This paper develops a model called Modified Elaboration Likelihood Model to understand how people process advertising on the Internet. An empirical study verifies the new model by examining several variables influencing voluntary exposure or clicking of banner ads. These variables include (a) level of personal and product involvement, (b) the size of a banner ad, (c) relevancy between the content of a vehicle and the product category of a banner ad, (d) attitude toward the vehicle, and (e) overall attitude toward Web advertising. The findings document significant relationships between these variables and clicking of banner ads and support the hypothesized model.

Chang-Hoan Cho is a doctoral student in advertising in The University of Texas at Austin. His current research interests include Internet advertising, advertising media and international advertising. His work has appeared in the Proceedings of Conference of the American Academy of Advertising, Association for Education in Journalism and Mass Communication, and several Korean publications.

The author would like to say special thanks to Dr. John D. Leckenby for his invaluable encouragement and advice. The author also wants to thank the anonymous reviewers and the editors of JCIRA for their helpful comments.

The growth of the Internet has been exponential. By the end of 1996, there were 10 million hosts on the Internet that connected 105,000 networks (Cyber Atlas, 1996 at URL: http://www.cyberatlas.com/news.html), supporting the total Internet users, estimated between 32 million and 50 million worldwide (Forrester Research 1996, 1997 at URL: http://www.forrester.com). The numbers are doubling every year (Forrester Research, 1996 at URL: http://www.forrester.com). Among the many segments of the Internet, advertising is becoming the one with the greatest growth, with 129.5 million online in the first quarter, an increase of 18% from the preceding quarter (IAB 1997, at URL: http://www.iab.net). Jupiter Communications (URL://www.webtrack.com) has recently projected the advertising expenditure for the Internet to grow to $5 billion by the year 2000. With the advent of the new Internet World Wide Web (WWW) as an advertising medium, understanding how people process advertising on the WWW has become the critical demand of Web advertisers. But there has been little research on advertising processing on the WWW.

This paper reexamines several existing theories about how advertising works in traditional media and then explores how they can be adjusted to explain information processing of Web advertising. The main theory is the Elaboration Likelihood Model (ELM), which will be applied to build a new model called Modified Elaboration Likelihood Model. To verify the new model, several hypotheses will be postulated and they will be empirically tested through an experiment.

Many researchers have formulated different models of the stages or hierarchy consumers go through before they ultimately purchase a particular brand. These models are called hierarchy-of-effects models. The hierarchy-of-effects model was first developed by Lavidge and Steinger in 1961, even though the term itself was first used by Plalda in 1966 to refer to Lavidge and Steinger's work (Barry, 1987). Lavidge and Steinger (1961) formulated six steps consumers follow before actual purchasing: 1) awareness, 2) knowledge, 3) liking, 4) preference, 5) conviction, and 6) purchase. Following this research, there have been many research studies on how advertising works; e.g., Krugman (1965), Ray et al.(1973), Houston and Rothschild (1978), Vaughn (1979), Petty and Cacioppo (1983), and so on. Among these research studies, Petty and Cacioppo’s (1983) Elaboration Likelihood Model (ELM) is considered to be the most recent and comprehensive model incorporating involvement into the information processing of advertising messages.

According to the ELM, there are two different persuasion routes that consumers follow when they come across persuasive communication: 1) the central route and 2) the peripheral routes. When consumers have high involvement (personal or situational) or high MAO (Motivation, Ability, and Opportunity) to process communication, consumers are willing or able to exert a lot of cognitive processing effort, called high elaboration likelihood. In this situation, central cues such as existing beliefs, argument quality, and initial attitude are important in determining persuasion effects (i.e., enduring positive attitude change or boomerang effects). In contrast to high MAO situations (high involvement), when MAO is low, consumers are either not willing or unable to exert a lot of processing effort. In this low elaboration situation, peripheral persuasion cues such as attractive sources, music, humor and visuals are determining factors of persuasion effects (a temporary attitude shift or retaining the initial attitude). Figure 1 shows the step-by-step process of central and peripheral routes to persuasion in the ELM.

The hierarchy-of-effects models are conceptually useful and thus have been accepted by many advertising academicians and practitioners until today (Preston, 1982). With the advent of the new interactive communication media, such as the Internet, however, the applicability of the existing advertising theories to the Internet is questionable because of different characteristics of the Internet (two-way interaction) from those of other traditional media (one-way exposure).

The concept of interactivity has a long history. But the Internet revivifies the discussion of interactivity because interactivity is considered to be the key advantage of the medium (Rafaeli and Sudweeks, 1997; Morris and Ogan, 1996; Pavlik, 1996). It is also becoming an increasingly important characteristics of marketing as illustrated in the concept of one-to-one marketing (Peppers and Rogers, 1993). Even though different people define interactivity differently, there are three dominant ways of defining interactivity: 1) interaction between senders and receivers (Flaherty, 1985; Cook, 1994), 2) interaction between humans and machine (Rice, 1984), and 3) interaction between message and its users (Steuer, 1992; Williams et al, 1988).

The Internet is believed to have capabilities to facilitate more interactions across these three dimensions than traditional media. For example, Steuer (1992) classifies various media into three levels of interactivity: high, medium, and low interactive media. Most traditional mass media, such as newspapers, films, radio, and broadcast TV, fall into the low interactive media. Interactive television, teleconferencing, e-mail, and BBS are examples of the medium interactive media. The high interactive media include video-games, the Internet, and other multi-user domains (MUDs). Then, considering the differences in the degree of interactivity of advertising media, what are the differences between the Internet and traditional media in terms of consumers’ advertising process?

Traditional hierarchy-of-effects models assume that the very first stage of the persuasion process is awareness through advertising exposure. Here, advertising exposure is involuntary and/or incidental because individuals involuntarily just happen to come across an ad in traditional media. In contrast, advertising exposure in the Internet can be either involuntary or voluntary, depending on the types of Web advertising.

There are many different types of Web advertising (e.g., banner ads, paid hyperlinks, corporate site, personal site with selling attempts, paid icons, etc.). Two current dominant forms are 1) "banner ads" and 2) "target ads" or linked sites from the banner ads (Hoffman et al., 1995). For banner ads, the traditional involuntary exposure concept can be applied; that is, banner ads on the Web are nothing but the traditional passive form of noninteractive advertising unless they are clicked and move users into the separate target ads. If the users are only exposed to the banners ads but do not click them to open to see linked target ads, it can be said that they are not interacting with the advertising messages or the advertisers, i.e., this is traditional one-way involuntary communication from advertisers to consumers.

As long as consumers voluntarily perform an action (i.e., clicking banners) to see the content of advertising messages, information processing is more active and intensive than passive exposure without voluntary action. This voluntary exposure will draw more attention to the messages and activate the cognitive learning process more intensively than involuntary exposure. In this sense, advertising exposure in the Internet is more voluntary or sought-out than traditional media because it requires more commitment with voluntary action (i.e., clicking). That is, communication in traditional media does not require voluntary action for active information processing; it is just a one-way passive process with no extra voluntary action (e.g., clicking banner ads) other than purchase. For example, even though people can read an headline and then decide to continue reading ads or not in magazine advertising, continuing reading ads does not require any extra action with more commitment (i.e., clicking and waiting for the full download). In contrast, consumers in the Internet must voluntarily perform an extra action for an active, conscious, and cognitive information process. In other words, in the Internet, voluntary action (i.e., clicking banners) is a pre-condition of active cognitive information processing.

After the initial action (i.e., clicking banners), consumers have the choice to perform more actions for further active information processing by interacting with messages (e.g., clicking to deeper sites, searching contents, providing feedback, purchasing products on-line, etc.). In this sense, more intensive and active information processing requires more interactions between consumers and messages or between consumers and advertisers. Therefore, we can say that information processing in the Internet requires more conscious cognitive effort, because the medium itself requires action to process information; that is, information processing in the Internet is more action-oriented and more interactive than that in traditional media.

This cognitive learning process through voluntary action in the Internet is more complex than that through nonaction-oriented, involuntary exposure in the traditional media. Consumers can take many different actions during the information processing: for example, they can click away from the messages whenever they want; they can search the content of messages; they can provide feedback to advertisers at the same time as their exposure; they can save the content of advertising messages or bookmark advertising sites for future reference or voluntary repeated exposure; and more.

Another difference of information processing in the Internet from that in the traditional media is the increased possibility of short-term advertising effects. In the traditional hierarchy-of-effects models (the high involvement learning model), product purchase is the ultimate stage of the communication process resulting from a series of pre-steps (awareness, knowledge, liking, preference, and conviction), and purchase usually takes place long after consumers’ exposure to advertising messages. In other words, in traditional hierarchy models, it is believed that advertising effects occur not in the short-term but in the long-term, where consumers go through a series of steps between unawareness of a particular brand and the actual purchasing of that brand. But in the Internet, it is more likely that purchase can take place at the same time as their exposure to advertising messages or within a relatively short period of time because consumers can place an order or request additional information (e.g., different models, price, etc.) instantly and directly via the medium (the Internet) rather than having to order through another method. This is a big extension from the existing marketing practices such as DM because the Internet can provide consumers with continuously updated product information more easily and without limitation of space and time, and thus the Internet can facilitate more instant transactions than existing marketing practices.

Based on the above-mentioned differences of information processing in the Internet from that in traditional media, it is possible to modify the traditional ELM for audience processing of advertising on the WWW. The modified ELM for consumer processing of Internet advertising is shown in Figure 2. The arrows indicate the flow of the process step by step. The following section will discuss individual steps of the modified ELM for Internet advertising.

As seen in Figure 1, the very first step in the information process in the ELM is persuasive communication. The model does not describe how consumers are exposed to persuasive communication and what variables mediate advertising exposure; that is, it does not differentiate ad exposure from vehicle exposure and does not explain certain mediating variables influencing advertising exposure. As Preston (1985) argued in his work, most measurement of advertising exposure had, in practice, been based on vehicle exposure, even though there are big differences between the two. To differentiate advertising exposure from vehicle exposure, the researcher specified vehicle exposure as the very first step of the model, while understanding that, in a strict sense, the advertising process begins only with advertising exposure. That is, the first step of the advertising process in the Internet is exposure to the vehicle on which ads are placed.

During the vehicle exposure (exposure to the home Web site where banner ads are placed), consumers may or may not be exposed to banner ads. This second step of the Modified ELM is called the "opportunity to process or being exposed to banner ads involuntarily." Many variables mediate this opportunity to process (involuntary exposure to banner ads). One important mediating variable is the downloading time taken to receive messages in the vehicle and advertising. Many factors affect downloading times: 1) server’s capacity, 2) modem speed, 3) file size, and 4) number of visitors at a specific time. If it takes too long to receive messages (downloading files), consumers may not wait to retrieve the messages and click away from the site. Therefore, vehicle exposure and ad exposure are affected by downloading times.

Another mediating variable affecting involuntary exposure to banner ads is the position of banner ads. When banner ads are located at the bottom of the site, consumers may not notice even the ads unless they scroll down to surf the whole site. The chance is that the information consumers are looking for in the vehicle is located at the top or middle of the Web site, so that consumers don’t have to scroll down to the bottom of the site and are thus not exposed to banner ads located at the bottom of the site. This is another reason why vehicle exposure and ad exposure are not the same.

During involuntary exposure to banner ads, consumers may form attitude toward the banner ads and the advertised brands. But in this stage, attitude formation or change is believed to be very weak because most current banner ads contain very limited amount of advertising messages and thus do not fully activate cognitive processing of advertising messages needed for strong attitude formation or change.

As long as consumers have an opportunity to be exposed to a banner ad, they have two choices: 1) to click the banner ad to request more information or 2) not to click it. The clicking of the banner ad is totally voluntary. Then, what are the variables determining the clicking of banners? The most important determining factor of clicking a banner is the level of involvement (i.e., personal relevance and product category involvement). In the traditional ELM (Figure 1), involvement was conceptually defined as "motivation" and "ability" to process advertising messages. But in the early stage of involuntary exposure to banner ads (before clicking banners), only motivation to process ad content would be appropriate because banner ads usually do not contain much information, so that ability to process (e.g., message comprehensibility) is not required at this stage. Ability to process would work as an important factor in the later stages of voluntary exposure to target ads (i.e., a linked site after clicking banners). In short, motivation to process ad content (i.e., level of involvement) is the most important determining factor for banner clickability.

Clicking banners is a voluntary action for the purpose of seeing more detailed advertising messages by requesting more information. This voluntary exposure to advertising messages is highly dependent on consumers’ level of personal and product involvement.

In high-involvement situations (high personal and product involvement), consumers have high motivation to process advertising messages due to high personal relevance, high product category involvement, and high need for cognition. In these situations, consumers are more likely to demand greater information to satisfy their intrinsic need for information and cognition; that is, they are more likely to request more information by clicking banners in order to see detailed ad content than consumers in low-involvement situations. This can be called the central route to voluntary exposure (clicking banners). Thus, the following hypothesis can be postulated:

In contrast to high-involvement situations, consumers in low involvement situations (low personal and product involvement) have low motivation to process advertising messages due to low personal relevance and low need for cognition. Therefore, they are less likely to request more information, i.e., less likely to click banners to see more detailed information. However, they follow another route to clicking banners--the peripheral route to voluntary exposure. When consumers are not highly motivated to process further ad content, they do not want to engage in message-related thinking; rather they are more likely to focus on available peripheral cues. In other words, favorability of peripheral cues will influence clickability of banner ads in low-involvement situations. In the case of banner ads, attention-getting or curiosity-generating peripheral cues would be novelty- or contrast-related components of banner ads, such as 1) large-sized banner, 2) bright colors, and 3) attention-getting animation.

The size of the stimulus is an important factor that can affect attention. Obviously, larger ads are more likely to be noticed than smaller ones. Thus, a full-page ad will have a higher chance of drawing attention than a half- or quarter-page ad. Likewise, a larger banner ad will draw more attention than a smaller banner ad. The theory underlying this rationale is that the increase in attention is in proportion to the square root of the increase in space (Rossiter and Percy, 1980). That is, if an ad is made four times bigger, attention will double. Therefore, a larger banner ad will work as a better peripheral cue to draw low-involved people's attention than a smaller banner ad. This theoretical linkage postulates the following hypothesis:

Similarly, dynamic animation on banner ads will also work as a good peripheral cue to draw low-involved people's attention. This reasoning leads to the following hypothesis:

However, in high-involvement situations, peripheral cues of banner ads do not make any difference in the clicking of banner ads.

There are many other variables mediating voluntary exposure to target ads (linked sites from banner ads) besides the level of involvement. The following are those mediating variables affecting clickability of banner ads:

1. Relevancy Between Vehicle and Ad

Advertisements can be placed on any advertising vehicles. However, the effect of advertising is believed to be maximum when the contents of the advertising vehicle are relevant to the product categories of the advertisements placed on the vehicle. This is true for banner ads in the Internet, too. The effects of banner ads may be minimal when the product categories of banner ads are irrelevant to the contents of the Web site where the banner ads are placed. One of the reasons for this need for relevancy when placing ads is that audiences of an advertising vehicle who are exposed to the vehicle voluntarily, because they are interested in the contents of the vehicle, are more likely to read ads when the ads match with their interests. In other words, ads placed on a specific advertising vehicle are more likely to be read by the audiences of the vehicle when product categories of the ads match with the contents of the vehicle. For example, audiences of C|Net, who visit the site to see the contents (i.e., computer-related information), are more likely to click banner ads for computer-related products on that site (e.g., PC, printers, software, etc.) than for other, irrelevant banner ads (e.g., clothing, soft drinks, etc.). This conceptualization leads to the following hypothesis (H3):

 

H3: The banner ad with higher relevance between its product category and the contents of the site where the ad is placed will generate more clicking of the banner.

 

2. Attitude Toward the Vehicle

Another mediating variable affecting voluntary exposure (clicking of banner ads) is general attitude toward the vehicle where the banner ads are placed. In traditional advertising, audiences who have a more favorable attitude toward a vehicle have a more favorable attitude toward the ads placed on the vehicle--attitude transparency from vehicle to ads. Similarly, in Internet advertising, visitors who have a more favorable attitude toward the vehicle (the home ad site where banner ads are placed) have a more favorable attitude toward banner ads on the vehicle, so that they are more likely to click the banner ads. However, this effect of attitude transparency from vehicle to ads will occur only when there is high relevancy between a vehicle and the ads placed on the vehicle (i.e., relevancy between the contents of the vehicle and the product categories of the banner ads). For example, if a person has a favorable attitude toward C|Net, he/she is more likely to transfer his/her favorable attitude to a banner ad for IBM Thinkpad than for Levi's, because he/she likes the computer-related contents of C|Net, which might make him/her like computer-related ads on C|Net, too.

Therefore, it is hypothesized that there is an interaction effect of two variables (relevancy and attitude transparency) on clickability of banner ads:

Another mediating variable affecting voluntary exposure (clicking of banner ads) is the overall attitude toward Web advertising. This is another case of attitude transparency. In traditional advertising, audiences who have a more favorable attitude toward advertising overall may have a more favorable attitude toward a specific ad--attitude transparency from whole to part. Similarly, in Web advertising, people who have a more favorable attitude toward Web advertising overall may have a more favorable attitude toward a banner ad and thus be more likely to click the banner ad. This reasoning leads to the following hypotheses:

Similar to exposure to banner ads, exposure to target ads is also affected by downloading time. If it takes too long to download the target ads, people may click away from the site. Similarly, people may bookmark the target ads as soon as they arrive there, before active cognitive learning, because of slow modem speed, high traffic, etc. Likewise, throughout the process of Web advertising, there exist several noise variables affecting information processing; e.g., modem speed, traffic rate, server capacity, distraction, etc. It is also true that clicking banner ads is not necessarily a pre-stage for exposure to target ads; that is, there exist many other ways of directly being exposed to target ads without clicking banner ads. For example, people may be exposed to target ads by self-initiated direct search or by clicking the highlighted URLs delivered to them through email system. However, it is believed that clicking banner ads is the dominant channel of being exposed to target ads.

Regardless of channels, people start voluntary and active cognitive processing of advertising messages as long as they are voluntarily exposed to the detailed advertising messages of the target ads. This active cognitive processing is on a more conscious level than the information processing through involuntary exposure to traditional advertising, because people perform an action (e.g., clicking banner ads or self-initiated search) totally voluntarily to process advertising messages. This voluntariness is true, regardless of whether people click the banner ads because they are highly motivated to process (i.e., high involvement) or they click banner ads because of the favorability of peripheral cues (low involvement). But the two different involvement situations yield two different routes to persuasion, as is true in the traditional ELM: 1) central routes and 2) peripheral routes to persuasion. The difference between the traditional ELM and the Modified ELM is the degree of activeness and consciousness in processing advertising messages. That is, the Modified ELM for Web advertising has more active and more conscious cognitive processing than the traditional ELM, because exposure to advertising messages (clicking banner ads) is totally voluntary in the Modified ELM.

.

In high-involvement situations, the ability to process is the necessary condition for active cognitive processing of advertising messages. As is true in the traditional ELM, several factors determine the ability to process detailed advertising messages on the first linked page from banner ads: "distraction," "message comprehensibility," "issue familiarity," "appropriate schema," etc.

First, if people are unable to process advertising information, they cannot start active message-related cognitive processing. In this situation (high involvement but no ability to process), as is true in the traditional ELM, people will turn their attention to peripheral aspects of advertising messages such as an attractive source, music, humor, visuals, etc. Contrariwise, when people have the ability to process, they start active and conscious cognitive processing or message-related cognitive thinking. As is true in the traditional ELM, there are two determining factors in this cognitive processing: 1) the initial attitude and 2) the argument quality of advertising messages. These two factors interact with each other so that they yield three different outcomes: 1) "favorable thoughts predominate," 2) "unfavorable thoughts predominate," and 3) "neither or neutral thoughts predominate."

In the case of the last outcome (neutral thoughts), people change to the peripheral route to persuasion by focusing on peripheral cues. If they like peripheral cues, they will temporarily shift their attitude; otherwise, they will retain their initial attitude. However, for the first two outcomes (either favorable or unfavorable thoughts predominate), people experience "cognitive structure changes," where two procedures occur: 1) "new cognitions can be adopted and stored in memory" and/or 2) "different responses are made more salient than previously." Based on this change in cognitive structure, people can have two different types of attitude change: 1) an enduring positive attitude change (persuasion) for those who have predominant favorable thoughts and 2) an enduring negative attitude change (boomerang) for those who have predominant unfavorable thoughts.

Compared to people in high involvement situations, according to the traditional ELM, those in low-involvement situations are less likely to engage in message-related thinking; rather, they engage in peripheral cues present in ads. This theory can be directly applied to the Internet. In other words, people in low-involvement situations who clicked banner ads (voluntary exposure), because of the favorability of peripheral aspects of the banner ads, do not engage in active cognitive processing. Rather, they focus on peripheral cues present in the advertising messages of the first linked site from a banner ad. If they don't like the peripheral cues, they will click away from the site--stop of voluntary exposure. In this case, they retain initial attitude they had before exposure to the ad. But if the peripheral cues in the first linked site from the banner are favorable, then "peripheral attitude shift" will occur.

1) High Involvement

According to the traditional ELM, when an attitude is formed on the basis of active cognitive processing (i.e., central routes with high elaboration), it endures longer and is more likely to predict behavior than when an attitude is formed through low elaboration. In other words, when people have actively processed information, the attitude is more likely to be based on strongly held beliefs, thereby resulting in a stronger conviction. This theory about attitude duration and the attitude-behavior relationship can be linked to cognitive processing of messages in Web advertising. That is, more active cognitive processing or higher-elaboration processing may yield 1) higher duration of attitude and 2) higher predictive power for future purchase than low-elaboration processing. Here, more active cognitive processing can be conceptually defined as the level of interactivity with advertising messages and advertisers. Examples of high level of interactivity are 1) clicking into deeper sites searching for more information, 2) providing feedback to advertisers, and 3) saving the contents (i.e., bookmarking) for future reference.

As mentioned before, in low involvement situations, people will have a "peripheral attitude shift" if the peripheral cues in the first linked site from the banner are favorable. But this peripheral attitude shift through low elaboration is less likely to endure and to predict reflective behavior than attitude change through high elaboration.

To test the above hypotheses, a between-group experimental design was used. The experiment was an off-line experiment with forced exposure manipulation. According to Preston (1985), the perfect advertising effectiveness measure should be related to the actual behavior. Similarly, the most concrete measure of clicking of banner ads is looking at users’ actual behavior, i.e., click-through data. However, recognizing the difficulty of getting the actual click-through data, this study employed a mental measure of clicking, i.e., people’s self-reported intention to click banner ads. Many previous research studies on advertising effectiveness have used various mental measures, such as recall, self-reported attitude toward the ad and the brand, and purchase intention.

A total of 203 undergraduate students in a large southwestern university, divided into two experimental groups, participated in the experiment. The experiment employed a between-group subject design, where each subject was randomly exposed to only one of two experimental treatments (I or II). Each subject was exposed to a set of banner ads and Web sites based on his/her experimental group (I or II).

Before each subject was exposed to experimental materials (three Web sites and three banner ads), each subject's level of involvement was first measured. Rather than looking only at one aspect of involvement, a multiple-indicator measure, consisting of "personal relevance" and "product category involvement" was constructed. The question items consisted of seven semantic differential scales: interesting/boring, important/not important, involving/not involving, relevant/not relevant, arousing/not arousing, whether planning to purchase in the next six months, and whether making subjects continue thinking about the product. Six product categories were used to measure the level of involvement. Three of them were actual product categories used in the experiment (i.e., computers, modems, and soft drinks) and the remaining were pseudo product categories not used in the experiment (i.e., automobiles, shoes, and telephones).

After this pre-measure of involvement, each subject in each experimental group was given a questionnaire to answer that was divided into three parts. For Part I, each subject was asked to answer questions concerning his/her attitude toward three Web sites and the banner ad placed on each site. First, each subject saw the very first Web site, including a banner ad located at the top of the site, and was asked to respond to several items measuring his/her attitude toward the Web site, self-reported probability of clicking banners, and attitude toward the banner ad on that site. After completing the question items for the first site and the banner ad, each subject was exposed to the second site, including a banner ad located at the top of the site, and then again asked to fill out the question items corresponding to this site and the banner ad. Each subject followed the same procedure for the third site.

After completing Part I, each subject was asked to continue with Part II of the questionnaire, which asked some questions concerning his/her overall attitude toward advertising and attitude toward Web advertising in general. Last, each subject moved onto Part III, which asked him/her several questions about his/her demographic information. The participation for each subject took approximately 25 minutes.

According to Mitchell (1986), professionally developed ads rather than mock ads are encouraged to be used in experimental research in order to elicit a more natural response from the subjects. Following this suggestion, professionally developed Web sites and banner ads were used in this experiment. A total of six banner ads were used in this experiment, three for each experimental group. Table 1.0 summarizes the three experimental banner ads for each experimental group (the sample banner ads are available at http://uts.cc.utexas.edu/~ccho/JCIRA/ELM/samplead.html).

In this study, involvement is the most important construct. Therefore, item internal consistency for the seven involvement scales was evaluated in a reliability test. The reliabilities were satisfactory ranging from .63 to .94. This study used a between-group experimental design because of its advantage; that is, there is no chance of one treatment contaminating the other, since the same subject never receives both treatments. However, the between-subject design must content with the possibility that the subjects in the two groups are different enough to influence the effects of the treatment. To guarantee that as few differences as possible exist between two groups, the researcher compared the groups in terms of their demographics and Internet usage. Table 1.1 indicates that two groups are very similar in terms of age, gender, major, Internet-surfing hours, and the purpose of surfing the Internet. Therefore, the results eliminate the possibility that the subjects in the two groups are different enough to influence the effects of the treatments (e.g., large vs. small banner ads, animation vs. no animation, etc.).

H1: People in high-involvement situations are more likely to click banner ads in order to request more information than those in low-involvement situations.

To determine whether the two groups (high vs. low involvement people) had a significantly different intention to click the banner ad, the research used between-group t-tests. Three different analyses were conducted based on three different product categories used in the experiment (computers, modems, and soft drinks). Table 1.2 shows the mean of intention to click the banner ads for IBM Thinkpad, US Robotics, and Gatorade. The mean clicking-intention score of high involvement subjects is significantly higher than that of low involvement subjects for all three products (M=3.4 vs. M=2.8 for the IBM Thinkpad ad, M=2.7 vs. M=2.3 for the US Robotics modem, and M=2.9 vs. M=2.5 for the Gatorade ad). All results were statistically significant (p £ .01). Therefore, H1 is supported.

H2a: In low-involvement situations, people are more likely to click a banner ad when it has a larger size than average banner ads.

First, based on their personal and product involvement level, 75 out of 203 subjects are categorized into low involvement people. To determine whether two different size ads for the same product (a large and a small US Robotics ad) had a significantly different possibility to be clicked by these 75 low-involved people, the researcher used a between-group t-test. Table 2.1 shows the mean scores of the two banner ads for US Robotics (large and small size banner ad) in terms of intention to click. The mean score of large banner ad (M=2.6) is higher than that of small banner ad (M=2.1). The result was statistically significant (p £ .01) and H2a is supported.

H2b: In low-involvement situations, people are more likely to click a banner ad when it has dynamic animation than when it has no dynamic animation.

To determine whether two different animation ads for the same product (an animated and a static Gatorade ad) had a significantly different possibility to be clicked by these 103 low-involved people, the researcher used a between-group t-test. Table 2.2 shows the mean scores of the two banner ads for Gatorade ad (animated and no-animated banner ad) for intention to click. The mean score of animated Gatorade banner ad (M=2.8) is significantly higher than that of static Gatorade banner ad (M=2.3). The result is statistically significant (p £ .01) and H2b is supported.

H2c & H2d: In high-involvement situations, both the size and dynamic animation of banner ads make no difference in clicking of banner ads.

To check whether high-involved people are also more likely to click the large banner ad and the animated banner ad, the researcher used another between-group t-test. As shown in table 2.3, for high-involved people, there was no significant difference in intention to click between the large and the small banner ad. Similarly, as shown in table 2.4, for high-involved people, there was no significant difference in intention to click between the animated and the static banner ad. The results imply that both the size and dynamic animation of banner ads are not important factors influencing intention to click banner ads for high-involved people. Therefore, both H2c and H2d are supported.

H3: The banner ad with higher relevance between its product category and the contents of the site where the ad is placed will generate more clicking of the banner.

To determine whether the banner ad with higher relevance is more likely to be clicked by people, the researcher used a between-group t-test. Here, the banner ad with high relevance and low relevance were IBM Thinkpad ad placed on C|Net site and the same ad placed on ESPN Sports Zone site respectively. Table 3 shows the mean scores of the ad with high relevance and the ad with low relevance. The mean score of high-relevance banner ad (M=3.3) is higher than that of low-relevance banner ad (M=2.9). The result was statistically significant (p £ .01) and H3 is supported.

H4: People who have a more favorable attitude toward a vehicle (home Web site) are more likely to click the banner ads on that site only when the product categories of the banner ads are relevant to the contents of the Web site.

High-relevance banner ads include 1) US Robotics ad on C|Net site, 2) IBM Thinkpad ad on C|Net site, and 3) Gatorade ad on sports section of Infoseek site. To test whether people who have more favorable attitude toward a vehicle where the ad is placed are more likely to click the banner ads for the three high-relevance banner ads, three one-way ANOVA were used. The independent variable is SAS (Site Attitude Score), which is the sum of twelve Likert scores measuring the attitude toward the site. The independent variable, SAS (i.e., the sum of the twelve Likert scores), was then categorized into two levels: high and low attitude scores (above and below the median SAS). The dependent variable is one Likert scale variable that measures each subject's intention to click the banner ad.

Table 4.1 and 4.2 show the result of ANOVA for the US Robotics ad and the IBM Thinkpad ad on C|Net site respectively. It indicates that there is a significant main effect of C|Net's SAS (Site Attitude Score) on the clickability of the US Robotics banner ad (F = 9.30*, d.f.= 1,196, p £ .01) and on the clickability of the IBM Thinkpad ad (F = 20.77*, d.f.= 1,196, p £ .01). Table 4.3 also shows that there is a significant main effect of Infoseek's SAS (Site Attitude Score) on the clickability of the Gatorade banner ad (F = 11.21*, d.f.= 1,196, p £ .01). The results imply that people who have more favorable attitude toward C|Net site are more likely to click the banner ads placed on the site. That is, there is attitude transfer from the site to the banner ad when the ad and the site are relevant to each other.

However, compared to the above three results of three high-relevance banner ads, the result of ANOVA for low-relevance banner ad shows no relationship between attitude toward the site and clickability of the banner ad placed on the site. The example of low-relevance banner ad is IBM Thinkpad ad placed on ESPN Sports Zone site. Table 4.4 shows that there is no significant main effect of ESPN Sports Zone's SAS (Site Attitude Score) on the clickability of the IBM Thinkpad banner ad (F = 3.16, d.f.= 1,196, p > .01). This means that people who have more favorable attitude toward ESPN Sports Zone site are not more likely to click the banner ad (i.e., IBM Thinkpad ad) placed on the site. That is, there is no attitude transfer from the site to the banner ad when the ad and the site are not relevant to each other.

These results indicate that those who have more favorable attitude toward the home ad site are more likely to have higher intention to click the banner ad on that site, but this is only when the product category of the banner is relevant to the contents of the site the ad is placed. Therefore, H4 is supported.

H5a: People who have a more favorable attitude toward Web advertising overall have a more favorable attitude toward a banner ad.

To determine whether two groups (those who have unfavorable attitude vs. favorable attitude toward Web advertising) had a significantly different attitude toward banner ads, the researcher used MANOVA. The results in Table 5.1 indicate the greatest disparity in the favorability of the Gatorade ad between two groups is found in "Gatorade ad has good visual effects." That is, "Gatorade ad has good visual effects" variable contributes most to the overall differences in attitude toward the ad between two groups. This means that those who have favorable attitude toward Web advertising are more likely to favor Gatorade ad's visual aspects. The favorability of Web advertising has also a significant effect on "Gatorade ad is irritating" "I like Gatorade ad," "Gatorade ad is eye-catching," "Gatorade ad draws my attention," and "I would enjoy seeing the ad again." The results are statistically significant (p £ .01).

H5b: People who have a more favorable attitude toward Web advertising overall are more likely to click banner ads.

Table 5.2 shows the relationship between the variable of "intention to click" and five variables measuring attitude toward Web advertising. As shown in Table 5.2, the two groups of subjects (high vs. low intention to click banner ads) have the greatest difference in terms of the three highlighted variables. "Web advertising is valuable in general" has the largest standardized coefficient, suggesting that this is the most important variable separating two groups (high vs. low intention to click). "Web Advertising supplies valuable information" and "Web advertising is necessary" are also important discriminators of the two groups. Wilks' Lambda equals .85 in this analysis. The average score for a group (group centroid) was - .41 and .42 for people with low intention and people with high intention to click banner ads respectively.

Table 6.1 shows the results of a factor analysis for IBM Thinkpad ad using eight checklist variables that measure attitude toward the banner ad. The figures are rotated loadings resulting from a Varimax rotation of the factor axes. As shown in table 6.1, the eight checklist items measuring attitude toward IBM Thinkpad ad were grouped into three factors : Factor I was visual-effects, Factor II was seeing-again, and Factor III was informativeness. The variables were grouped under the factor of which they had the highest correlation coefficient (factor loadings shown in bold). Grouping the eight variables into these three factors retained 86 percent of the original total variance of the eight variables. Table 6.2 shows the mean surfing-hours per week of female and male respondents. The mean of male respondents (M = 4.2 hours) is higher than that of female respondents (M = 3.1 hours). The result was statistically significant (p £ .05).

This paper mainly explored two different routes from involuntary exposure to attitude formation: 1) central routes for high-involvement situations and 2) peripheral routes for low-involvement situations. This paper also looked at three mediating variables affecting voluntary exposure (clicking banner ads): 1) relevance between contents of the Web site and product categories of banner ads, 2) attitude toward a home Web site on which banner ads are placed, 3) overall attitude toward Web advertising. Based on this Modified ELM, this paper generated 7 hypotheses, and all hypotheses were empirically supported.

This study is pioneering in the sense that it is the first formal research on information processing of advertising on the WWW. However, the greatest weakness of this study is that the samples are not representative to the general population even though college students are one of the largest segments of Internet users. The picture would have been different if the research drew the samples from the general population. Another weakness of this study is that mental measures (i.e., intention to click banner ads) do not usually represent actual behavioral measures (i.e., actual click-throughs). Therefore, it would be valuable to study actual clicking behaviors of the general population.

It would be also valuable to empirically test other aspects or stages of the Modified ELM not tested in the current study. For example, according to Modified ELM, more active cognitive processing or higher-elaboration processing is supposed to yield 1) higher duration of attitude and 2) higher predictive power for future purchase than low-elaboration processing. Here, more active cognitive processing can be conceptually defined as the level of interactivity with advertising messages and advertisers. Examples of high level of interactivity are 1) clicking into deeper sites searching for more information, 2) providing feedback to advertisers, and 3) saving the contents (i.e., bookmarking) for future reference. This rational generates the following hypotheses for future research:

There are several possible operational measures of the amount of interactivity in these hypotheses. The length of stay in an interaction ad (i.e., a linked site from a banner ad) can be used as an operational measure of the different amount of interactivity, because the more people interact with the ad, the more likely that they will stay longer in the ad. However, caution needs to be taken when using length of stay as an observable measure of amount of interactivity because of possible artificial inflation in this measure. For example, a consumer can be distracted from an interaction ad by a phone call or a knock on the door, making him/her to attend to other tasks while the ad is still up on the screen. Another measure of advertising exposure as amount of interaction is the number of pages or screens the users click into. The deeper users click into the sites (visiting more linked sites), the more they interact with the ad.

To test these hypotheses, multiple methods with multiple kinds of data, as suggested in Williams et al's (1988) Research Methods and the New Media, can be employed in the future research. That is, a simulation-based experiment in a laboratory setting can be used. It can combine online and offline measurement techniques (unobtrusive capturing of usage data with online techniques, and possible control and manipulation as in offline assessment). In other words, this future experiment may allow careful observation of subjects' activities during the experiments to measure levels of interactivity (e.g., # of clickings, duration time, # of pages, etc.) so that a future researcher can test the effects of different levels of interactivity on the duration of attitude change and reflective behavior (Future H1a and H1b).

Despite the increasing significance of Internet advertising, there has been no research on consumers' information processing of Internet advertising. In this sense, this paper provides some groundwork in this field. Most studies on Internet advertising have been conducted by Web publishers on audience measurement data, i.e., how many people visit their sites, or how many people are exposed to banner ads, etc. But this kind of result-oriented data does not provide the understanding of consumers' step-by-step information processing, e.g., why people click banner ads and why they click one banner ad more than another. In conclusion, information processing of Internet advertising is too important to leave unstudied, therefore, more future studies on this area are strongly recommended.

N=203

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

p £ .01

p £ .01

**Attitude toward the home ad site was categorical variable with two levels (favorable and unfavorable attitude)

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

p £ .01

**Attitude toward the home ad site was categorical variable with two levels (favorable and unfavorable attitude)

**Intention to click banner ads was measured by 5-point Likert item of "I will click the banner ad to further see the detailed description of the ad"

Wilks’ Lambda = 0.62, Approximate F-ratio = 14.5*, d.f. = (8, 191), p £ .01

** All items (5-point Likert scales) were coded by giving a big number to positive agreement with positive statement and negative agreement with negative statement.

**The independent variable (attitude toward Web advertising) had two levels (favorable and unfavorable attitude).

Wilks’ Lambda = 0.85, Chi-Square = 31.7*, d.f. = 5, p £ .01

** 5 items measuring attitude toward Web advertising were measured by 5-point Likert scale.

Total variance explained with three factors: 86%

** All items (5-point Likert scales) were coded by giving a big number to positive agreement with positive statement and negative agreement with negative statement.

p £ .05

Figure 1

Elaboration Likelihood Model of Persuasion

Figure 2

Modified Elaboration Likelihood Model of Persuasion

 

 

Reference

 

Abnernethy, Avery M. (1991), "Differences between advertising and program exposure for car radio listening," Journal of Advertising Research, April/May, 33-42.

Advertising Research Foundation (1961), Toward Better Media Comparisons, Audience Concepts Committee, New York.

Barry, Thomas (1987), "The Development of the Hierarchy of Effects: A Historical Perspective," in Current Issues and Research in Advertising, James H. Leigh and Claude R. Martin, Jr., ed., Clemson, SC: CtC Press, 251-295.

Cook, William A. (1994), "Is It Interactive Media, or Hyperactive Media?" Journal of Advertising Research, 34 (Spring), 7-9.

Cyber Atlas (1996), [URL: http://www.cyberatlas.com/news.html].

Flaherty, David H. (1985), Protecting Privacy in two-way electronic services, White Plains, New York.

Forrester Research (1996, 1997), [URL: http://www.forrester.com/].

Hoffman, D. L., T.P. Novak, and P. Chatterjee (1995), "Commercial Scenarios for the Web: Opportunities and Challenges," Journal of Computer-Mediated Communication, Special Issue on Electronic Commerce, 1 (December).

Houston, Michael J. and Michael L. Rothschild (1978), "Conceptual and Methodological Perspectives on Involvement," in Research Frontiers in Marketing: Dialogues and Directions, Subhash C. Jain, ed., Chicago: American Marketing Association, 184-187.

Hoyer, W.D. and D. J. MacInnis (1990), Consumer Behavior, 2nd ed., Boston, MA: Houghton Press.

IAB (Internet Advertising Bureau) (1997), [URL: http://www.iab.net/].

Jupiter Communication (1996), [URL: http://www.webtrack.com/].

Krugman, Herbert E. (1965), "The Impact of Television Advertising: Learning without Involvement," Public Opinion Quarterly, 29, 349-356.

Lavidge, Robert C. and Gary Steiner (1961), "A Model for Predictive Measurements of Advertising Effectiveness," Journal of Marketing, 25(October), 59-62.

Mitchell, A. A. (1986), "Theoretical and Methodological Issues in Developing an Individual Level Model of Advertising Effects," in Advertising and Consumer Psychology, Olson, J.C., and K. Sentis, ed., New York: Praeger Publishing Co.

Morris, Merrill and Christine Ogan (1996), "The Internet as Mass Medium," Journal of Computer Mediated Communication, 1(4), [URL: http://www.cwis.usc.edu/dep/annenberg/vol1/issue4/vol1no4.html].

Pavlik, John V. (1996), New Media technology: Cultural and Commercial Perspectives, Boston, MA: Allyn & Bacon.

Peppers, Don and Martha Rogers (1993), The One to One Future: Building Relationships One Customer at a Time, New York, NY: Currency Doubleday

Petty, Richard E. and John T. Cacioppo (1981), "Central and Peripheral Routes to Persuasion:

Applications to Advertising," in Advertising and Consumer Psychology, L. Percy and A. G. Woodside, ed., Lexington, MA: Lexington Books, 3-24.

------------ and ------------(1983), "Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement," Journal of Consumer Research, 19(September), 135-146.

Preston, Ivan L. (1982), "The Association Model of the Advertising Communication Process," Journal of Advertising, 11(Summer), 3-15.

------------ (1985), "The Developing Detachment of Advertising Research from the Study of Advertisers' Goals," in Current Issues and Research in Advertising, James H. Leigh and Claude R. Martin, Jr., ed., Clemson, SC: CtC Press

Raraeli, Scheizaf and Fay Sudweeks (1997), "Networked Interactivity," Journal of Computer Mediated Communication, 2, [URL: http://www.usc.edu/dept/annenberg/vol2/issue4/rafaeli.sudweeks.html].

Ray, Michael L., Sawyer, Tothschild, Heeler, Strong, and Reed (1973), "Marketing Communication and the Hierarchy of Effects," in New Models for Mass Communication Research, Peter Clark, ed., 147-176.

Rethans, Arno J., John L. S. and Lawrence J. Marks (1986), "Effect of Television Commercial Repetition, Receiver Knowledge, and Commercial Length: A Test of the Two-Factor Model," Journal of Marketing Research, 13 (February), 50-61.

Rice, Ronald E. (1984), "New Media Technology: Growth and Integration", in The New Media R. Rice & Associates, ed., Beverly Hills: Sage.

Rossiter, J. R., and L. Percy (1980), "Attitude Change through Visual Imagery in Advertising," Journal of Advertising, 9, 10-16.

Steuer, Jonathan (1992), "Defining Virtual Reality: Dimensions Determining Telepresence," Journal of Communication, 42 (Winter), 73-93

Vaughn, Richard (1983), "How Advertising Works A Planning Model," Journal of Advertising Research, 23, 22-28.

Williams, Frederick, Ronald E. Rice, and Everett M. Rogers (1988), Research Methods and the New Media, New York: The Free Press.