Internet Advertising: Is Anybody Watching? Xavier Drèze Wharton School, University of Pennsylvania François-Xavier Hussherr * Mediametrie July 2003 This is a Draft, do not circulate or quote without prior consent from authors. *Xavier Drèze is an Assistant Professor at the Wharton School, University of Pennsylvania ([email protected]). François-Xavier Hussherr is Director of Internet and New Media activities at Mediametrie, Paris, France ([email protected]). The authors would like to thank Voilà and France Telecom for funding this research and Philippe Taupin for his help with the Eye-Tracker device. Internet Advertising: Is Anybody Watching? Abstract Click-through rates are still the de facto measure of Internet advertising effectiveness. Unfortunately, click-through rates have plummeted. This decline prompts two critical questions: (1) why do banner ads seem to be ineffective; (2) what can advertisers do to improve their effectiveness? To address these questions, we utilized an eye-tracking device to investigate online surfers’ attention to online advertising. Then we conducted a large-scale survey of Internet users’ recall, recognition, and awareness of banner advertising. Our research suggests that the reason why click-through rates are low is that surfers actually avoid looking at banner ads during their online activities. This implies that the larger part of a surfer’s processing of banners will probably be done at the pre-attentive level. If such is the case, click-through rate is an ineffective measure of banner ad performance. Our research also shows that banner ads do have an impact on traditional memory-based measure of effectiveness. Thus, we claim that advertisers should rely more on traditional brand equity measures such as brand awareness and advertising recall. Using such measures, we show that repetition affects unaided advertising recall, brand recognition, and brand awareness and that a banner’s message influences both aided advertising recall and brand recognition. Keywords: Internet, Advertising, Awareness, Brand Equity, Eye-tracking, Click-through. 2 Introduction Advertisers were one of the early proponents of the Internet, embracing its dual promise of global reach and one-to-one targeting. This should not come as a surprise as advertisers have long used every conceivable vehicle to display their messages in front of the gazing eyes of potential customers, be it magazines, television, or racecars. However, after a promising start, the burst of the Internet bubble has forced many companies to curb their plans for further expansion of their online advertising spending. Industry experts nevertheless believe that this set back will be short lived and project a resurgence of online advertising in 2004 (Forrester 2001, Internet Advertising Bureau 2002, The Wall Street Journal Online 2003) with a full recovery by 2007 (Forrester 2002). A possible obstacle to this resurgence of online advertising is its apparent lack of effectiveness. Indeed, the most widely used measure of online advertising effectiveness is the percentage of the total number of ad exposures that induce a surfer to actually click on a banner in response to an advertised message (Forrester 2001, 2002). This measure is known as the click-through rate (Novak and Hoffman 1997). Click-through rates started in 1996 at around seven percent. However, they have declined to around 0.7% in 2002 (DoubleClick 2003). This is problematic because advertisers typically do not knowingly allocate budgets to media that are not effective. Past research on advertising effectiveness cautions us regarding click-through as a valid effectiveness measure. Briggs and Hollis (1997) have shown using Milward Brown’s Brand Dynamics™ system (Dyson, Farr, and Hollis, 1996), that banner ads can have an impact on consumers’ attitudes toward a brand independent of click-through. 3 Given the doubts one may have concerning online advertising effectiveness, we believe that the following two questions must be answered before marketers can pour money back into the online realm: (1) why do banner ads seem to be ineffective; and (2) what can advertisers do to improve their effectiveness? The purpose of this paper is to answer these two questions. We first show that banner ads will most likely operate at the pre-attentive processing level (Shapiro, MacInnis, and Heckler 1997). Thus, traditional effectiveness measures are more appropriate than click-through rates. We then use memory-based measures to study some of the factors that might impact banner ad effectiveness. The remainder of the paper is organized as follows. The first section discusses the results of a study that utilized an eye-tracking device to determine whether web surfers see banner ads and which factors increase or decrease the probability that a banner ad is seen. We use the results from this first study to generate hypotheses about the characteristics of banner ads that might increase or decrease viewers’ attention. The second section relates the results of the follow-up study that tested the hypotheses generated in the first study on a broader sample of web surfers (807 respondents). The study explored the effects of Internet advertising on recall, recognition, and awareness. We then consider the results of both studies and discuss their managerial relevance. Finally, we close with concluding remarks, a discussion of the limitations of our methodology and results, and directions for future research. Study 1: Eye-tracking The Internet differs from traditional media in at least one significant way. When an advertiser uses Television or Radio to deliver his messages, he preempts the program being broadcasted (e.g., a sitcom or song) and uses all the bandwidth of the medium to transmit his message (see Drèze and Zufryden 2000). This means that by default, the viewer or listener is paying attention 4 to the advertisers, and the message is only interrupted if the listener zaps away. Zapping, however, is quite infrequent. Siddarth (2002) reports zapping rates for commercials of less than 3%. By contrast, online banner ads share their bandwidth with other elements of the pages in which they are being displayed. A banner ad typically occupies less than 10% of the area of a web page on a standard VGA computer screen (640x480 pixels). Therefore, the attention of the web surfer is generally focused on other elements of the page. The task of the banner ad is thus first to grab a surfer’s attention and second to induce the surfer to click on the ad. If surfers never look at a banner, they cannot click on it! Shared bandwidth might explain why click-through rates are low, but not why they are declining. There is some evidence that some online surfers dislike banner ads (Bass 1999). One can thus hypothesize that, as surfers gain more familiarity with the medium, they learn to differentiate informational content from advertising. Ultimately, this would give them the ability to disregard banner ads. Janiszewski (1998) shows that peripheral vision allows subjects to recognize objects that are located outside their focal point of attention (as measured by the eye- tracking device). This ability, coupled with the fact that most banner ads have the same shape (468x60 pixels) provides web surfers with the ability to train themselves into recognizing banner ads for what they are without having to actually focus on them. Both of these explanations assume that surfers learn over time and develop strategies to avoid devoting attention to advertising. Given this possible learning and avoidance behavior, we start our investigation by measuring the extent to which surfers pay attention to banner ads. We begin by formulating the following two hypotheses: H : Internet users avoid looking at banner ads. 1 5 H : The more time users have spent on the Internet; the less they pay attention to 2 banner ads. To test these hypotheses, we asked a group of subjects to look at various web pages while hooked up to an eye-tracking device that records their eye movements and fixations. The use of eye-tracking devices in marketing studies is not new. Russo and Leclerc (1994) studied in-store brand choice, Fischer et al (1989) studied warning labels on tobacco ads, Janiszewski (1998) looked at exploratory search behavior with catalogs, Kroeber-Riel (1979) investigated the effect of arousal on advertising copy processing, and Lohse (1997) studied Yellow Page advertising. Most of the eye-tracking studies, including ours, use Pupil Center/Corneal Reflection (PCCR) monitoring devices to track the eye movement of their subjects. Study design Our study was conducted using information portals as a background. The cover study was an ergonomic research on the design for one of the largest French portals: Voilà (www.voila.fr). The subjects were asked to perform five searches using three portals: Voilà, an alternate layout for Voilà (henceforth called Voilà Bis), and Voilà’s largest competitor. Three of the searches related to general topics (e.g., find information about ‘Le Louvre’), the other two related to individuals (e.g., find the phone number of ‘Jean Dupont’). Each of the three general- topic searches was made using a different portal. The two other searches were made with Voilà and Voilà Bis. To accomplish the task they were assigned, the subjects would click on a link or enter a search string. For the three general-topic searches, this would prompt the display of an answer page (see Figure 1 for the answer page to Voilà) containing the information they were asked to look for. They were then asked to indicate with the mouse where the information they were 6 looking for was located. The answer pages were designed to match the look and feel of the question page. In other words, the Voilà search page led to a Voilà answer page of similar design; the Voilà Bis search page led to a Voilà Bis answer page, and likewise for the competing portal. ===================== Insert Figure 1 about here ===================== The order in which the pages were shown was rotated across subjects with the only restriction being that the three search/answer page pairs were kept unbroken. Eight banner ads were integrated within the design of the eight web pages (see Figure 1 for the ‘Club Internet’ banner of Voilà Answer). At no time before or during the search task on these eight pages was any reference to banners ads made. We collected our data in two steps. First, we used an eye-tracking device to collect eye movements and fixations during the experiment. Second we asked our subjects to fill out a short survey after completing their assigned task on the eight web pages. The survey asked questions about their Internet savvy, the experimental process (e.g., did they encounter any stress), their preferences regarding the various pages to which they were exposed, and a series of questions regarding the banner ads they saw (e.g., do you remember seeing any banner ads). ===================== Insert Figure 2 about here ===================== 7 To simplify the analysis of the eye-tracking data, each page was dissected in a series of mutually exclusive rectangular zones. One zone was created for each paragraph of text, banner ad, or graphical element of the page (see Figure 2 for the zone definition of Voilà Answer). The eye-fixation data were then coded by zone. Hence, for every subject we have a list of each zone that they fixated on during the experiment and for how long. Of the 60 subjects that were recruited for the experiment, 11 had to be eliminated because they suffered from heavy nystagmus or because the calibration of the eye-tracking device could not be performed satisfactorily on them. This left us with 49 usable subjects. The subjects were selected through a street intercept in the center of Paris and paid €15 to participate in the study. Analysis The primary goal of this first experiment was to measure the extent to which surfers actually look at the banner ads that are embedded within the web pages. Each of our 49 subjects was exposed to 8 banner ads (one per page). Looking at the zones that were focused on by our subjects, we find that every subject looked at one or more banner ads (i.e., nobody managed to avoid every ad). On average they looked at 3.96 banners during the experiment, which yields a probability of being seen of 0.49 for each individual banner. This probability is low relative to other media such as television (>90%, per Siddarth 2002) or Yellow Page ads (89% for small display ads, 93% for large display ads, per Lohse 1997). To test Hypotheses 1 and 2, we created a data file that lists all of the zones that a subject might focus on (i.e., 111 zones x 49 subjects). Each zone is associated with variables describing their location, shape, and content (see Appendix 1 for a description of the zone description variables). We then ran a logit regression using as the dependent variable an indicator that was set to 1 if the subject fixated on the zone, and 0 otherwise. As test variables we used an Ad 8 dummy (1 if the zone is a banner ad, 0 otherwise), an Expert dummy (1 if the subjects have been on the Internet at least 25 times, 0 otherwise), and an interaction term. To control for page layout as well as possible differences across gender or age, we use the variables described in Appendix 1 as control variables along with two demographic control variables (Gender and Age –over or under 40 years old). ===================== Insert Table 1 about here ===================== The results shown in Table 1 are very revealing. As one could have expected, a zone’s location and size are important. The positive coefficient on Area shows that the bigger a zone is, the more likely it is to capture subjects’ attention. The significant interaction between the page and the zone interaction shows that the page layout is important. Similarly, the zone’s content is important as evidenced by the significant content dummies. The negative coefficient on the Ad dummy provides support for Hypothesis 1. It indicates that viewers avoid looking at ads. It also indicates that they are able to recognize that an item is an ad without having to look at it directly. Although the Expert dummy is marginally significant (p=0.11), the interaction term between expertise and banner is not significant (p=0.46). Hence, we do not find support for Hypothesis 2. Although we did not find support for Hypothesis 2, it is still interesting to contrast the behavior of our various demographic groups. During the experiment, as well as throughout our analysis, we found significant differences in behavior between novices and experts as well as between young and older surfers. To illustrate these differences, we ran a series of regressions on the number of fixes, number of zones looked at, and time spent during fixes across these 9 groups. For each page looked at by each respondent, we regressed the three dependent variables against an expertise dummy, an age dummy, and a gender dummy, as well as seven control variables to account for the differences across pages. As the results in Table 2 show, experts tend to process each page by making fewer fixes, looking at fewer zones, and spending less time then novices. An illustration of this can be seen in Figure 3 where we show an example of eye movements for both an expert and a novice. The expert clearly makes fewer fixations than the novices. Further, the eye movement seem more systematic with less back and forth movement. Older people look at the same number of zones as young people, but it takes them longer and they fixate on a larger number of positions. Finally, males and females seem to behave similarly. ============================= Insert Table 2 and Figure 3 about here ============================= As part of the debriefing questionnaire, we asked our subjects if they remembered seeing any banner ads. Only 46.9 percent of the subjects indicated they did. After asking them if they remembered seeing any ad, we showed our subject four banners and asked them if they recalled seeing the ads during the test. Two of the ads were fake ads that had not been part of the test; the other two were real. We did not find significant differences in recognition level between the fake and the real ads (m=23.5% vs. 18.4, respectively, p=0.38). The number of false positives we encountered is similar to those reported in other studies. Janiszewski (1990a) reported 21 percent of false positives (119 subjects, 5 different ads). The French division of the Internet Advertising Bureau and SOFRES (1999) reports false positive levels of 17 percent (6,872 subjects, 14 ads).