online banner advertising?
Investigation of the relationship between banner
design elements and click-through rate.
What factors determine the effectiveness of
online banner advertising?
Investigation of the relationship between banner
design elements and click-through rate.
Management summary
The European market for online banner advertising has grown to a 4 billion Euros market in 2009. A lot of academic research is performed on the topic of online banner advertising effectiveness and multiple studies have investigated the effects of certain design elements. However, most studies only focus on one or two factors and also use data from a limited number of industries. The contribution of this research is that a total of ten design elements that are proven in either offline advertising or online advertising are combined in one model in order to determine which factors determine online banner advertising effectiveness. Furthermore data is used from four different industries, making the results more generalizable.
An extensive literature study has been performed to indentify the factors that could be of influence on the online advertising effectiveness. The factors are be based either on research on advertising in the more traditional media or on research in the field of online advertising. From this research the following design elements are identified: size, image presence, words used, colors used, shape of advertisement, animation type, length of animation, call to action presence, obtrusiveness and interactivity. Besides the factors that could influence online advertising effectiveness, this online advertising effectiveness itself is defined as the click-through rate. The factors are combined in a conceptual model that tests the effects on advertising effectiveness by making use of 14 hypotheses.
a negative effect on advertising effectiveness. Furthermore there are significant differences among the different industries, both the kitchen appliances industry and the non-profit industry obtain a higher click-through rate than the car insurance industry.
It is recommended that online marketing agencies build in images and interactivity components in their banner advertisements. Adding an image to the banner advertisement will increase the comprehensiveness of the advertisement, which in turn leads to a higher click-through rate. The interactivity component creates some kind of curiosity with consumers at the pre-click level, inducing consumers to interact with the banner before they are evoked to click through to the advertisers website. Besides adding these components is it suggested that the vertical banner shape is used as much as possible and that banner size is reduced to a minimum because bigger banners decrease the click-through rate.
Preface
Almost five years ago, when I was still a Bachelor of Economics and Management student who didn’t knew which specialization to choose, I started working on the marketing department of a local wholesale company. Within a few days of work on the marketing department I realized that marketing was the specialization direction I should choose and that this was the field I wanted to work in after graduation. After two years of work I became responsible for the e-commerce, shifting my interests from the more traditional marketing to online marketing. The idea for this thesis was born after I launched a few online banner advertisements campaigns which led to astonishing low click-through rates and leaving my with the questions “What went wrong?” and “What should I do to get better results?” With this thesis not only my own questions were answered but hopefully also a contribution is made for online marketing companies such as Storm Marketing Consultants.
Many thanks for supervising this thesis to Sonja Gensler and Thorsten Wiesel. Sonja always provided structured and directed feedback within a day, keeping the process up to pace and improving the quality of this thesis. Furthermore I want to thank the whole team at Storm Marketing Consultants. Not only for their ideas and feedback on my thesis but also for the help on the model specification and data collection. Special thanks goes to supervisor Haico Pols who made it possible for me to do my thesis at Storm Marketing Consultants, which provided me with all the data needed for this research.
Enjoy reading!
Table of Contents
1. Introduction...7
2. Theoretical Framework...11
2.1 Measuring online banner advertising effectiveness...11
2.2 Factors that could determine online banner advertising effectiveness ...14
2.3 Conceptual model ...25 3. Research Design...27 3.1 Data collection ...27 3.2 Variables of interest ...28 3.3 Mathematical model...31 4. Results...33 4.1 Statistical validity...33 4.2 Model fit...36
4.3 Investigation of the parameters...37
4.4 Hypothesis evaluation...39
5. Conclusions, recommendations, limitations and future research...42
5.1 Conclusions...42
5.2 Recommendations...45
5.3 Limitations ...47
5.4 Directions for future research ...48
References...50
1. Introduction
Even in the times of economic recession the spending on online advertising is ever growing. In 2009 the European online advertising industry grew by 4.5% to a total amount of 14.7 billion Euros (IAB Europe, 2010). Not only the total spending on online advertising is growing but also the portion of online advertising spending as part of main media expenditures is rising. In Western-European countries such as The UK and The Netherlands the spending on online advertising is already making up for approximately 25% of total advertising spending (IAB Europe, 2010). These numbers show that the Internet and the online advertising opportunities can count on an ever-increasing interest from marketing managers. But are all these online advertising billions well spent? And how should marketing managers design their online marketing campaign to get a positive ROI?
This research focuses on the banner advertising section of online marketing. Banner advertisements where used for the first time in 1994 (Robinson, Wysocka and Hand, 2007), when the World Wide Web was still in it’s infancy. Since then it has grown to a 4 billion Euros market in Europe in 2009, accounting for 30% of the total online advertising expenditures (IAB, 2010). However, how can marketers make sure that they spend their budget for banner advertising in such a way that it is as effective as possible? There is academic research that states that banner advertisements have a negative connotation associated with it because many consumers perceive them as being annoying (Yang and Ghose, 2010) and thus that banner advertising doesn’t increase sales. Furthermore Drèze and Hussherr (2003) argue that more than half of the online users may not even pay attention to online advertisements. On the other hand Manchanda et al. (2006) claim that banner advertising increases purchase behavior and thus is effective. Also found is that web advertisements build a more favorable attitude towards the brand, even if consumers did not process the banner consciously (Yoo, 2008). But which banners do the consumers process consciously and which banners do increase purchase intentions? Are there commonalities in the banner design that can explain the differences between an effective banner and a non-effective banner? Or is the non-effectiveness of a banner more dependent on the website were it is placed? And when is a banner effective? When it generates a click-through or when it generates brand awareness?
i.e. whether a banner is horizontal versus vertical and animated versus not animated, is investigated by Drèze and Hussherr (2003). Furthermore Manchanda et al. (2006) studied the effects of banner advertising repetition on Internet purchase behavior. However, the number of articles that have investigated the factors that determine the effectiveness of banner advertising is fairly limited. Also the research is often done within a specific industry and thus results cannot always be generalized, e.g. Robinson, Wysocka and Hand (2007) who only studied the online casino industry. Also there is discussion about how to measure online advertising effectiveness, whether it should be the click-through rate (i.e. Robinson, Wysocka and Hand, 2007), purchase intentions (Manchanda et al., 2006) or something more from the traditional marketing such as brand awareness (Fulgoni and Mörn, 2009).
The goal of this research is to bring together the factors that determine the effectiveness of online banner advertising. Thus combining the factors that have already been identified and proven to be effective for the more traditional media together with factors that recently have been studied in the online world. The contribution of this research to existing literature is that it proposes and tests a complete model with a combination of factors whereas other literature has often focused on only one or two factors. Furthermore previous studies only used data from one or two industries, making it hard to generalize the results. In this research data from four different industries is used. After this study it should be clear to online marketing agencies how to design online banner advertisements in order to make the banners as effective as possible. Furthermore it is investigated how online banner advertising effectiveness should be measured. These goals and contributions can be represented by the following research question:
What are the factors that determine the effectiveness of online banner advertising and how can this effectiveness be measured?
2. Theoretical Framework
In this part of this paper the theoretical framework is constructed. First the dependent variable, advertising effectiveness is discussed. After that it is determined which factors could have an influence on the effectiveness of online banner advertising. The factors are based on literature either from the more traditional print magazine advertising or the literature that has focused on online marketing. All the factors are concluded with a hypothesis of what the expected effect on banner advertising effectiveness would be. The chapter is concluded with a graphical representation of the theoretical framework.
2.1 Measuring online banner advertising effectiveness
The research question of this paper is to determine which factors influence the effectiveness of online banner advertising. But before these factors are identified it should first be determined what is meant with advertising effectiveness. In the last decade there has been much discussion about which measurement should be used to measure the effectiveness of banner advertising (e.g. Manchanda et al., 2006; Drèze and Hushher, 2003; Chatterjee, Hoffman and Novak, 2003).
very hard to assign for example a certain percentage of sales increases to a specific advertisement. Therefore brand recall, brand recognition and brand awareness are still the predominant measurements for traditional advertising effectiveness.
One of the biggest advantages of online marketing is that much more is measurable compared to print advertising. However, just as with the more traditional marketing, there has been a lot of discussion about how online banner advertising effectiveness should be measured (Manchanda et al., 2006). At first online advertising agencies were using print media exposure metrics to calculate online advertising rates the same as they did offline, in for example CPM (Hoffman and Novak, 2000). But soon online marketers were asking for more heuristic metrics of performance that the Internet could provide such as the click-through rates and the conversion rate. Chatterjee, Hoffman and Novak (2003) define a click-through as a click on the banner advertisement by a consumer after which the consumer is redirected to the advertisers website. Assumed is that a click-through to the advertisers assigned website is a sign of interest of the Internet user in the advertised product or service. Thus a click-through can be seen as a customer, whose interest has been triggered by the banner advertisement, making click-through rate a measurement of advertising effectiveness. Next to obtaining a click-through most banners also have a specific pre-set goal on the advertisers website, e.g. an e-mail registration or a purchase. The conversion rate measures whether this goal is completed or not after the consumer has clicked on the banner. Therefore the conversion rate can be seen as an actual sales or newly acquired customers due to the banner advertisement, thus measuring online advertising effectiveness in figures that could be used to justify the costs of marketing activities.
probable brands. Therefore it is hard to express advertising effectiveness in terms of sales or other economic figures if these kinds of measurements are used. On the other hand the measurements used for online marketing such as the click-through rate and the conversion rate measure almost at the last stage of purchase. In this stage the consumer has already evaluated the different brands and the decision is already made to click on a specific banner advertisement.
Both measurement methods have advantages and disadvantages. The biggest disadvantage of the traditional effectiveness measurements is how they are measured. Measurements such as brand recall and brand awareness require a laboratory experiment in order to be measured. These laboratory experiments require a lot of time and money because participants are needed and special test situations have to be simulated. Disadvantage of the click-through is that it is, according to Manchanda et al., (2006), a measurement of a visit to the website and not a measure of advertising effectiveness. Furthermore click-through rates have dropped from 7% in 1996 to 0.7% in 2002 (Drèze and Hussherr, 2003), putting click-through rate as a measure of banner advertising effectiveness under pressure.
the click-through rate the conversion rate is an even better measurement of advertising effectiveness because real consumer actions are measured. Furthermore, by comparing the click-through rate and the conversion rate, banners can be identified that for example generate very high click-through rates but fail to deliver consumers that convert the pre-set goals and thus only generate website traffic.
2.2 Factors that could determine online banner advertising effectiveness In determining the factors that could be of influence on the effectiveness of online banner advertising earlier research made use of two separate categories (e.g. Baltas 2003). The factors in the first category are the so-called creative factors, factors that can be totally altered by the advertising agency in designing a banner. The other category consists of the factors that are related with the placement and use of the banner, thus information as the website they appear on, the place on that website and the number of competitive banner advertisements on a specific website. However, the media landscape for online banner advertising has changed lately. Because of the ever-growing character of the Internet it is impossible for online marketing agencies to investigate, select and contact the websites they want to advertise on by hand. To accommodate for this problem media-buying networks have emerged. These networks buy advertising space on a large number of websites and sell this space on a cost-per-click (CPC) basis. A disadvantage of these media-buying networks is that advertising agencies lose control of on which websites and where on the websites they want to advertise. So the second category of factors that is proposed by Baltas (2003) is out of control of the marketing agencies nowadays. Therefore this research will focus on the factors that are in control of the marketing agencies, the factors based on banner design.
effective for online banner advertising. Furthermore specific factors that are only related to online banners are also taken into account. The contribution of this study to existing research is that it combines multiple factors that could be of influence on advertising effectiveness whereas other studies are often limited to only one or two factors. Furthermore in this study the interactivity variable is accounted for, interactivity is something quite new to the banner design. Interactivity is something that takes place at a pre-click level, it is designed in order to create curiosity which after the interaction should lead to a click-through to the advertisers website.
In table 1 below the factors that have been identified in previous research are summarized, both factors that are proven to be effective for the more traditional media as well as the online specific factors. All factors are further elaborated in this section.
Table 1: Factor identification
Factor Study
Size of advertisement e.g. Finn, 1988; Kelly and Hoel, 1991; Homer 1995
Image presence e.g. Childers and Houston, 1984; Unnava and Burnkrant, 1991; Finn, 1988 Number of words used e.g. Donthu, Cherian and Bhargava , 1993; Schweiger and Hruschka,1980
Number of colors used Kvitastein and Gronmo, 1991; Meyers-Levy e.g. Sparkman and Austin, 1980; Gronhaug, and Peracchio 1995
Shape of advertisement e.g. Drèze and Hussherr , 2003; Hussein, Sweeney and Mort, 2010
Animation type of advertisement Donthu and Hershberger, 2003; Chandon, e.g. Drèze and Hussherr, 2003; Lothia, Chtourou and Fortin, 2003
Size of the banner advertisement
One of the most extensively researched determinants of advertising effectiveness is the size of the advertisement. The general conception is that advertisement size usually improves memorization, which leads to brand recall (Chandon, Chtourou and Fortin, 2003). Studies on print advertising confirm that bigger advertisements lead to more attention, a bigger change of being seen and remembered and thus increasing brand recall (Finn, 1988; Kelly and Hoel, 1991). However, the positive effects of size on brand attitude are less evident. Homer (1995) showed that size has an inverted u-shape effect on brand attitude, size does increase the brand attitude but only until a certain level after which the advertisement is seen as manipulation.
Where the effect of advertisement size on brand recall is undoubtedly proven for print advertising there is no clear effect of size on click-through rates for online banner advertising. Robinson, Wysocka and Hand (2007) found a positive effect of banner size and click-through rates, even as Chandon, Chtourou and Fortin (2003). However, both Drèze and Hussher (2003) and Cho (2003) did not find any significant effects. Based on the findings for print advertising and because of the fact there are no negative effects found for the relationship of size on click-through rates a positive relationship is expected. These expectations are based on the research in the traditional media that have shown that large banners attract more attention, which in turn leads to a higher advertising effectiveness. Contrary to the findings in print advertising, it is not assumed that banner advertisements size will have an optimal value after which the size is negatively related with the effectiveness. This because of the fact that banner advertisements are restricted to particular sizes by media agencies,
Call to action e.g. Rossiter, 1981; Hofacker and Murphy, 1998
Obtrusiveness e.g. Heckler and Childers, 1992; Campbell, 1995; Goldfarb and Tucker, 2010
therefore abnormally large banners are not allowed. These expectations result in the following hypothesis:
H1: Size positively affects online banner advertising effectiveness.
Content of the banner advertisement
The content of a banner advertisement can consist out of images, text or both. The effect of the presence of images has also been extensively researched for traditional media but the results are mixed. Finn (1988) found positive relations between image presence and comprehension for 3 out of 5 studies and the effect on memorization was positive 8 out of 12 times. It has been proven that images generate more mental codes than verbal content and that these mental codes increase memorization of advertisements (Childers and Houston, 1984; Unnava and Burnkrant, 1991). However, Schweiger and Hruschka (1980) found no effect between the presence of a picture and the number of reactions on an advertisement. This is in line with the findings of Donhtu, Cherian and Bhargava (1993) who show that there is no significant relation between the image size and memorization for outdoor advertising. In the field of banner advertising effectiveness the amount of research is fairly limited, Chandon, Chtourou and Fortin (2003) found no effect of image presence on click-through rates. Following the work of Finn (1988) it is expected that an image in a banner advertisement will help consumers to comprehend a banner advertisement. This increased comprehension will give consumers more confidence that the advertised product is relevant for their needs and thus inducing a click-through. This leads to the following hypothesis:
H2: Image presence positively affects online banner advertising effectiveness.
assumed to increase the effectiveness of online banner advertisements because the bigger banner help to attract attention and the image component helps to increase advertisement comprehension. This results in the following hypothesis:
H3: The effect of image presence on online banner advertising effectiveness becomes more positive when banner size increases.
Another part of the content is the number of words used in an advertisement. As with the use of images also the research on the effect of the number of words is not consistent. Donthu, Cherian and Bhargava (1993) found a negative effect for the number of words on brand recall, contradicting with Schweiger and Hruschka (1980) who found a positive effect for the average number of words in an advertisement on inquiries based on the advertisement. For the online media environment Baltas (2003) found a negative effect of the message length on direct response. According to Krishnamurthy (2000) this is because long messages require that consumers pay close attention, which they rarely do. Furthermore, Baltas (2003) found that the effect of number of words on direct response follows a parabolic curve indicating a certain maximum number of words. Whereas Baltas (2003) suggest using short messages Robinson, Wysocka and Hand (2007) found that longer messages resulted in a higher click-through rate because more words gives the consumer a better comprehension of the advertised product. It is expected that an increased number of words not only could help a consumer in understanding the advertised product but also could help in persuading a consumer to click-through to the advertisers website. Although a positive effect of number of words on banner advertising effectiveness is expected also the results of Baltas (2003) are kept in mind and a hypothesis is formulated to test if there is a maximum number of words that should be used.
H4a: A larger number of words positively affect online banner advertising effectiveness.
Colors used in the banner advertisement
Next to pictorial and textual content the use of colors is also very important in advertisement design. Advertisements that make use of colors are found to elicit more attention than advertisements without colors. This higher attention will lead to higher brand awareness and recall (Sparkman and Austin, 1980). For the traditional media research on the effects of the use of color goes way back. Sparkman and Austin (1980) found that sales could increase with 41% if a one-color advertisement was used instead of a black-and-white advertisement when advertising price-reduced items in newspapers. Furthermore Gronhaug, Kvitastein and Gronmo (1991) found a positive effect between the use of colors and advertisement recognition and Meyers-Levy and Peracchio (1995) found that under low effort processing colored advertisement will outperform black-and-white advertisements.
For the effect on online banner advertising effectiveness the research is not that extensive yet as for traditional media. Moore, Stammerjohan and Coulter (2005) state that the use of colors will increase the recall and recognition of a banner advertisements when its colors are incongruent with the website and when the colors are congruent the attitude towards the advertisement is increased. The use of colors also has a positive effect on click-through rates but only until a certain level of colors, when too many colors are used the effect becomes negative (Lohtia, Donthu and Hershberger 2003). In accordance with prior research it is expected that increasing the number of colors will attract more attention from consumers and that this increased attention, just as size, will lead to a higher click-through rate. Although banners with a high number of colors will attract more attention than competitive banners that are not as colorful it is expected that there is a certain maximum. After this maximum the banner does stand out but is purposely avoided by consumers because of the redundant use of colors. This leads to the following hypotheses:
Shape of the banner advertisement
The factor shape refers to the physical shape of the banner. This can be for example a horizontal, vertical of square banner. According to Hussein, Sweeney and Mort (2010) most banners are horizontally or vertically shaped, accounting for 75% and 14% of the total banners respectively. However, not much research has been done on the effect of banner shape on advertising effectiveness. Drèze and Hussherr (2003) found that vertical banners generate a higher brand recall than horizontal banners. It is expected that vertical banners have a higher click-through rate than horizontal banners and other shaped banners because they are longer visible along side the website as consumers scroll down. Because they are longer visible they generate more attention, which probably will lead to a higher banner advertising effectiveness. This expectation leads to the following hypothesis:
H6: Vertical banners have a higher banner advertising effectiveness than horizontally or otherwise shaped banners.
Type of animation in the banner advertisement
Hussein, Sweeney and Mort (2010) state that there are multiple types banners, a banner can be for example static, animated or dynamic. A static banner is one that does not move or change contents every time when it loads and only contains one JPEG or GIF image. According to Chatterjee, Hoffman and Novak (2003) a static banner is passive because it does not interrupt the activity of the user of the website. When a banner consists of multiple JPEG or GIF images that are shown in succession the banner is called animated. The last type of banner is the dynamic banner. Dynamic banners are made-up of audio, video, Java or Flash. These banners almost look like TV commercials and have been designed to impact consumers the same way as a TV commercial does (Koegel, 2003).
there is not much research on animation in traditional literature. The results of studies of the effects of animation in online banner advertising are mixed. Robinson, Wysocka and Hand (2007) did not find significant changes in click-through rates whereas multiple other studies do find a positive relationship (Drèze and Hussherr, 2003; Lothia, Donthu and Hershberger, 2003; Chandon, Chtourou and Fortin, 2003). These positive effects are expected because animated banners give advertisers the opportunity to put more information in the same place. Advertisers can also create dramatic effects by delivery the message more sequentially (Drèze and Hussherr, 2003). It is expected that this increased flexibility and increased possibility to present information could lead to increased attention to banner advertisement and increased comprehension of the advertised product, which should translate to a higher banner advertising effectiveness. Thus, according with the existing literature the following hypothesis is proposed:
H7: Animated banners have a higher banner advertising effectiveness than static banners.
It is assumed that the combined attention grabbing effect of both a large banner and an animated banner will have a positive effect on the online banner advertising effectiveness (Chandon, Chtourou and Fortin, 2003). This combined increased attention grabbing is expected to lead to a higher online banner advertising effectiveness, resulting in the following hypothesis:
H8: The effect of animation on online banner advertising effectiveness becomes more positive when banner size increases.
Length of the animation
repeated and consumers have more time to absorb the information provided, these findings are supported by Fabian (1986).
However, for online banner advertising there has been little to no research in this field. Length of the animation can be seen as the time that a banner plays or the number of frames (i.e. number of single images) a banner consists of. Robinson, Wysocka and Hand (2007) found no significant effects for the length of animation on click-through rates. On the other hand Baltas (2003) found a negative effect for the number of frames in the animation on click-through rates. This negative effect is explained from the fact that multiple frames increase the complexity of the advertisement and that this increased complexity can have a negative effect on consumers’ attitude and response to the advertisement (Bruner and Kumar, 2000). A more practical problem according to Baltas (2003) could be that the download time for an animated banner is too long and consumers do not wait for the banner to load. However, because the study by Baltas (2003) took place in the early 2000’s the download speed is not a problem anymore in this study. Internet speeds have risen to such a level that all banner advertisements, with no regards to the length, can be downloaded instantly. It is expected that the findings on traditional media also apply for online banner advertising, thus that longer advertisements are able to generate a higher advertising effectiveness due to the longer absorption period. Therefore the following hypothesis is expected:
H9: Length of animation positively affects online banner advertising effectiveness.
Call to action
avoided. The use of these kinds of headings does not trigger consumers to read the advertisement whereas determines in headlines, adjectives in headlines and nonimperative headlines do increase overall readership.
However, the existing literature on the use of call to actions in online banner advertisements contradicts with the findings of Rossiter (1981). Hofacker and Murphy (1998) found that banners with only the text “Click here” performed four times better in generating click-throughs than all banners with nonimperative headlines. This positive effect is explained by the curiosity factor that comes with the phrase “Click here”, consumers are tempted to click on it. Because it is easy to maneuver around on the Internet there is not a big risk involved with clicking on a banner, therefore consumers give in with their curiosity (Hofacker and Murphy, 1998). Furthermore, Chandon, Chtourou and Fortin (2003) also found a positive effect between the presence of the phrase “Click here” and the obtained click-through rate. Combining the fact that a clear call to action evokes some kind of curiosity with the consumers and the fact that clicking on banner is not a high-risk action it is expected that including a call to action in the banner design will lead to a higher online advertising effectiveness. Therefore the following hypothesis is formulated:
H10: The presence of a call to action positively affects online banner advertising effectiveness.
Obtrusiveness
in such a way that they request an action from the consumer, the consumer should close the banner first before the web navigation can be continued.
From studies on print advertising can be concluded that highly visible advertisements are recalled more (Heckler and Childers, 1992). However, research in consumer behavior shows that obtrusive advertisements can make consumers think that they are manipulated by the advertisers, reducing purchase intentions (Campbell, 1995). For online banner advertising both Goldfarb and Tucker (2010) and Cole, Spalding and Fayer (2009) studied the effect of obtrusiveness on purchase intentions. Both studies found that obtrusive banners generate a higher purchase intention because they generate higher attention than banners that are just part of the website, whether or not these other banners are animated. As with the other attention generating variables such as size and use of colors it is expected that making a banner obtrusive will make it stand out more and thus increasing the click-through rate because the banner is harder for consumers to ignore. These expectations lead to the following hypothesis:
H11: Obtrusiveness positively affects online banner advertising effectiveness.
Interactivity
No academic research has been done on the presence of interactivity elements on banner advertising effectiveness. It is expected that the presence of an interactivity element has a positive effect on advertising effectiveness in the form of click-through rate. This because interactivity elements create some kind of curiosity among consumers, e.g. what can I win if I spin the wheel or what will my car insurance be with this insurance agency? It is found that creating curiosity in online advertising leads to positive effects on focus and memory (Menon and Soman, 2002). Thus if interactivity creates curiosity as expected the focus on the banner will be higher, probably leading to more click-throughs. This expectations lead to the following hypothesis:
H12: The presence of an interactivity item positively affects online banner advertising effectiveness.
2.3 Conceptual model
3. Research Design
In this chapter the research design is given. First it is discussed how the data is obtained, where it consists of and what the implications are of using this data. Then the various variables are constructed that are used in the mathematical model. The chapter is concluded with the final mathematical model what is used for testing the different hypotheses.
3.1 Data collection
The data that is used in this research is real-life data from actual online advertisements campaigns. The data is provided by Storm Marketing Consultants, a Groningen based online marketing firm with customers from all over the Netherlands that are active in many different industries. Data in the dataset is coming from campaigns that have run in the period from July 2010 to July 2011 and consists of 305 banners that vary in design elements, among industries and among media-buying networks
banner design, not only on industry characteristics. However, because differences between the four industries are expected the company is added to the model as a control variable. The control variable is constructed as a dummy variable with the car-insurance company being the base category.
As already stated in chapter two the online media landscape nowadays consists of various media-buying networks that control the advertising space on a set of websites. Online marketing companies buy advertising space with these media-buying networks not knowing on which websites their banners will be shown. Every media-buying network has a very distinctive set of websites where they control advertising space on. Because of this distinctiveness all the campaigns of the customers of Storm Marketing Consultants make use of different media-buying networks. Just as for companies it seems logical that one media-buying network generates higher click-through rates than the other. To account for this difference in click-through rates among networks, the network has been added to the dataset as a control variable. There are four networks (IMX, ClickDistrict, Telegraaf and Facebook) that have been used by the various campaigns; the media-buying network is constructed as a dummy variable with four dummies and the network IMX as the base category.
3.2 Variables of interest
Furthermore multiple independent variables are constructed to test the different hypotheses. These variables are constructed based on previous literature and the available data in the dataset. The variables are in line with the factors that were determined in the conceptual model in the previous section. In table 2 below the various variable construction and their sources can be found.
Table 2: Variable construction
Variable Variable
construction Variable measurement Source
Size Surface
measurement
Multiplying the pixel height and pixel width.
Wedel and Pieters (2000)
Image Presence 0/1 variable Observing whether or not an image is present.
Chandon, Chtourou and Fortin (2003)
Image * Size Interaction effect
Multiplying the image size with the image
presence variable.
Chtourou and Chandon (2000)
Words Count variable
Counting the number of words in the banner
advertisement.
Baltas (2003)
Words2 Quadratic effect
Square of the words to account for the maximum
number of words.
Baltas (2003)
Colors Count variable
Counting the number of colors in the banner
advertisement.
Colors2 Quadratic effect
Square of the colors to account for the maximum
number of words.
Lohtia, Donthu and Hershberger 2003 Shape Dummy variable with three categories Observing whether a banner is horizontal, vertical or otherwise shaped. Horizontal is the
base category. Hussein, Sweeney and Mort (2010) Type Dummy variable with three categories Observing whether a banner is static, GIF animated or Flash animated. Static is the
base category. Hussein, Sweeney and Mort (2010) Animation * Size Interaction effect
Multiplying the banner size with the sum of the dummies GIF animated
and Flash animated.
Chandon, Chtourou and Fortin, 2003 Length of animation Number of frames
Counting the number of frames a banner advertisement consists
of.
Robinson, Wysocka and Hand (2007)
Call to action 0/1 variable Observing whether or not a call to action is present.
Interactivity 0/1 variable
Observing whether or not an interactivity component is present.
-
3.3 Mathematical model
The control variables from section 3.1 and the constructed variables from section 3.2 are combined in a mathematical model that can be found in equation 1 below, also the specification of the different parameters is given. The mathematical model is a full model, including all campaigns from all companies that have run on all networks. In the next chapter this mathematical model is tested.
€
CTR = α + β1Size + β2Image + β3Image * Size + β4Words + β5Words 2
+β6Colors
+β7Colors 2
+β8Vertical + β9Other + β10GIF + β11Flash +β12Animation * Size
+β13Frames + β14CTA + β15Obtrusiveness + β16Interactivity + β17Utility
+β18KitchenAppliances + β19Non Pr ofit + β20Telegraaf + β21ClickDistrict
+β22Facebook + ε
Equation 1: Mathematical model of effects of design elements on click-through rate.
The parameters in equitation 1 above can be specified as follows: CTR = Click through rate.
α = intercept.
Size = records the size of the banner.
Colors = counts the number of colors used. Vertical = records if a banner is vertical shaped.
Other = records if a banner is shaped other than horizontal or vertical. GIF = records if the banner type is animated with GIF or not.
Flash = records if the banner type is animated with Flash or not.
Animation = measurers whether a banner is animated (GIF or Flash) or not. Frames = records the number of frames a banner consists of.
CTA = records if a call to action is present.
Obtrusiveness = set to one if a banner is obtrusive, zero otherwise. Interactivity = set to one if a banner is interactive, zero otherwise. Utility = set to one if the industry is utility companies, zero otherwise.
KitchenAppliances = set to one if the industry is kitchen appliances, zero otherwise. NonProfit – set to one if the industry is non-profit organizations, zero otherwise. Telegraaf = set to one if network used is Telegraaf, zero otherwise.
4. Results
This part of the paper covers the results of the tests performed on the mathematical model proposed in chapter 3 of this research. This linear additive model is tested by performing an OLS regression with the use of the statistical software package SPSS. In the first part the statistical validity of the model is tested after which the model fit is discussed, the individual parameters are evaluated and it is checked whether the hypotheses are supported or not.
Before the actual testing is done one more alteration to the dataset is made. The dependent variable, the click-through rate, is multiplied with 1000. This because of the fact that the average click-through in the dataset is 0.073%, so in numbers this is an average of 0.000743. Because this number is so small it is hard to identify the size of the effects after the linear regression is done because the unstandardized betas in SPSS only have three decimals. The significance and sign are not changed if the dependent variable is multiplied by a constant, only the sizes of the effects are better interpretable. Of course to obtain the real size of the effect on the click-through rates the sizes of the effects have to be divided by 1000 again.
4.1 Statistical validity
In order to make sure that the conclusions that are drawn based on the SPSS output are statistically sound the regression analysis has to be validated over the requirements for a solid OLS analysis: multicollinearity, autocorrelation, heteroskedasticity and normality.
Multicollinearity
correlating variables are the variables that are used to calculate the interaction effects, the interaction effects itself and the squared terms of the variables “Words used” and “Colors used”. This is not a surprising result, according to Cortina (1993) it is not unlikely to find strong multicollinearity when interaction variables are used. The interaction variable tends to move strongly together with the separate variables that are multiplied in the interaction effect.
To remedy multicollinearity Leeflang et al. (2010) suggest a few solutions. One of them is to eliminate predictor variables with a statistically insignificant t-ratio. However, most of the t-ratios of the correlation variables are above the critical value of t (0.05, 22) = 1.7171 and the adjusted R2 drops from 0.264 to 0.261 if the variables are removed. Hence, the model is better with the variables included and removing the variables is not a good solution. Another option to correct for multicollinearity is reformulate the model or the predictors. A possible way to do this is to center all the independent variables around their own means. After recoding the variables the model is estimated again and all the VIF values are below ten so there is no more sign of multicollinearity.
Autocorrelation
Heteroskedasticity
To test for heteroskedasticity the dataset is divided into two samples based on the size of the banner advertisement. The average advertisement surface is taken and used as a split point resulting in 107 observations below the average and 198 banners whose size is above average. For these two samples the Goldfeld-Quandt test (Leeflang et al., 2010) is used to test for the equality of variances of the residuals. The Goldfeld-Quandt test leaves an F-value of 25.258, which results in a p-value of 0.000 (F=25.258, 85,176). Because of this low p-value the hypothesis that the variances of the residuals are equal is rejected, breaking the assumption of homoskedasticity.
To correct for this heteroskedasticity GLS is applied. All the variables in the dataset are divided by the standard deviation of the unstandardized residual of the sample they belong to. After applying GLS the Goldfeld-Quandt test is performed again resulting in an F-value of 1.091, which results in a p-value of 0.67 (F=1.091, 176,85) indicating that the variances of the residuals are indeed equal.
Normality
Figure 2: Normal Q-Q Plot of unstandardized residuals
The visual detection of non-normality is supported by the results of the Kolomogorov-Smirnov and the Shapiro-Wilk statistics (Leeflang et al., 2010). Both tests have a p-value of 0.000 so that the H0 hypothesis, which assumes that the residuals are normally distributed, is rejected. Outliers can influence the Kolomogorov-Smirnov and the Shapiro-Wilk statistics but it is not likely that the removal of the two outliers to the right will eliminate all non-normality. Because it is very hard to remedy non-normality the model is estimated with knowledge that there is non-normality in the residuals. This non-normality makes the p-values of the parameters less reliable, so parameters with p-values that are close to the significance level have to be looked upon with caution.
4.2 Model fit
the variance in the dependent variable can be explained by the independent variables in the model. When the R2 is adjusted for the number of predictors in the model it
becomes 0.410.
4.3 Investigation of the parameters
The final model has a total of 23 parameters (including constant). From these 23 parameters 5 parameters are significant at a 5% significance level besides the constant and 1 parameter is significant at the 10% level. All the parameters are summarized in table 3 below together with their significance levels and their unstandardized betas, the statistically significant parameters are in bold.
Table 3: Individual parameters
Parameter Beta P-value
Colors Squared .021 .172 Vertical .317 .080* Other -.004 .946 GIF .098 .453 Flash -.067 .626 Animation*Size .000 .557 Frames -.023 .446 Call-to-action -.015 .798 Obtrusiveness -.046 .444 Interactivity .015 .000
* = significant at the 10% confidence level.
From table 3 can be concluded that from the 16 variables that are used for hypothesis testing there are four significant individual parameters indicating that there are four design elements that influence the click-through rate significantly. These parameters are the banner size, the image presence, whether or not a banner is vertical and the interactivity component. These individual parameters and their influence on the hypothesis are further evaluated in section 4.4.
model, the media-buying network, is not significant on any of the three dummy variables. Thus there is no significant difference in the click-through rate between any of the three networks in the model and the network that is selected as base category.
4.4 Hypothesis evaluation
Hypothesis H1 assumed a positive effect of the banner size on the advertising effectiveness of that banner. The independent variable size was significant (p = 0.031) at the 5% significance level, indicating that there is indeed an effect between size and click-through rate. However, size is not positive as expected based on previous studies. The unstandardized beta of size is too small to draw any conclusions on but if the standardized beta is examined it is concluded that the effect is negative (standardized beta is -0.162). Thus H1 is not supported.
Image presence has a significant and positive effect on advertising effectiveness (p = 0.000), supporting H2. The interaction effect between size and image presence that is assumed by H3 is not supported because the interaction term is not statically significant (p = 0.637). The variables that are constructed to test H4a, H4b, H5a and
H5b, words used and colors used (p = 0.428 and p = 0.989) as well as the squared
terms of both variables (p = 0.960 and p = 0.172), are not significant at either the 5% or 10% level. So it seems that the number of words used and number of colors used have no effect on advertising effectiveness and there is also no maximum number of words and colors that can be used before the effect becomes negative.
If the significance level for parameter testing is stretched to 10% support can be found for H6, the positive effect of the vertical shape of a banner on the click-through rate (p = 0.080). However, it is concluded in section 4.2 that there still is non-normality in the dataset so the p-values can be disturbed. Therefore H6 is accepted but with caution. The three hypotheses that are related to the animation effect of a banner, H7,
H8 and H9, are not supported. Both the dummy variable created to record the
between animation and size are not significant (p = 0.557) as is the variable that counts the number of frames (p = 0.446). Thus so significant difference in click-through rate is found between the banners that are static or flash animated or the banners that have only 1 frame versus banners that have multiple frames.
No support is found for H10 because the variable that records the presence of a call-to-action has no significant effect (p = 0.798) on the advertising effectiveness. Also the obtrusiveness variable is not significant (p = 0.444) so H11 is not supported. Support is found for H12, the positive effect of interactivity on advertising effectiveness, because of the positive and significant effect of the interactivity variable (p = 0.000).
All hypothesis, their expected effects and conclusions from the performed tests are summarized in table 3 below.
Table 3: Hypotheses evaluation
Hypothesis Variable Expected effect Supported?
H1 Size + No*
H2 Image presence + Yes
H3 Size * Image + No**
H4a Words used + No**
H4b Words squared - No**
H5a Colors used + No**
H5b Colors squared - No**
H6 Vertical shape + Yes***
H7 Flash + No**
H9 Number of frames + No**
H10 Call to action + No**
H11 Obtrusiveness + No**
H12 Interactivity + Yes
* = significant but effect opposite of expected effect ** = obtained effect is not significant
5. Conclusions, recommendations, limitations and future research
In this final chapter conclusions are drawn based on the results of the research and recommendations are made for Storm Marketing Consultants and online marketing agencies alike. Furthermore the limitations of the study are discussed and directions for future research are given.5.1 Conclusions
The goal of this research was to construct a model that could be used to identify the relevant design elements when it comes to online banner advertisements. The design elements were mainly based on findings from research in the field of the more traditional media. Besides the elements that have proven to be effective in the traditional media, elements were added that specifically come with online advertising. All these design elements were combined in a linear additive model where the click-through rate is used to measure online advertising effectiveness. This research differs from other studies on online banner advertising effectiveness in a way that this model combines a total of ten design elements where other papers often focus on only one. To test the model data was used from the database of Storm Marketing Consultants. Results from online advertising campaigns from four different companies that are all active in different industries were included. From the obtained results it can be concluded that the size of the banner, the image presence, the shape of the banner and the presence of an interactivity component significantly influence the online banner advertising effectiveness.
consumers maybe process the content unconsciously and then make a decision whether to click on it or not.
Where Chandon, Chtourou and Fortin (2003) found no significant effect of image presence on click-through rate, a significant positive effect is found is this study. The positive effect of image presence in a banner advertisement on online advertising effectiveness means that including an image in the banner design will lead to a higher click-through rate. Finn (1988) found positive relationships between image presence and consumers comprehension of the advertisements. This positive relationship can imply that consumers better understand that a banner advertises for a product that suits their needs if an image is present to clarify the advertised product, this increased understanding could lead to a higher click-through.
Banner advertisements that are vertically shaped have a higher click-through rate than banners that are horizontally or otherwise shaped. This is in line with the findings of Drèze and Hussherr (2003) who found a higher brand recall for vertically shaped banner advertisements. Vertical banner advertisements are mostly placed next to the content of the website which makes them longer visible than horizontal banners as consumers scroll down. Because they are longer visible they will attract more attention by consumers and because more consumers consciously process the banner the online banner advertising effectiveness will be higher.
Besides the four design elements that have shown a significant effect on online advertising effectiveness, there are also six design elements that seem to have no influence. No significant effect is found for the number of words used in the banner advertisement on click-through rate. This was not expected because a lot of previous studies found, although contradicting, significant effects (e.g. Baltas, 2003 and Robinson, Wysocka and Hand, 2007). Also the expected positive effect of the use of colors is not confirmed whereas the effect is often confirmed in research in the field of traditional media (Gronhaug, Kvitastein and Gronmo, 1991) and the online environment (Lohtia, Donthu and Hershberger 2003). Furthermore no effect has been found of the type of the banner and the number of frames used on online advertising effectiveness. This means that there is no significant difference in click-through rate between static banners, GIF banners or flash banners. This result is in line with the results of Robinson, Wysocka and Hand (2007) who also found no significant effect but it contradicts with for example Drèze and Hussherr (2003) who did found a significant positive effect. Contradicting with Goldfarb and Tucker (2010) and against expectations also obtrusiveness shows no significant effect on click-through rate. However, this result is quite in line with the rest of the results of this research that showed that bigger banners have a negative effect on click-through rate and that other high attention-grabbing elements such as animated banners also have no effect on online advertising effectiveness.
the results can also be concluded that there is no significant difference in click-through rates among the different media-buying networks.
When referring back to the research question that has been proposed in the introduction it can be said that the best way to measure online advertising effectiveness is still the click-through rate when compared to the mindset measurements such as brand recall and brand awareness. Furthermore it is found that of the design elements there are not a lot of variables that contribute to the advertising effectiveness significantly. The main conclusion about the design elements is that creating banner advertisements that are just attention grabbing is not going to lead to a higher click-through rate. This is based on the fact that size has a negative effect on click-through rate and that all the attention grabbing design elements such as high number of colors, an animated Flash banner or an obtrusiveness component have no significant effect on online advertising effectiveness. What does increase click-through rate is designing a banner in such a way that it is evoking some kind of curiosity. Adding an image or adding an interactivity component, in which the consumer is evoked to really interact with the banner at the pre-click level, are ways to create this curiosity.
5.2 Recommendations
total of ten different sizes are used, which also vary across shape. From these ten different formats only two are vertical banners, the sizes 120x600 and 160x600. Following the results of the research it is recommended to design the banner advertisement in the 120x600 format. Although this recommendation seems straightforward most media-buying network don’t allow companies such as Storm Marketing Consultants to only provide the vertical banners. They mostly require all the sizes and shapes they use on their advertising websites for every banner campaign. To overcome this problem online marketing agencies should create all banners sizes required by the media-buying network but on the other hand they should increase the cost-per-clicks bids for the vertically shaped banners. Increasing the cost-per-click bids for vertical banners above the industry average will lead to the situation that these banners will be shown more often than banner of the same size with a lower cost-per-click bid. On the other size the bids on big and horizontal banners should be lowered resulting in the fact that the banners that yield a lower click-through rate are shown less often than the banners of competitors of the same size with industry average bids. In this way the banner advertisement with respectively the highest click-through rate are shown more often than those with a lower click-click-through rate.
The second recommendation is that images should be included in all banner designs. According to the results adding an image will lead to a higher click-through rate, probably because images are proven to help consumers to better comprehend a banner advertisement. This increased understanding of the advertisement can give consumers the feeling that the advertised product fits their needs, thus inducing a click-through.
5.3 Limitations
As with every research there are some limitations to this study. The first limitation comes from the amount of data that is available. The first purpose of this research was to create two dimensions, which both consists of various elements that could influence online advertising effectiveness. Next to the in this research tested design elements there are also elements that are related to the placement and use of the banner. Examples of these elements are the place on the website, the number of competitive banners on these websites and the kind of website the banner is displayed on. However, for this research the only data that was available was the data given by the different media-buying networks and this was fairly limited. The media-buying networks only provided the click-through rate for every unique creative banner design in a specific campaign, no data was provided about the place on the website or which website the banner was displayed on. Thus the only thing that was available was how the banner looked like, on which the design elements the banner is based, and which click-through rate in generated. Another piece of missing data was the conversion rate so no information is given about the actual goals a specific banner completed. This can lead to the fact that the variables that in this research are proven to have a positive effect on click-through rate do not have a positive effect on conversion rate. So it can be that the recommended design elements only generate website traffic and do not help in completing the goals which are set for a specific banner advertisement campaign.
past there was no opportunity to influence the banner design and then check for changes in online advertising effectiveness. Also because the data is only from companies that are based in the Netherlands the results are not generalizable across other countries.
In section 4.1 it is already concluded that there still is some non-normality in the residuals after the dataset is corrected for multicollinearity and heteroskedasticity. This is also a limitation to the research because the model is estimated with this in mind, so the obtained p-values can be biased.
5.4 Directions for future research
Although this research already contributes to existing literature because it combines multiple design elements in order to determine online advertisement effectiveness, whereas most other papers focus on only one, there are still a lot of opportunities for future research. First of all the model of this paper could be extended with elements that determine the effects of placement and use on the click-through rate. At the moment of writing this paper Storm Marketing Consultants is working on setting up a so-called Ad-server that could be used to measure a lot more than just the click-through rate the media-buying agencies are providing. This Ad-server is capable of giving information for example about the place on the website where the banner is shown, the type of the website the banners is shown on, the view-time (i.e. how long a banner was visible to the consumer) and the conversion rate. With this data the current model could be extended and online advertising effectiveness could be a measure of both click-through rate and conversion rate.
References
Baltas, George (2003), “Determinants of Internet Advertising Effectiveness: An Emperical Study,” International Journal of Marketing Research, 45 (4), 505-513.
Bruner II, Gordon C. and Anand Kumar (2000), “Web Commercials and Advertising Hierachy-of-Effects”, Journal of Advertising Research, 40 (1/2), 35-42.
Campbell, Margaret C. (1995), “When Attention-Getting Advertising Tactics Elicit Consumer Inferences of Manipulative Intent: The Importance of Balancing Benefits and Investments,” Journal of Consumer Psychology, 4 (3), 225-254.
Chandon, Jean Louis, Mohamed Saber Chtourou and David R. Fortin (2003), “Effects of Configuration and Exposure Levels on Response to Web Advertisements,” Journal
of Advertising Research, 43 (2), 217-229.
Chatterjee, Patrali, Danna L. Hoffman and Thomas P. Novak (2003), “Modeling the Clickstream: Implications for Web-Based Advertising Efforts,” Marketing Science, 22 (4), 520-541.
Chen, Jianqing, De Liu and Andrew B. Whinston (2009), “Auctioning Keywords in Online Search,” Journal of Marketing, 73 (July), 125-141.
Cho, Chang-Hoan (2003), “The effectiveness of banner advertisements: involvement and click-through”, Journalism and Mass Communication Quaterly, 80 (3), 623-645.
Childers, Terry L. and Micheal J. Houston (1984), “Conditions for a Picture-Superiority Effect on Consumer Memory”, Journal of Consumer Research, 11 (2), 643-654.
Cole, Sally G., Leah Spalding, and Amy Fayer (2009) “The Brand Value of Rich Media and Video Ads,” DoubleClick Research Report, June.
Cortina, Jose M. (1993), “Interaction, nonlinearity and multicollinearity: implications for multiple regression”, Journal of Management, Winter
Deighton, John (1996), “The Future of Interactive Marketing,” Harvard Business
Donthu, Naveen, Joseph Cherian and Mukesh Bhargava (1993), “Factors Influencing Recall of Outdoor Advertising,” Journal of Advertising Research, 33 (3), 64-72.
Drèze, Xavier and François-Xavier Hussherr (2003), “Internet Advertising: Is Anybody Watching?,” Journal of Interactive Marketing, 17 (4), 8-23.
Edwards, Steven M., Hairong Li and Joo-Hyun Lee (2002), “Forced Exposure and Psychological Reactance: Antecedents and Consequences of the Perceived Intrusiveness of Pop-Up Ads,” Journal of Advertising, 31 (3), 83-95.
Fabian, George S. (1986), “15-Second Commercials: The Inevitable Solution”,
Journal of Advertising Research, 26 (4), RC3-RC5.
Finn, Adam (1988), “Print Ad Recognition Readership Scores: An Information Processing Perspective,” Journal of Marketing Research, 25 (2), 168-177.
Fulgoni, Gian M. and Marie Pauline Mörn (2009), “Whither the Click? How Online Advertising Works,” Journal of Advertising Research, 49 (2), 134-142.
Goldfarb, Avi and Catherine Tucker (2010), “Online Display Advertising: Targeting and Obtrusiveness,” Marketing Science, forthcoming
Gronhaug, Kjell, Olav Kvitastein and Sigmund Gronmo (1991), “Factors Moderating Advertising Effectiveness as Reflected in 333 Tested Advertisement,” Journal of
Advertising Research, 31 (5), 42-50.
Heckler, Susan E. and Terry L. Childers (1992), “The Role of Expectancy and Relevancy in Memory for Verbal and Visual Information: What is Incongruency?”
Journal of Consumer Research, 18 (March), 475-492.
Hodges, Jane (1997), “Online classifieds heat up”, Advertising age, 68(22), 14-19.
Hofacker, Charles F. and Jamie Murphy (1998), “World Wide Web banner advertisement copy testing,” European Journal of Marketing, 32 (7/8), 703-712.
