1 Introduction
1.1 Effects of Banner Advertising on the Internet
Since the Internet magazine HotWired and 14
sponsors placed the first advertising graphics on the Web in
1994,
banners and buttons have become a
prevalent form of promotion on the Internet. The main goal of
Web banners is to inform users about the existence of
particular Web sites and to persuade customers to visit the
advertised sites. Thus, banners should attract the users'
attention, and they need to appeal to them to trigger the
desired user reaction. Normally, this is realized by
animation and interactive features.
1.1.1 Click-Through Rates for Web Banners¹
Since advertisements began appearing on the Internet, the
user reaction to banners has been disappointingly low. Williamson (1996) reported an
average click-through rate of 2.1%. According to Nielsen (1997; 2000),
click-through rates have since then constantly decreased from
2% (1995) to 0.5% (Oct. 1998), and reached 0.2% in May 2000.
Sutherland (1999) also found that
click-through rates had declined to 0.5% in 1999. Search
engines like Infoseek and WebCrawler were
found to have lower click-through rates (1.1% and 0.7%,
respectively) than content providers like HotWired (2.8%
according to Williamson, 1996). A
study by DoubleClick (1996)
revealed that banners have a substantial "burnout rate": The
probability of clicking on a banner drops from a maximum of
2.7% to less than 1% at the fourth exposure.
1.1.2 Acceptance of Web Banners and Impact on Purchase Intentions
Some studies found advertising banners to be reasonably successful in terms of acceptance, however. An Internet survey by market researcher COBUS with 1178 German Internet users showed that banners with an appealing design do attract the users' attention (Leest 1996; Belz 1997). Nearly half the respondents reported that they looked at advertising banners if they were well designed. Less than 25% felt that banner advertisements were disruptive; only a third of the respondents said they ignored banners altogether. The Internet Advertising Bureau (1998) found similar results: 60% to 70% of the respondents of this survey approved advertisement on the Internet. Using two matched samples, Schlosser et al. (1999) compared attitudes to Internet advertising and to advertising in general. Although their results showed that, overall, Internet advertising was more disliked than advertising in general, Internet advertising was found to be superior in terms of customer trust. Wong (2001) surveyed 80 Internet users on their attitudes towards advertising banners and found that 80% had a favorable opinion towards banners on the Web.
Furthermore, it was shown that banner advertisements can have
an impact on the purchase intentions of customers and a
positive influence on their assessment of the advertised
products (Briggs and Hollis 1997).
Likewise, a study by Gruner + Jahr
Electronic Media Service and Media Transfer (1999) with
2000 Internet users revealed that brand names were rated more
positively after users had looked at advertisement banners
for those brands.
1.1.3 Do Users Suffer from Banner Blindness?
A few reasons might be considered to explain why click-through rates are exceptionally low despite a reasonable acceptance. One explanation might be that the majority of users are just not interested in the advertised product or Web site, and thus have no reason to click on a banner. Users may also have learned that advertising banners often do deliver what they promise. Consequently, they consciously ignore them (Nielsen 1997, Drèze and Hussherr 1999). Another reason, as several studies suggest, is that many users simply do not notice banners on Web sites.
Benway (1998; 1999) as well as Benway and Lane (1998) report a series of usability studies in which they coin the term "banner blindness". Their subjects' task was to locate single information units on a Web page that could be found faster by clicking on a graphical full banner. In Benway and Lane's two experiments the vast majority of users did not use the banners for information retrieval. The first of these experiments showed that subjects had significantly more difficulty finding information when it was accessible via a banner than when it could be accessed via a text link. Benway and Lane concluded from the results that users are very likely to overlook information that is placed in graphical banners.
To find out if the banners' design has an impact on whether or not the information is noticed, Benway and Lane conducted a second experiment. They asked 72 subjects to find information while some features of the banners (positioning on the page, colors, animation, similarity to an advertising banner) were varied. A recognition test was carried out afterwards and it was found that more than 75% of the subjects (55 out of 71²) were not able to recognize banners that could have been relevant for the search task. Design and positioning of the banners did not have a significant effect on banner recognition. In addition, the test Web site also contained banners that were not relevant to the search task. In a post-experimental questionnaire, 80% of the subjects reported that they had not noticed these banners. Thus, it seems that a clue for the low click-through rates lies within the (non-)perception of banners.
"Banner blindness", however, does not seem to occur in general. Bachofer (1998) found considerably higher recall and recognition scores for banner advertisements than those reported in the Benway experiments. He exposed 71 subjects to several Web pages of a German online magazine that also contained graphical banner ads. Using an eye tracking device, he found that the mean time for watching a banner is 1.1 sec. On average subjects focused on a banner 1.5 times while looking at a page. The frequency and duration of fixation on the banners was found to be correlated with the performance in a subsequent recognition test. Overall, subjects correctly reproduced 11% of the formal elements for each banner. The positioning of banners did not have a significant effect on recall and recognition in Bachofer´s study either. Compared to the usual performance in recognition tests in advertising research, this percentage is similar to the recognition rate of ads studied incidentally in print magazines (Perfect and Askew 1994). The similarity of banner ads to other forms of advertisement is further corroborated by Ipsos-ASI (1999) who found that immediate recall of an online static banner ad was comparable to that of a 30 sec. television commercial.
Similarly, the Stanford Poynter Project (2000) found that 45% of advertising banners placed on the pages of an online magazine were looked at for more than 1 sec., a percentage that was regarded as surprisingly high. In a study similar to that of Benway (1998), Bayles (2000) found rather high banner recognition scores: 74% of her 35 subjects recognized the two banners presented in her experiment correctly. In contrast to Benway's study, subjects in Bayles' experiment were asked to complete search tasks on individual Web pages, each one containing a banner ad. Thus, no navigation or browsing was involved in Bayles' experiment. As explained in the next section, this is a crucial difference, which might explain the differences in banner recognition found.
Bachofer´s (1998) results
cannot be directly compared to those of Benway (1999) due to differences in
research methodologies, yet it can be assumed that in
Bachofer's study the perception of banners was substantially
higher, since for each banner at least parts were correctly
recalled.
1.2 A Psychological Perspective on Different Navigational Behaviors in Web Sites with Banner Ads
1.2.1 Navigation Style and Banner Perception
What might be the reason for the contradictory results concerning the (non-)perception of banners on the Web? An important difference between Benway's and Bachofer's studies is the difference in their instructions. While Benway asked the subjects to locate specific information on a Web site, Bachofer's subjects were instructed to visit single Web pages on a bookmark list and to investigate them as they liked. Thus, the two studies induced opposing navigational behaviors.
The task in Benway's study was to find short items of information such as prices for products, phone numbers or names. This task most likely triggered a navigation style known as "searching" (Canter et al. 1986), "search browsing" (Cove and Walsh 1988) or "directed browsing" (Kuhlen 1992; Tergan 1995). When users have a precise image of the information they are looking for and of its supposed form (e.g. numbers or single words), they tend to navigate in a very directed and structured manner to find the target as quickly as possible.
In Bachofer's study, on the other hand, subjects did not
search for information. In fact, they did not even really
navigate; they only called up individual Web pages one after
the other. This activity can best be characterized as
"associative browsing" (Tergan 1995)
or "undirected browsing" (Kuhlen
1992): users do not have a concrete plan to find a
particular object but instead are guided by the appeal of the
information offered and respond to those stimuli that catch
their attention first.
1.2.2 Schema Activation and Goal Directed Search for Information
We assume that navigation style exerts a significant influence on attention focusing and information processing. The directed search for information, as induced in Benway's experiments, focusses users' attention on content that is expected to contain relevant information to answer the search tasks. The underlying process is a top-down process, i.e. users direct their attention consciously towards areas of the Web site where they anticipate relevant information. This process is guided by cognitive schemata that play an important role for the direction of visual attention. They structure the information processing and help the user anticipate what information needs to be voluntarily selected (Neisser 1979).
Because Benway's subjects were highly experienced in using the Internet (90% reported that they use the Internet daily or several times per week), it can be assumed that they possess elaborate schemata of Web site structures and Web page elements. In addition, they probably also have a schema of the form and function of advertising banners. When the users search for information, banner related schemata are most probably not activated or are inhibited because the information the users are looking for is normally located somewhere in the text of a Web page but rarely in a banner. Thus, there is no priming for rectangular flashing graphics with text and the users' attention is hardly directed towards these objects. Instead, the users continue to read the text beneath the banner right away (Norman 1999). Consequently, banners are rarely processed at a higher level and are, therefore, hardly recalled or recognized in subsequent tests.
The existence of Web related schemata is supported by the
studies of Morkes and Nielsen (1997),
which compared reading behavior on the Web with print media.
It was found that subjects who purposefully look for
information rarely read the text of a Web page word by word
but that they scan the page for relevant bits of information.
When a new page is called up, schemata direct the users'
focus to the center of the screen where the main text is
located, before their attention is directed towards the
heading or the navigation elements.
1.2.3 Orientation Reaction and Aimless Browsing
On the other hand, an animated banner forces an orientation reaction, especially if its color contrast changes or if it simulates motion. An important function of an orientation reaction is to focus attention on newly appearing or changing stimuli.
When browsing associatively as in Bachofer (1998), a high degree of such
involuntary attention focusing might be anticipated because
the users are not looking for specific information but are
guided by the appeal of the different features on a Web page.
Under this condition, attention focusing follows a bottom-up
process: at first, a stimulus attracts users' attention
involuntarily, and consequently, voluntary attention is
directed toward this stimulus (Prinz
1992). Also, Web site-related search schemata are not
activated as strongly as in information retrieval to guide
browsing behavior. Thus, banners are more likely to be
noticed because, as mentioned above, changes in color
contrast and movement will cause an orientation reaction,
followed by voluntary attention focusing. Consequently, under
this condition banners should be remembered more easily than
when the user is searching for information.
2 Hypothesis
We assume that in the case of aimless browsing there is a
higher chance that orientation reactions towards banner ads
prevail against activated search schemata. Banner ads will
then be processed at a deeper level than under the condition
of goal directed search for information. On the basis of the
supposed differences in perception and information
processing, we expect a group of aimless browsing subjects
(ALB group) to achieve better results in subsequent banner
recall and recognition tests than a group of subjects who
search for information in a goal directed way (GDSI group).
For the significance level an alpha of 5% will be used.
3 Method
3.1 Subjects
The subjects recruited for this study were expected to be familiar with the function of Internet browsers and to have spent at least part of their online time with goal directed search for information. This way we intended to make sure that the assumed schemata for Web searching already existed. All of the 32 subjects were students, 19 to 43 years old (mean = 24.8 years, females: 19, males: 13). We also asked the subjects to assess the mean time per week they spent on the Web. Subjects had to distinguish between time spent for aimless browsing and goal directed search for information (ALB/GDSI time). None of the subjects was unclear about this distinction. As the accuracy of users' self-assessment of their different types of Web use can be questioned, the values are interpreted only as a rough estimate of their browsing and search times. The mean for ALB time was 60.8 min.; for GDSI it was 244.6 min. The distribution for these variables in the treatment groups is shown in Table 1.
ALB group | GDSI group | |
---|---|---|
Subjects | 16 | 16 |
Males / females | 6 / 10 | 7 / 9 |
Mean age (SD) | 24.6 (6.12) | 25.0 (3.72) |
Mean ALB time in min. (SD) |
72.2 (95.20) | 49.4 (88.09) |
Mean GDSI time in min. (SD) |
159.4 (131.78) | 208.1 (147.57) |
3.2 Materials
The experimental Web site was a modified version of the Web site of the Center for Media Research (CMR), which was locally stored on the hard disk of a network server. In the content frame [Fig. 1, part 3], it was possible to investigate 55 HTML pages including texts, tables and pictures concerning the CMR's employees, projects, course contents and research interests. The support frame [Fig. 1, part 2] was meant to make navigation easier by showing the title of the respective submenu. In the navigation frame [Fig. 1, part 1], subjects were presented with 16 animated full graphical banner ads (see Supplementary Files to view the view the banners), one after the other, each one for a duration of 45 sec. The banner ad sequence was randomly varied for each subject. The banner ads were placed in the navigation frame and the browser window was used in maximized screen mode, so this presentation area was always on screen during the browsing or searching phase.
The sample of 16 banner ads that became the final
experimental ads were selected from a total of 83 ads
collected from various Web sites and archives. We decided to
use real banner ads and to modify them slightly in color and
speed of single picture changes to cover the variety of real
Web banner ads. Also, since Bachofer tested just two animated
ads and did not get clear results for recognition nor recall,
we wanted to include a broader variety of banners. Color
contrasts and the speed with which banners were changed, as
well as color contrasts between banner and frame background,
were subjects of a later exploratory data analysis that was
supposed to yield more information on the role of design
features for recall and recognition of banner ads.
3.3 Procedure
To induce goal directed search for information, subjects were asked to find as many answers as possible to a list of questions³, which had been handed out on an extra sheet of paper. Subjects were asked to note brief answers on the paper. Each answer could only be found on one Web page, and subjects were free to answer the questions in any order they liked.
The task for the subjects of the aimless browsing condition was to look through the Web pages according to their own interest. They were asked to find out about the contents of the Web pages and, if they found something interesting, to read the pages partly or completely as they liked. No further restrictions or instructions were given to this group because we did not want to impose an external goal on them.
After 15 min. of exploring the Web site's contents, a blank page opened automatically. The subjects were told to take a break of 7 min.
After the break, subjects were asked to complete the recall and recognition test. The free recall test was divided into three subsections concerning details of the presented banner ads:
- advertised companies, products or services;
- advertising copies and slogans;
- pictorial motives and basic colors.
For each banner they remembered subjects were asked to write down what company or product was advertised, the slogan or text of the banner and any of its colors. Only answers that could be related by the researchers to one of the banners presented during the experiment where counted for the subsequent analysis. While the recall questions were paper-and-pencil based, the recognition test was supported by an HTML page (Figure 2) presenting the experimental banner ads in a similar context but again the sequence of appearance was randomly varied. Subjects then had to look at the bannersd and judge on a five-point rating scale how sure they were that they had seen it in the test before. A low scale rating (0) indicates that the person is certain that he/she has not seen the ad before, a high scale rating (4) means that he/she is certain of having seen it.
Figure 2. Screenshot of the support site for the recognition test (click on image for larger version)
4 Results
We carried out four univariate analyses of variance on the recall and recognition scores, each of them including navigation style as the fixed factor and five covariates (age, gender, previous exposure to the Web site, mean time for information search, mean time for aimless browsing).
According to the hypothesis, first the number of recalled banner details for each category and subject was compared between the two treatment groups (Table 2). As we expected, the reproduced details of advertised companies, products or services was significantly higher for the ALB group than for the GDSI group (F = 9.004, p = .006). The mean for the ALB group was 1.00 (SD = 1.41) while it was .06 for the GDSI group (SD = .25). For the second dependent variable, the advertising copies and slogans, the hypothesis could not be accepted (F = 1.359, p = .255). The mean was .63 for the ALB group (SD = 1.31) and .25 for GDSI group (SD = .45). The recalled pictorial motives and basic colors again showed a significant difference between groups in the predicted direction (F = 7.899, p = .009). Here, the ALB group reached a mean of 1.87 (SD = 1.59) and the GDSI group a mean of .63 (SD = .72). In this analysis the covariate "mean time for aimless browsing" also showed a significant impact on the recall scores (F = 5.798, p = .024).
The last analysis examined the certainty judgments of having seen the banner ads during the experimental phase, that is, the mean recognition scores. For this dependent variable the difference between treatment groups was also significant in the predicted direction (F = 10.470, p = .003). For the recognition performance the ALB group showed a mean of 1.52 (SD = .92) while the GDSI group reached a mean of .75 (SD = .57).
ALB group | GDSI group | ||
---|---|---|---|
Recalled advertised companies, products or services |
1.00 (1.41) | 0.06 (0.25) | p<.01 |
Recalled advertising copies and slogans |
0.63 (1.31) | 0.25 (0.45) | n.s. |
Recalled pictorial motives and basic colors | 1.87 (1.59) | 0.63 (0.72) | p<.05 |
Recognized banner ads | 1.52 (0.92) | 0.75 (0.57) | p<.01 |
5 Discussion
The results show that as expected, the recall as well as the recognition scores for banners were higher when subjects browsed aimlessly than when they performed a goal directed search for information. Thus, the hypothesis that navigation behavior has an impact on the (non-)perception of Web banners can generally be accepted. It seems that people who are browsing aimlessly are more susceptible to perceiving a Web banner, because the banners' color contrast or animation trigger an orientation reaction that is followed by a bottom-up process of information processing. Subjects who search for information, on the other hand, seem to apply cognitive schemata that suppress a deeper processing of Web banners.
It can thus be concluded that, especially on Web sites that users access to look for specific information (e.g. search engines), it is likely that banners are overlooked. In particular, if the desired information is normally not contained in a banner or a similar graphical element, users will probably fade out areas of the screen where banners are placed. In contrast, the likelihood that a banner is perceived seems to be higher on pages that users preferably visit to browse for information in a more unstructured manner (e.g. online magazines). However, as can be seen from the mean number of correctly reproduced banner elements as well as the recognition scores, the recall and recognition of banners is extremely low, questioning the effect of banner advertisement in general.
Contrary to the hypothesis, it was found that there was no difference in recall for advertising copies. This finding might be explained by the "picture superiority effect" (Nelson et al. 1976) and the "dual coding theory" (Paivio 1986). i.e. that recall for pictures (when consciously perceived) is usually much better than for textual information. It is possible that even when browsing aimlessly, the perception period for the banner ads is too short for a deeper processing of animated advertising texts. This hypothesis needs to be further investigated by using an eye tracking device, but it seems likely in the light of Bachofer's (1998) results and the results from the Stanford Poynter Project (2000).
Figure 3 shows the mean recognition scores for the single
banner ads. Recall and recognition scores cannot be directly
compared as recall scores are based on the number of
correctly reproduced features for the banner and recognition
scores are based on a confidence rating (see Procedure). The mean scores are quite
low for both groups, probably because each banner was on
screen for only 45 sec. and was then not shown again during
the experiment. In other studies that yielded higher
recall/recognition scores banners usually stayed on screen or
were shown repeatedly (e.g. Bachofer
1998; Bayles 2000).
It can also be seen there were distinct differences between
the banners. An exploratory linear regression analysis
investigating the effects of banner design on the recognition
scores for each treatment group did not yield any significant
results, however. The only interesting finding was a tendency
for the GDSI group to be more likely to recognize banner ads
if the color contrast was low within the banner and high
towards the background, giving rise to the suspicion that a
design that follows the rules of gestalt psychology can
enhance banner recall and recognition.
5.1 Limitations of the study
The results of this study are limited in several ways. The
most important limitation of the study lies in the design of
the experiment itself. While conducting the test, we observed
that some subjects of the GDSI group spent considerable time
looking at the sheet with the list of search tasks. Thus, the
difference in recall and recognition scores might be partly
attributed to differences in exposure time of the banners
because some subjects in the GDSI group might have spent less
time looking at the screen. For further studies we suggest
that the search task should be presented on screen to make
sure that exposure times for both groups are identical. We
did not control the individual attention performance nor the
prior knowledge of subjects regarding the banners presented
in this study. Therefore, differences in attention as well as
prior knowledge between the two groups might have distorted
the results. To minimize these error effects, subjects were
assigned randomly to the two conditions in our study.
However, in subsequent research, differences in attention
performance and prior knowledge should be explicitly
controlled. Finally, we did not use control banner ads in the
recognition test. Although the method used in this study is a
common way of testing recognition in advertisement research,
it would be interesting to know if the results could be
replicated using a test with distracter items.
6 Conclusion
Most studies investigating the impact of banner advertisements on the Web look at the effect of banners from an economic perspective, analysing click-through rates and examining design features that may have a positive impact on click-through behavior. So far, the psychological processes underlying the (non-)perception of banners have rarely been investigated. The results of this study suggest that trying to find the "optimal" banner design is perhaps asking the wrong question. Rather than focusing on banner design, we should focus our efforts on understanding the users and their intentions when browsing a Web site. This study shows that for further experiments in this field it is important to distinguish between the induced navigational styles in order to clarify the effects of banner ads.