; Rutz, Bucklin & Sonnieret, 2012). However, a key limitation according to Rutz and Trusov (2011, p.790) is that: “These models assume that online shoppers who use a certain keyword are homogeneous in their preferences and responses to marketing instruments of paid search. From our perspective, this is an ad hoc assumption akin to segmenting consumers a priori.” Hence, just like the performance of the keywords is questionable, the effectiveness of SEA also needs further analysis to assess its effectiveness.

Lastly, this study will look at the difference between branded and unbranded keywords. Li and Kannan (2014) show that 73% of paid conversions are based on branded keywords and 27% are based on unbranded. This shows that more than 50% of conversions in their study are made through the branded path, which is explained by a strong brand name. Due to their findings, this paper will also include the keyword difference in the analysis. The study aims to show significant results, which will contribute to the company’s bidding activity and SEA

12 usage. Furthermore, this study will look at the volume and rates separately from each other to operationalise SEA. These terms will be explained in the next section and the hypotheses will be presented respectively.

2.1 SEM Clicks

The first measure that has to be explained is clicks. Clicks indicate how many times a customer has clicked on a link and has ended on the homepage of a website. This is a

measure of volume that is recorded when assessing the traffic of a website. Fulgoni and More (2009) argue that clicks are crucial measures for learning and predicting user responses, especially for internet-based organisations including e-commerce, online advertising, and web search. These authors also have proven that a low number of clicks doesn’t mean that they have no effect on the success of the advertisement. They examined 139 online display advertising campaigns across various industries and found that even if the amount of clicks is small it still has a positive effects on: “visitation to the advertiser's website; the likelihood of consumers conducting a search query using the advertiser's branded terms; consumers' likelihood of buying the advertised brand online; consumers' likelihood of buying at the advertiser's retail store” (Fulgoni & Morn, 2009). Other authors argue that the number of clicks is indeed an unreliable measure of the effectiveness of an advertisement (Dinner, Van Heerde & Neslin, 2014). However, this study believes that clicks are a good measurement to assess the traffic and the brand awareness of an organization. For example, branded clicks can show if there is growth in the specific organization name, which leads to growth in brand recognition. Because of these reasons, this study asserts that TV advertisements will increase the number of clicks made through a search engine. As discussed in previous section, channel type is believed to have an effect on search engine advertising. Therefore, this research believes that channel type will influence the number of clicks made online. Hence, hypothesis 1 follows:

13

Hypothesis 1a [H1a]: Channel type has an effect on the number of

branded clicks.

Hypothesis 1b [H1b]: Channel type has an effect on the number of

generic clicks.

Also, the study assumes that a higher the GRP indicates more potential customers are watching the ad. If the target audience is reached, then this should have a positive influence on search engine marketing. In this case the focus lies of the number of clicks. Therefore, hypothesis 2 follows:

Hypothesis 2a [H2a]: Advertisement GRP is positively related to

branded clicks.

Hypothesis 2b [H2b]: Advertisement GRP is positively related to

generic clicks.

2.2 SEM Conversion

Everything in a for-profit organization is driven by sales and how to maximize profits. In online marketing, sales are called conversions. Meaning the consumer converted to your product or service. According to Neal and Bathe (1997), advertisement effectiveness is judged either by the examination if the advertisement led to an increase in the number of potential consumers who would consider the product or by the number of consumers that already made the sale and became clients. The conversions are automatically recorded in the data systems of organisations. This study believes this is a very important measure to take into consideration when trying to answer the research question. The difference between branded and generic is also taken into account for this measure. This study believes that the difference in channel type will lead to some significant outcomes, which will show the relevance of channel type on sales. Therefore, hypothesis three follows:

14

Hypothesis 3a [H3a]: Channel type has an effect on the number of

branded conversions made.

Hypothesis 3b [H3b]: Channel type has an effect on the number of

generic conversions made.

This study also believes that the relationship between advertisement GRPs and conversions made exists and that this relationship is positive. It is believed that the more potential customers are reached, the more conversions will be made online. Hence, hypothesis four is as following:

Hypothesis 4a [H4a]: Advertisement GRP is positively related to

branded conversions.

Hypothesis 4b [H4b]: Advertisement GRP is positively related to

generic conversions.

2.3 Click-through rate

Search engine effectiveness is often measured by the click-through rate. In order to maximise revenue and consumer satisfaction, the methods used in estimation must predict the expected user behaviour. For each online advertisement that is displayed, must maximise the numbers that user will act or click on the advertisement. For example, if an advertisement was displayed 1000 times in the past and has only received 100 clicks, the marketing system estimates its click-through rate to be 0.1 (Richardson, Dominowska & Rango, 2007). Other authors have also used CTR to measure the effectiveness of online advertising in their research (Lewis & Reiley, 2014). Ren and colleagues (2007) also mention that the likeliness of a user to convert is measured by the predicted probability of click, which is the click-through rate or conversion and will be evaluated in the next section. Furthermore, Hoffman and Novak (2000) state that the CTR is a preferred measure over the other two normally used

15 measures, which are (1) conversion rate and (2) impressions generated. Hence, this paper agrees with Hoffman and Novak (2000) and the other authors by using CTR as an indication of SEA success. This study will show that as for the volume measurements, channel type will also influence the click-through rate. Hence, hypothesis five follows:

Hypothesis 5a [H5a]: Channel type has an effect on the branded

click-through rate.

Hypothesis 5b [H5b]: Channel type has an effect on the generic

click-through rate.

This study also believes that the click-through rate will increase when the advertisement GRP increases. Based on past literature that was presented in previous sections it is assumed that the GRPs will positively have an effect on the CTR. Therefore, hypothesis six is as follows:

Hypothesis 6a [H6a]: Advertisement GRP is positively related to branded

click-through rate.

Hypothesis 6b [H6b]: Advertisement GRP is positively related to generic

click-through rate.

2.4 Conversion rate

Lastly this research shows that conversion rate also has to be taken into account when looking at SEA. It is assumed that Pricewise seeks to display ads only to the quality match between the content of the advertisement and the user’s intent. In other words, under perfect circumstances, the conversion rate would be very close to one. Advertisers try to make the fit between the advertisement and the user as perfect as possible to reach the best conversion rate. As mentioned before, the effectiveness of online advertising is generally measured by (1) impressions generated, (2) percentage of click-throughs, and (3) inducted sales or conversion rates. The accuracy of these measures is an ongoing debate in academia and between

16 marketeers, where each company chooses one or more of these measures to assess SEA (Danaher & Mullarkey, 2003; Manchanda, Dube, Goh & Chintagunta, 2006). Even though Hoffman and Novak (2000) say that click-through rate is the preferred measure, other authors use conversion rate in their research to access success of SEA (Pan, 2015; Yang, Lin, Carlson & Ross, 2016). Hence, this paper will also include conversion rate in the analysis to examine how the results vary between conversion and click-through rate. As other hypothesis stated above this study is interested in the relationship of channel type to SEA, hence, hypothesis seven follows:

Hypothesis 7a [H7a]: Channel type has an effect on

branded conversion rate.

Hypothesis 7b [H7b]: Channel type has an effect on

generic conversion rate.

Further this study will try to show that the higher the GRPs of the advertisements indicate higher the conversion rates. Therefore, this research predicts hypothesis eight:

Hypothesis 8a [H8a]: Advertisement GRP is positively related to

branded conversion rate.

Hypothesis 8b [H8b]: Advertisement GRP is positively related to

generic conversion rate.

Figure 1 presents the conceptual framework of this research with the main effects and their hypotheses.

17 Figure 1: Operationalised model of key variables, effect, and hypotheses.

18

Methodology

1. Procedure

This research will be conducted in a quantitative manner. Firstly, the hypotheses have been stated, which is followed by an analysis to prove or reject them. Hence, the study will be done using the deductive approach. The data was retrieved from the database of Pricewise, from a period of approximately six months, starting from September 2017, during the TV campaigns. This resulted in a data sample of 172 days. The definition of variables will be determined in this section. The model presented in Figure 1, shows the hypotheses. To test these, a correlation analysis will be conducted to study the strength of the relationship between the variables. Secondly, the direct relationships between TV advertisement and SEM will be tested with a linear regression, also known as the one-way ANOVA. In the next section the variables are evaluated and the data collection of these variables is explained.

2. Sample and Data Collection

The sample consists of data from the dataset of Pricewise and includes information from a period of six months. The number of observations was 172 days in total. The data could not have been bigger because the company started with broadcasting TV advertisements in September 2017, meaning there was no prior data.

3. Variables and Measures

As seen in Figure 1 this research tests four main relationships. All four relationships have the same independent variable, traditional TV advertisement, measured by the Gross Rating Point (GRP) taking into account the difference between commercial and government-owned channel is taken into account. The dependent variables include data from search engine

19 advertisements (SEA) split into branded and generic SEA. The four operational variables are: number of clicks, number of conversions, conversion rate, and click-through rate

3.1 Independent variables

1. Channel type

Channel type is one of the variables that represent the independent variable, TV advertising. Data was gathered from the big dataset of Pricewise. The variable includes the exact number of ads shown on a commercial (RTL Nederland) or government-owned (Ster) channels. Due to a small sample size, the data was simplified and dichotomous variables were created to show if the ad was simply shown or not (1 = ad shown; 0 = ad not shown). This was done separately for the two channels. Hence, this study used two different variables for commercial and government-owned channels for future hypothesis testing.

2. Gross Rating Point (GRP)

Gross rating point is the other variable that was used to operationalise TV advertisements. This variable was given by the organisation as a sum of all GRPs per day. Since channel type was made into dichotomous variables, GRP was converted into an average GRP per day. This was done by dividing the total GRPs per day by the total number of ads shown in that particular day. The difference between government-owned and commercial channels was also made and the two separate variables were made respectively. These were used throughout all four models to test the hypotheses.

3.2 Dependent variables

1. Clicks

The first dependent variable that was used in this research was the number of clicks registered through search engine marketing. This variable is divided into branded and generic clicks. Clicks are recorded the minute a customer lands on the “home page” of Pricewise.nl. In

20 this research a distinction is made between branded and generic clicks. Branded clicks mean that a person landed on the homepage through searching the exact company name and generic click means that the person came through searching a category, industry, or product name. This data was saved in the Business Intelligence software and was provided to this research untouched as exact numbers.

2. Conversion

The second dependent variable was number of conversions made. This variable was also divided into branded and generic conversions. Conversions were recorded when a customer landed on the “Thank you” page of Pricewise.nl. The difference between branded and generic was recorded through the customer path; specifically, when a consumer arrived on the last “Thank you” page through a branded keyword of advertisement, it was coded as branded. If a consumer made the purchase through a generic keyword or ad it was coded as generic. This study used two different datasets to test the hypotheses, generic and branded conversions.

3. Click-through rate

The third dependent variable of this study was the click-through rate (CTR). As all the other dependent variables, CTR was also given as two different datasets for generic and branded CTRs. This variable was already calculated prior to the data collection for this study. However, it is important to understand how this rate is measured. CTR is the number of clicks that the ad receives divided by the number of times the ad was shown.

4. Conversion rate

The last dependent variable of this study was conversion rate (CR). This variable was also split into branded and generic CRs. This variable was given to this research as a percentage. It was calculated by taking the number of conversions (sales) and dividing that by the number of total ad clicks that can be tracked to a conversion during that same time period.

21 The data was given as two separate datasets of generic CR and branded CR and was also used in this form for hypothesis testing.

3.3 Control Variable

A number of factors may influence the results of an empirical study, therefore they have to be controlled. This study chose the objective of a TV advertisement as a control. In this case, the objectives of the different ads are to raise awareness of the brand and product. This type of marketing campaign is used to help users understand their need or desire for a product or service and most importantly provide them with the solution.

22

Results

1. Statistical procedure

The previous section discussed the design and the methods used in this research. The next section will present the results. This study is divided into four sections, with four different dependent variables. Each section represents a model, which will be presented with the corresponding correlations and regression analysis, followed by a sub conclusion and a hypothesis evaluation. Each section will mention if the variables needed to be transformed or not. Data was retrieved from the Business Intelligence software of Pricewise. For the statistical analysis the data was transferred into SPSS. For all the variables descriptive statistics, skewness and the normality tests were completed and will be explained in correspondence to each model.

2. Results

This study is structured in four different models with four different dependent variable, however the independent variables are the same for each model. The independent variables are channel type and GRP. These two variables did not have any outliers and were normally distributed. Once channel type was made into dichotomous variables it was left untouched. However, GRP was slightly skewed to the left, therefore, it was log transformed.

2.1 Model 1

Model 1 uses clicks as the dependent variable. This variables was standardised to have better coefficient in the regression analysis. It was also divided into branded and generic and was used separately in the model testing. Firstly the correlation analysis with all the variable included in this model was made, which can be seen in Table 1. Next, a one-way ANOVA was performed to see the main effects between the dependent variable and the independent

23 variables, which are tested in hypothesis one and two. The result of these tests can be further seen in Table 2.

Results Model 1

The mean, standard deviation and correlations of the variables can be seen in Table 1. The most interesting outcome of this correlation analysis is that the independent variables indeed have an effect on the dependent variable. It can be seen that CC airing is positively related to branded clicks (0.214, p<0.01). However, CC airing is not related to generic clicks. Also it can be seen that GC airing is positively related to generic and branded clicks (0.194,

p<0.05; 0.425, p<0.01). Interestingly GRP seems to be positively related to branded clicks,

but not to generic clicks (0.377, p<0.01; 0.150, p<.05; 0.035, ns; 0.007, ns).

Table 1. Descriptive statistics and correlations between the variables with clicks

Variable Mean SD 1 2 3 4 5 1. CC airing .506 .501 2. GC airing .471 .501 .350** 3. GRP CC -.597 .884 .662** .166* 4. GRP GC .134 .740 .375** .654** .117 5. Clicks Generic 2654 816.9 -.051 .194* .035 .007 6. Clicks Branded 655 348.2 .214** .425** .377** .150* .613** ** Correlation is significant at the 0.01 level (2-tailed).

* Correlation is significant at the 0.05 level (2-tailed).

CC = Commercial Channel, GC = Government-owned channel, GRP = Gross Rating Point

One-way ANOVA was chosen to test whether the model is significantly good at predicting the outcome. From SPSS it is observed that the models are significant. For the model with branded clicks as the dependent variable F(172)=19.853, p=0.000 and for generic clicks F(172)=4.116, p=0.003. The models can be seen in Table 2 with the coefficient, standard errors and betas of the independent variables. The independent variables CC airing, GC airing, GRP CC, and GRP

24 GC are all significantly correlated with clicks branded with, respectively, B=0.250**, p<0.01; B=0.449, p<0.001; B=0.384, p<0.001; B=-0.113, p<0.09. However, it can be seen that for the dependent variable, generic clicks, the outcomes are different. The coefficients’ results showed that GC airing and GRP CC positively correlate with generic clicks, with B=0.265, p<0.01 and B=0.148, p<0.09 and CC airing and GRP GC do not correlate and are not significant.

Table 2. Regression results for 2 models with clicks as the dependent variable.

Model 1.1 Model 1.2

Dependent variable _{Clicks Branded Clicks Generic} Coefficient SE Beta Coefficient SE Beta Constant -.493*** .099 -.120 .115 CC airing .498** .149 .250** -.096 .172 -.048 GC airing .896*** .137 .449*** .529*** .159 .265*** GRP CC .553*** .103 .384*** .213 .119 .148 GRP GC -.222* .129 -.113* -.252 .150 -.128 R2 _{0.322 0.090} N=171, *p<.05, **p<.01, ***p<.001

CC = Commercial channel, GC = Government-owned channel, GRP = Gross Rating Point

Sub conclusion Model 1

Concluding, with the statistical evidence hypothesis 1 is partly proven. Hypothesis 1a has enough statistical support, to see that the difference in channel type has a positive effect on brandied clicks. Meaning, that is an ad is shown on either a commercial or a government-owned channel, the number of clicks in SEM increase. Hypothesis 1b is rejected because commercial airing does not seem to have a significant correlation to generic clicks. Furthermore, hypothesis

25 two has also been partly proven, showing that Hypothesis 2a is supported, whereas hypothesis 2b rejected. Figure 2 presents a summary of the model and the betas with the p-values for each hypothesis.

26

2.2 Model 2

Model 2 uses conversion as the dependent variable. This variable was also standardised to have better coefficient in the regression analysis. It was also divided into branded and generic and was used separately in the model testing. Firstly, the correlation analysis with all the variable included in this model was made, which can be seen in Table 3. Next, a one-way ANOVA was performed to see the main effects between the dependent variable and the independent variables, which are tested in hypothesis three and four. The result of these tests can be further seen in Table 4.

Results Model 2

The mean, standard deviation and correlations of the variables can be seen in Table 3. The most interesting outcome of this correlation analysis is that some of the independent variables indeed have an effect on the dependent variable. It can be seen that CG airing is positively related to branded conversions (0.260, p<0.01). However, CG airing is not related to generic clicks. CC airing is also not related to either generic or branded conversions. Also it can be seen that GRP CC is positively related to branded conversions (0.221, p<0.01). It is interesting to see that GRP GC is both negatively related to generic and branded conversions

(-0.174, p<0.05; -0.186,P<0.05). Next, it is seen that CC airing is not related to both generic

27

Table 3. Descriptive statistics and correlations between the variables with conversion.

Variable Mean SD 1 2 3 4 5 1. CC airing .506 .501 2. GC airing .471 .501 .350** 3. GRP CC -.597 .884 -.431** -.252** 4. GRP GC .134 .740 .245** .086 -190* 5. Conversion Generic 213.304 90.539 -.041 .122 .160* -.174* 6. Conversion Branded 70.460 38.709 .059 .260** .221** -.186* .833**

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

CC = Commercial Channel, GC = Government-owned channel, GRP = Gross Rating Point

One-way ANOVA was chosen to test whether the model is significantly good at predicting the outcome. From SPSS it is observed that the models are significant. For the model with branded conversions as the dependent variable F(172) = 10.402, p = 0.000 and for generic conversions F(172) = 3.635, p = 0.007. The models can be seen in Table 4 with the coefficient, standard errors, betas, and the significance of the independent variables. The independent variables, GC airing, GRP CC, and GRP GC are all significantly correlated with conversions branded with, respectively, B=0.320, p<0.001; B=.322, p<0.001; B=-0.200, p<0.01. However, it can be seen that for the dependent variable, generic conversions, the outcomes are very similar, but the correlation showed less correlation. The coefficients’ results showed that GC airing and GRP CC positively correlate with generic conversions, with B=0.185, p<0.05 and B=0.180, p<0.05. Whereas, GRP GC is negatively correlated with generic conversions, with B=-0.167, p<0.05. CC airing did not show any correlation with either of the dependent variables.

28

Table 4. Regression results for 2 models with conversion as the dependent variable.

Model 2.1 Model 2.2

Dependent variable _{Conversion Branded Conversion Generic} Coefficient SE Beta Coefficient SE Beta Constant -.273 .108 -.008* .116 CC airing .270 .162 .135 .026 .173 .013 GC airing .639*** .149 .320*** .370* .160 .185* GRP CC .464*** .112 .322*** .260* .120 .180* GRP GC -.391** .141 -.200** -.328* .151 -.167* R2 _{0.199 0.080} N=171, *p<.05, **p<.01, ***p<.001

CC = Commercial channel, GC = Government-owned channel, GRP = Gross Rating Point

Sub conclusion Model 2

Model two tested for hypothesis three and four and all these hypotheses were partially proven. Hypothesis 3a predicted that channel type has an effect on branded conversions made. This hypothesis is partially supported. Commercial channel airing shows no correlation with conversions made. However, government channel airing shows significant results to prove the hypothesis, showing a positive correlation. Hypothesis 3b stated that channel type has an effect on generic conversions made, this hypothesis was also partially supported. Again, commercial channel airing showed no correlation to branded conversions. Whereas, government-owned channel airing had enough statistical evidence to show a positive correlation. Furthermore, hypothesis 4a was partially supported, this hypothesis predicted that GRP is positively related to branded conversions. Both GRPs has a significant correlation, however, government-owned

29 channel GRPs showed a negative relationship to branded conversions. Which was not the case for commercial channel, showing a positive relationship and partially proving the hypothesis. The results predicting hypothesis 4b show similar results as the hypothesis 4a. It predicted that GRP is positively related to generic conversions made. It can be further seen that commercial channel GRP supports the hypothesis, showing positive correlation to generic conversions. But, government-owned channel again showed a negative correlation to generic conversion. To summarise Model 2 Figure 3 was made, showing corresponding betas and p-values.

30

2.3 Model 3

The dependent variable for model 3 is the click-through rate. This variable was also divided into branded and generic and was checked for outliers and distribution. Branded through rate was normally distributed and did not need any transformations. Generic click-through rate showed some outliers, however, these outliers did not lead to a smaller significance in the regression analysis, therefore they were left untouched and not removed from the dataset. Firstly the correlation analysis with all the variable included in this last model was made, which can be seen in Table 5. Next, a linear regression was performed to see the main effects between the dependent variable and the independent variables, which are tested in hypothesis seven and eight. The result of these tests can be further seen in Table 6.

Results Model 3

The mean, standard deviation and correlations of the variables of Model 4 can be seen in Table 5. Interestingly the only variable that is related to both generic and branded click-thought rates in CC airing, other independent variable seen to have to relation to the dependent variables. CC airing is positively related to CTR branded, but negatively related to CTR generic

(0.173, p<0.05; -0.246, p<0.01). CTR generic is not related to GC airing, GRP CC, and GRP

CG, with respectively (-0.117, ns; 0.131, ns; 0.010, ns). Also CTR branded does not seem to be related to GC airing, GRP CC, and GRP GC, with the corresponding values (-0.117, ns;

31

Table 5. Descriptive statistics and correlations between the variables with click-through rate.

Variable Mean SD 1 2 3 4 5 1. CC airing .506 .501 2. GC airing .471 .501 .350** 3. GRP CC -.597 .884 -.431** -.252** 4. GRP GC .134 .740 .245** .131 -190* 5. CTR Branded .601 .066 .173* -.117 .017 .241 6. CTR Generic .184 .029 -.246** -.117 .131 .010 .065

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

CC = Commercial Channel, GC = Government-owned channel, GRP = Gross Rating Point, CTR = Click-Through Rate

Model 3 also used one-way ANOVA to test hypotheses five and six. It is observed that the models are significant. For the model with branded click-through rate as the dependent variable F(171) = 3.459, p = 0.010 and for generic conversion rate F(171) = 3.037, p = 0.019. The models can be seen in Table 6 with the coefficient, standard errors, betas, and the significance of the independent variables. There are three significant correlations in this model. First, CC airing is positively correlated to CTR branded, with B=0.265, p<0.01 and negatively related to CTR generic, with B=-0.240, p<0.01. Whereas, GC airing is only related to CTR branded and the relationship is negative, with B=-0.195, p<0.05. All the other relationships showed no correlation with each other, this can be further seen in the table below.

32

Table 6. Regression results for 2 models with click-through rate as the dependent variable.

Model 3.1 Model 3.2

Dependent variable _{CTR Branded CTR Generic} Coefficient SE Beta Coefficient SE Beta Constant .598*** .008 .192*** .003 CC airing .035** .011 .265** -.014** .005 -.240** GC airing -.026* .011 -.195* -.002 .005 -.033 GRP CC .009 .008 .096 .001 .003 .034 GRP GC .009 .010 .069 .005 .004 .080 R2 _{.077 0.068} N=171, *p<.05, **p<.01, ***p<.001

CC = Commercial channel, GC = Government-owned channel, GRP = Gross Rating Point CTR = Click-Through Rate

Sub conclusion Model 3

Model 3 tested hypothesis five and six, starting with hypothesis 5a, which predicted that channel type has an effect on the branded CTR. This relationship was supported by showing that both channel types had significant correlation results. Commercial channel showed that it is positively correlated with branded CTR, but government-owned channel is negatively correlated with branded CTR. Furthermore, hypothesis 5b predicted that channel type has an effect on generic CTR. This hypothesis was partially supported. Commercial channel showed that it is negatively correlated with generic CTR, but government-owned channel showed no significance, hence does not support the hypothesis. Next hypothesis 6a was tested, which predicted that GRP is positively correlated with branded CTR. It was not supported, because of lack in statistical evidence to draw a significant conclusion. And lastly hypothesis 6b predicted tested whether GRP is positively related to generic CTR. This was also

33 rejected due to insufficient significance. The summery of Model 4 can be seen in Figure 4 with the corresponding betas and p-values.

34

2.4 Model 4

The dependent variable of Model 4 is conversion rate. The variable branded conversion rate was given by the company as a percentage, and needed to be transformed into a numeric variable, hence, was divided by 100. Same was done to the generic variable. After checking for normality it was acknowledged that the data had a few outliers that had to be removed in order for the data to be more normally distributed. This also made the regression analysis more reliable and increased the significance of the model. The new sample includes 162 data points. Firstly the correlation analysis with all the variable included in this model was made, which can be seen in Table 7. Next, a linear regression was performed to see the main effects between the dependent variable and the independent variables, which are tested in hypothesis five and six. The result of these tests can be further seen in Table 8.

Results Model 4

The mean, standard deviation and correlations of the variables of Model 3 can be seen in Table 7. In general the interesting outcome of this model is that Branded conversion rate is negatively related to three out of the four independent variables, whereas, generic is only related to one. CR Branded has a strong negative relation to CC airing, CG airing and GRP GC, with respectively (-0.318, p<0.01; -0.294, p<0.01; -0.241, p<0.01). CR generic is positively related to GRP CC (0.184, p<0.05). Also it can be seen that CR generic is not related to CC airing, GC airing, and GRP GC (-0.123, ns; -0.101, ns; -0.129,ns).

35

Table 7. Descriptive statistics and correlations between the variables with conversion rate.

Variable Mean SD 1 2 3 4 5 1. CC airing .506 .501 2. GC airing .471 .501 .350** 3. GRP CC -.597 .884 -.431** -.252** 4. GRP GC .134 .740 .245** .131 -190* 5. CR Branded .111 .026 -.318** -.294** .134 -.241** 6. CR Generic .073 .009 -.123 -.101 .184* -.129 .451**

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

CC = Commercial Channel, GC = Government-owned channel, GRP = Gross Rating Point, CR = Conversion Rate

Hypothesis seven and eight were tested in model 4 also using the one-way ANOVA. From SPSS it is observed that the models are significant. For the model with branded conversion rate as the dependent variable F(162) = 9.568, p = 0.000 and for generic conversion rate F(162) = 2.650, p = 0.035. The models can be seen in Table 8 with the coefficient, standard errors, betas, and the significance of the independent variables. The independent variables CC airing, GC airing, are all significantly correlated with branded conversion rate with, respectively, B=0.320, p<0.001; B=-0.200, p<0.00. However, for the other independent variable, GRP, this model showed no correlation with branded conversion rate. For generic conversion rate the coefficients’ results showed that CC airing and GC airing are positively correlated, with B=0.251, p<0.01; B=0.042, p<0.001. However, GRP CC and GRP GC is not correlated with generic conversion rate.

36

Table 8. Regression results for 2 models with conversion rate as the dependent variable.

Model 4.1 Model 4.2

Dependent variable _{CR Branded CR Generic} Coefficient SE Beta Coefficient SE Beta Constant .123*** .003 .075 .001 CC airing -.017*** .004 -.320*** -.005** .002 -.251** GC airing -.011** .004 -.200** .001*** .002 .042*** GRP CC -.005 .003 -.138 .000 .001 .017 GRP GC -.007 .004 -.135 .000 .001 -.015 R2 _{0.195 0.063} N=162, *p<.05, **p<.01, ***p<.001

CC = Commercial channel, GC = Government-owned channel, GRP = Gross Rating Point CR = Conversion rate

Sub conclusion Model 4

Model 4 tested hypothesis five and six. Where hypothesis 7a predicted that channel type has an effect on branded CR. This hypothesis was supported, however showed strange outcomes. Commercial and government-owned channel airings, both showed a negative correlation with branded CR. Next hypothesis 7b was also tested, which predicted that channel type has an effect on generic conversions. This was also supported, but showed two different correlations for the channel types. Commercial channel is negatively correlated with generic CR and government-owned channel has a positive correlation with generic CR. Furthermore, hypothesis 8a tested whether GRP is positively correlated with branded CR. This hypothesis had not enough statistical evidence to support it. Same for hypothesis 8b, which predicted that GRP is positively related to generic CR. This hypothesis also did not show significant statistical

37 results, hence it is rejected. The summary of Model 4 with its betas and p-values can be seen in Figure 5.

38

Discussion

This section will elaborate the results found in the analysis. The hypotheses and the models of this study will be concluded. Furthermore, the practical implications of this research will be stated and the limitations will be addressed. The final section will discuss the implications for future research and academia. This paper will end with a short conclusion.

1. Discussion of findings

1.1 Hypotheses findings

This study aimed to contribute to the literature concerning the traditional advertising and search engine marketing and, in particular, the effects of TV channel type and GRP on SEA. Specifically, it tested whether TV advertising increases the volumes and rates of SEA. This study tested eight hypotheses, which will be elaborated further in this section.

The first hypothesis predicted that channel type would have an effect on the number of clicks made. As all the other hypotheses this was divided into branded and generic. It was proven that channel type has a positive effect on branded clicks made. This means that when the advertisements were shown the number of branded clicks increased. There was an assumed support before the testing, because if the advertisements are made correctly they should increase the number of times consumers search by the specific brand name and thus increase the number of clicks made. Also, the first hypothesis predicted that channel type would have an effect on generic clicks. This was a difficult hypothesis to support before the analysis because it is logical to assume that by having advertisements about a specific product increase the number of generic clicks; however this was only an assumption. The assumption was partially supported by the fact that only government-owned channels seemed to have a positive effect on generic clicks. This is difficult to explain as there might be multiple consumer

39 characteristics that vary between the channels. It might be that people who watch the programs on government-owned channels are more focused and are more engaged with the advertisements. This difference also exists for the branded clicks, even though both channel types seemed to have a positive effect on branded clicks. It is seen that government-owned channel has a much higher coefficient and seems to increase the number of clicks by almost 50% if the ad was shown.

The second hypothesis predicted that GRPs are positively related to clicks made. To make the testing more reliable this was also split into branded and generic. It was assumed that the bigger the GRP meant the more targeted audience an advertisement would reach, hence it was logical to assume that a bigger GRP would positively influence branded and generic clicks. This hypothesis was partially proven. The analysis showed that GRPs only seemed to have an effect on the branded clicks and not on the generic. The effect size also varied between commercial and government-owned channels. Strangely enough, the commercial channel showed a strong positive effect on branded clicks, whereas the government-owned channel showed a low negative effect. This result is difficult to comprehend, because not a lot of literature can be found on this subject. The only explanation that this study can provide is that the audience of the commercial channels is more diverse, hence, the higher the GRP in a commercial channel the more varied the potential customers are. The audience of the government-owned channel is more uniform, so even with a high GRP the target audience would only represent a small portion of the potential consumers. Furthermore, it is proven that the GRPs have no correlation with generic clicks. This might be explained by the meaning and purpose of GRPs. They are supposed to predict the audience that would be interested in the brand and not the sector or category. However, no literature is available to support this outcome, therefore, this should be tested further.

40 Hypothesis three tested whether channel type has an effect on conversions made. As with the previous hypotheses, this was also divided into branded and generic. This hypothesis was partially proven, showing a strong positive effect for branded conversions and a weak positive effect for generic conversions. The outcome for the commercial channel again showed no effect for branded or generic conversions. The testing showed a strong effect for the owned channel. This means that when the commercial was aired on a government-owned channel, the commercial increased the number of conversions made. The effect is stronger for branded conversions. The support for these arguments lies closely to the support mentioned in the previous paragraphs. Also, the company that provided the data mentioned that the average age of their consumers lies in the higher end of the age scale, with an average consumer being 50 years old. If that is the case then it supports the hypothesis, because the average age of viewers on a government-owned channel is higher than on the commercial channel (Mediamonitor, 2017). Channel type once more seemed to have little effect on the generic side of the conversions made. There was a low positive effect of the government-owned channel on generic conversions, but the effect is relatively small in comparison to the branded conversions.

Hypothesis four tested whether GRPs had a positive effect on conversions made. It showed various results and was partially statistically proven. The effect of GRPs on branded conversions was strong and is similar to the result of hypothesis two. It is seen that the commercial channel’s GRP is positively related to conversion, meaning that the higher the GRP, the more branded conversions take place. The effect on branded conversions was stronger than for generic, which was predicted and is a logical outcome because a good advertisement is more likely to promote the brand and not only the category itself. Contrarily, the data indicated that the government-owned channel’s GRP negatively influenced branded and generic conversions. The effect is not as strong as for the other channel but still substantial.

41 This, again, can only be explained by the diversity of potential customers that are present in the audience because it is unlikely that a high GRP would result in a decrease in sales. If that is the case, then the estimation of GRP is incorrect and the targeted audience is not reached; However, it is almost certainly not the case for Pricewise.

Hypothesis five predicted that channel type has an effect on the click-through rate. This hypothesis is partially supported; three of the four relationships that are included in this hypothesis have showed a significant effect. For the branded click-through rate, the outcome was different for each channel type. The commercial channel showed a positive effect and the government-owned channel showed a negative effect on the click-through rate. For the generic click-through rate the effect was only significant for the commercial channel. These results are very mixed and show various outcomes, hence are difficult to comprehend. There is one assumption that can explain these outcomes. Once the volume of search engine marketing increases due to the interest in the brand after a TV advertisement, the click-through rate might decrease because the volumes increased and the clicks made are not growing as fast as the volume. This is proven by the data used in the model, however it cannot be backed up with literature, hence further research is needed. This might explain the negative effects of TV advertisements on click-through rate. However it is difficult to explain the positive effect that commercial channels shows with respect to branded click-through rate.

Hypothesis six tested whether GRP had a positive effect on click-through rate. This hypothesis was rejected since none of the relationship showed correlation. This means that GRP does not seem to have any effect on the click-thorough rate. This outcome was not predicted before because logically the variables would have a positive effect on each other, especially, because clicks showed a relationship with GRPs.

Furthermore, hypothesis seven predicted that channel type would have an effect on conversion rate. This hypothesis was supported, all four relationships showed a significant

42 correlation between the independent and the dependent variables. However, the outcomes were different for each channel types. Branded conversion rate had a negative correlation with both channel types. This means that the conversion rate decreased once the TV commercial had been aired. This, again, can be explained by the relatively small increase in conversions in relation to a high increase in volume, leading to a smaller conversion rate. Next, the generic conversion rate showed a strong negative correlation with the commercial channel, but a weak positive correlation with the government-owned channel. The difference between the two is small and in general the conclusion can be made that the effect is negative or non-existent, showing the same result as branded conversion rate.

Last, hypothesis eight predicted that GRPs are positively related to conversion rate. This hypothesis had insufficient statistical evidence for support and had to be rejected. This means that the GRPs have no effect on the conversion rate. This is a similar outcome to hypothesis six, and has to be evaluated further for more understanding o this outcome. The hypotheses have been summarised and put in a Table 9 for a clear overview with the corresponding key information.

43

Table 9: Hypotheses summary

Hypothesis Number

Description Supported Effect size

CC/GC H1a Channel type has an effect on

Branded Clicks

Supported + Strong / + Strong H1b Channel type has an effect on

Generic Clicks

Partially Supported No /+ Strong H2a GRP has a positive correlation

with Branded Clicks

Partially Supported + Strong / - Low H2b GRP has a positive correlation

with Generic Clicks

Not Supported No / No

H3a Channel type has an effect on Branded Conversions

Partially Supported No / + Strong H3b Channel type has an effect on

Generic Conversions

Partially Supported No / - Low H4a GRP has a positive correlation

with Branded Conversions

Partially Supported + Strong / - Medium H4b GRP has a positive correlation

with Generic Conversions

Partially Supported + Low / - Low

H5a Channel type has an effect on Branded CTR

Supported + Medium / - Low H5b Channel type has an effect on

Generic CTR

Partially Supported - Medium / No H6a GRP has a positive correlation

with Branded CTR

Not Supported No / No

H6b GRP has a positive correlation with Generic CTR

Not Supported No / No

H7a Channel type has an effect on Branded CR

Supported - Strong / - Strong H7b Channel type has an effect on

Generic CR

Supported - Strong / + Strong H8a GRP has a positive correlation

with Branded CR

Not Supported No / No

H8b GRP has a positive correlation with Generic CR

44

1.2 Model findings

This study finds it important to discuss the models separately from the hypotheses. Firstly, the R squared will be evaluated. This measure shows how close the data points are to the fitted regression line. The models used in this study show a relatively low R2_{. However,}

this is not a problem in this study, because marketing studies generally do not show high R2 _in

there analyses. Table 10 was made to summarise all the models, it is presented with the corresponding R2_{, F values, dependent variables and hypotheses. One of the main findings}

from the models is that the difference in R2 _{is always substantial between branded and generic}

dependent variables. For example, model 1.1 uses branded clicks as the dependent variable and the R2 _{is 0.322. Whereas, model 1.2 uses generic clicks as the dependent variable and the R}2

is significantly lower at 0.090. This shows that for the generic variables the model is less accurate and the data points lay further from the regression line. In other words, it means that the branded effect is stronger than generic and this research shows that the TV advertisements have a greater impact on branded SEM. In practice, this is a good outcome because brand building plays a big role in an organisation’s success and the results of this research lead to that outcome.

Secondly, this research indicates that the difference between volume and rate needs to be elaborated because the results showed interesting outcomes. As mentioned in the hypotheses discussion, the results for conversion and click-through rates are less significant than for the volume measures, which were clicks and conversion records. This study shows that the reason behind this is that the exact volume is a more reliable measure. The rate on its own can stay the same or decrease because the conversions do not rise at the speed as the volume. When an interview was taken from one of the SEA specialists at Pricewise, he stated that they see that the some conversions follow a few days after the click has been made. He explained this by saying that potential customers need time to make a decision to place a sale. Therefore, since

45 the data was taken daily, these conversions are not recorded on the day the volume has risen so the conversion rate decreases. Other authors who agree that the click-through rate and conversion rate only account for direct effects of advertising have also raised this argument. However, Google ads may have an indirect effect on other consumers in the future (Hollis, 2005; Huang & Lin, 2006; Manchada et al., 2006). Furthermore, the rates on their own say very little about the success of the advertisements. Therefore, both of the measures are very important to consider when assessing the success of a campaign. For this reason, both were used in this research.

Table 10: Models summary.

Number Dependent variable R2 _{F Hypotheses}

1.1 Branded Clicks 0.322 19.853

P=.000

H1a & H2a

1.2 Generic Clicks 0.090 4.116

P=.003

H1b & H2b

2.1 Branded Conversion 0.199 10.402

P=.000

H3a & H4a

2.2 Generic Conversion 0.080 3.635

P=.007

H3b & H4b

3.1 Branded CTR 0.077 3.459

P=.010

H5a & H6a

3.2 Generic CTR 0.068 3.037

P=.019

H5b & H6b

4.1 Branded CR 0.195 9.568

P=.000

H7a & H8a

4.2 Generic CR 0.063 2.650

P=.035

46

2. Practical Implications

This study has shown interesting significant results with practical implications that can be exercised in an organisation. Since the company operates online, it is of high importance that their search engine marketing works to its full potential. Firstly, it is seen that the TV advertisement increases the volume and traffic on search engine advertising. This means that the TV works great as a channel to raise awareness for the company and its product. It is also seen that the conversions and clicks increase if an ad has been shown, whereas the rates sometimes show a negative effect. This is explained by the disproportionate increase in volume with respect to the actual amount of conversions or clicks.

Secondly, this study has proven that the TV advertisements have a greater impact on the branded SEM versus generic SEM. This is advantageous to the company, because the target audience is responding to the ads in the way the company intends. Pricewise can assume that their brand building is working and that they are not promoting the industry but their own name and increasing their brand awareness. When an interview was taken with the marketing officer of Pricewise, he mentioned that their brand awareness has been rising since the commercials have aired on TV, this is now also supported by this research. However, some might argue that generic search engine marketing is also important to address the urgency of the service and to create recognition of the category, in order for the potential consumer to start thinking about the purchase.

Thirdly, it has been proven that the channel types have an effect on SEM. In some cases the government-owned channel showed more significant results than the commercial channel and in other cases it was the other way around. This study indicates that this effect is explained by the difference in audience characteristics. The choice of which channel to broadcast lies in the company’s goals. If they want to address a more diverse audience and broaden their consumer characteristics, a commercial channel is the better choice. However, if Pricewise